Release [run release-20261111T164520Z]

#22
by Battam - opened
This view is limited to 50 files because it contains too many changes. See the raw diff here.
Files changed (50) hide show
  1. CHANGELOG.md +18 -0
  2. README.md +33 -33
  3. README.zh.md +33 -33
  4. campaign_files/2026-10-a-spreadsheets-with-answers-2/montreal-open-data-workbooks/README.md +0 -2
  5. campaign_files/2026-10-c-repros-2/moshenfeld-pld-accounting-repro/README.md +0 -2
  6. campaign_files/2026-10-c-repros-2/mw-rank12-exact-repro/README.md +0 -2
  7. campaign_files/2026-10-g-solver-instances-2/mmlib/README.md +0 -4
  8. campaign_files/2026-10-gt-browser-tests-2/mutter-monitor-config-tests/README.md +0 -22
  9. campaign_files/2026-10-m-issue-trackers-2/mongodb-jira/README.md +0 -4
  10. campaign_files/2026-10-m-issue-trackers-2/morebus-passenger-gtfs/README.md +0 -4
  11. campaign_files/2026-10-m-issue-trackers-2/mozilla-holiday-calendars/README.md +0 -8
  12. campaign_files/2026-10-m-issue-trackers-2/mvgo-gtfs/README.md +0 -4
  13. campaign_files/2026-10-m-issue-trackers-2/mythtv-mythtv-users/README.md +0 -4
  14. campaign_files/2026-10-md-media-and-documents/mmark-testdata-table/README.md +0 -4
  15. campaign_files/2026-10-md-media-and-documents/myst-parser-renderer-fixtures-tables/README.md +0 -4
  16. campaign_files/2026-10-md-media-and-documents/myst-to-typst-tests/README.md +0 -4
  17. campaign_files/2026-10-ml-serving-configs-2/mitre-menelaus/README.md +0 -6
  18. campaign_files/2026-10-ml-serving-configs-2/ml341-05/README.md +0 -4
  19. campaign_files/2026-10-ml-serving-configs-2/ml341-08/README.md +0 -4
  20. campaign_files/2026-10-ml-serving-configs-2/ml341-09/README.md +0 -4
  21. campaign_files/2026-10-ml-serving-configs-2/ml341-10/README.md +0 -4
  22. campaign_files/2026-10-ml-serving-configs-2/ml437-01/README.md +0 -4
  23. campaign_files/2026-10-ml-serving-configs-2/ml437-02/README.md +0 -4
  24. campaign_files/2026-10-ml-serving-configs-2/ml437-03/README.md +0 -4
  25. campaign_files/2026-10-ml-serving-configs-2/ml437-04/README.md +0 -4
  26. campaign_files/2026-10-ml-serving-configs-2/ml437-05/README.md +0 -4
  27. campaign_files/2026-10-ml-serving-configs-2/ml437-06/README.md +0 -4
  28. campaign_files/2026-10-ml-serving-configs-2/ml437-07/README.md +0 -4
  29. campaign_files/2026-10-ml-serving-configs-2/ml437-08/README.md +0 -4
  30. campaign_files/2026-10-ml-serving-configs-2/ml437-09/README.md +0 -4
  31. campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-3dunet-brats/README.md +0 -4
  32. campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-3dunet-kits/README.md +0 -4
  33. campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-bert/README.md +0 -4
  34. campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-dlrm/README.md +0 -4
  35. campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-gptj/README.md +0 -4
  36. campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-rnnt/README.md +0 -4
  37. campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-vision/README.md +0 -4
  38. campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-whisper/README.md +0 -4
  39. campaign_files/2026-10-ml-serving-configs-2/modelcloud-gptqmodel/README.md +0 -8
  40. campaign_files/2026-10-ml-serving-configs-2/musl-libc-test-math/LICENSE +3 -0
  41. campaign_files/2026-10-ml-serving-configs-2/musl-libc-test-math/README.md +24 -0
  42. campaign_files/2026-10-ml-serving-configs-2/nab-artificialwithanomaly/README.md +0 -4
  43. campaign_files/2026-10-ml-serving-configs-2/nab-realawscloudwatch/README.md +0 -4
  44. campaign_files/2026-10-ml-serving-configs/mlperf-inference-llama31-8b/README.md +0 -4
  45. campaign_files/2026-10-s-security-detection-2/modsecurity-v3-regression-tests/README.md +0 -8
  46. campaign_files/2026-10-s-security-detection-2/mq-hackmac-2021-public-challenges/README.md +0 -12
  47. campaign_files/2026-10-s-security-detection/modsecurity-v3-regression/README.md +0 -4
  48. campaign_files/2026-10-sc-science-reference-data-2/modern-robotics-library-docstring-examples/README.md +0 -4
  49. campaign_files/2026-10-sc-science-reference-data-2/molgw-testsuite/README.md +0 -4
  50. campaign_files/2026-10-sc-science-reference-data-2/mopac-tests/README.md +0 -6
CHANGELOG.md CHANGED
@@ -3195,3 +3195,21 @@ Changes to the content of this index, oldest first. There is one entry for each
3195
  - hf-training-records: refused
3196
  - npm: refused
3197
  - stackexchange-api-posts: refused
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3195
  - hf-training-records: refused
3196
  - npm: refused
3197
  - stackexchange-api-posts: refused
3198
+
3199
+ ## 2026-11-11: sources held back
3200
+
3201
+ - Each source below keeps its newest version that passed every check; its newer batch waits until it passes them:
3202
+ - borealis-dataverse-workbooks-cc-by: changed pointer group file data/borealis-dataverse-workbooks-cc-by/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 2 key(s) of the old version are in no data file of source borealis-dataverse-workbooks-cc-by after the commit (row 4 from batch 2026-10-a-spreadsheets-with-answers, row 5 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
3203
+ - borealis-dataverse-workbooks-cc0: changed pointer group file data/borealis-dataverse-workbooks-cc0/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 1 key(s) of the old version are in no data file of source borealis-dataverse-workbooks-cc0 after the commit (row 7 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
3204
+ - cfpb-complaints: refused
3205
+ - data-gov-uk-workbooks: changed pointer group file data/data-gov-uk-workbooks/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 4 key(s) of the old version are in no data file of source data-gov-uk-workbooks after the commit (row 5 from batch 2026-10-a-spreadsheets-with-answers, row 6 from batch 2026-10-a-spreadsheets-with-answers, row 12 from batch 2026-10-a-spreadsheets-with-answers, 1 more); the removal record holds none of them, and none fails a publish check now
3206
+ - data-gov-uk-workbooks-noassertion: changed pointer group file data/data-gov-uk-workbooks-noassertion/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 1 key(s) of the old version are in no data file of source data-gov-uk-workbooks-noassertion after the commit (row 6 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
3207
+ - dataversenl-workbooks-cc-by: changed pointer group file data/dataversenl-workbooks-cc-by/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 1 key(s) of the old version are in no data file of source dataversenl-workbooks-cc-by after the commit (row 17 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
3208
+ - dataversenl-workbooks-cc0: changed pointer group file data/dataversenl-workbooks-cc0/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 3 key(s) of the old version are in no data file of source dataversenl-workbooks-cc0 after the commit (row 7 from batch 2026-10-a-spreadsheets-with-answers, row 13 from batch 2026-10-a-spreadsheets-with-answers, row 16 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
3209
+ - dvn-fvhlgg-repro: changed pointer group file data/dvn-fvhlgg-repro/2026-10-c-repros/part-00000.parquet: 3 key(s) of the old version are in no data file of source dvn-fvhlgg-repro after the commit (row 5 from batch 2026-10-c-repros, row 7 from batch 2026-10-c-repros, row 16 from batch 2026-10-c-repros); the removal record holds none of them, and none fails a publish check now
3210
+ - eu-institution-workbooks-cc-by: changed pointer group file data/eu-institution-workbooks-cc-by/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 3 key(s) of the old version are in no data file of source eu-institution-workbooks-cc-by after the commit (row 2 from batch 2026-10-a-spreadsheets-with-answers, row 4 from batch 2026-10-a-spreadsheets-with-answers, row 5 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
3211
+ - hdx-workbooks-cc0: changed pointer group file data/hdx-workbooks-cc0/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 2 key(s) of the old version are in no data file of source hdx-workbooks-cc0 after the commit (row 4 from batch 2026-10-a-spreadsheets-with-answers, row 5 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
3212
+ - hdx-workbooks-odbl: changed pointer group file data/hdx-workbooks-odbl/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 1 key(s) of the old version are in no data file of source hdx-workbooks-odbl after the commit (row 4 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
3213
+ - hf-training-records: refused
3214
+ - npm: refused
3215
+ - stackexchange-api-posts: refused
README.md CHANGED
@@ -24,25 +24,25 @@ configs:
24
  <!-- artifact-pointer-layout: v2 -->
25
  <!-- Generated by artpool.index_projection from meta/, cards/*.yaml, the field registry and its templates. Every release regenerates it, so edits to this file do not last. -->
26
 
27
- <img class="not-prose" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/banner.svg" alt="Public web, in two layers (task material in collections; corpora), one row per file (address, size, SHA-256, license), your selection, fetched and verified">
28
 
29
  This dataset is the public index of the **Training Data Foundation**: find public files to build agent tasks and environments from, and fetch them with one command.
30
 
31
  English | [简体中文](README.zh.md) · [Data page](https://huggingface.co/spaces/Battam/training-data-foundation) · [Usage video](https://huggingface.co/spaces/Battam/training-data-foundation#usage-video)
32
 
33
  <picture class="not-prose">
34
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/numbers-narrow-light.svg">
35
- <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/numbers-wide-light.svg" alt="1,805 material sets with 4,190 units in 14 collections; 139 corpora with 1,216,362,128 rows; 4.45 PB pointed to, as the index holds addresses, not copies; 99.6% of the 34,707 rows compared match their recorded size.">
36
  </picture>
37
  <picture class="not-prose">
38
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/numbers-narrow-dark.svg">
39
- <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/numbers-wide-dark.svg" alt="1,805 material sets with 4,190 units in 14 collections; 139 corpora with 1,216,362,128 rows; 4.45 PB pointed to, as the index holds addresses, not copies; 99.6% of the 34,707 rows compared match their recorded size.">
40
  </picture>
41
 
42
  ## Get data in one command
43
 
44
  ```sh
45
- uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/tools/fetch.py --form spreadsheets --sample 3 --out ./data
46
  ```
47
 
48
  This fetches 3 units of every spreadsheet material set into `./data`, one folder per unit, and checks each file against the size and hashes its row records.
@@ -52,7 +52,7 @@ the same command takes the same units of this release.
52
  <details>
53
  <summary>What the command takes, and in which order</summary>
54
 
55
- The command and the queries below are pinned to the release tag `release-2026-11-11.14`.
56
  [uv](https://docs.astral.sh/uv/) reads the script's dependencies, at the versions published before the day of this release.
57
 
58
  The command returns units: 3 units from each material set whose form is `spreadsheets`, each in a folder of its own named after its `unit_id`.
@@ -79,7 +79,7 @@ Choose material sets from the catalog with [DuckDB](https://duckdb.org/):
79
 
80
  ```sql
81
  SELECT source, collection, title, units, files, top_types, license.tier AS license_tier, needs_auth
82
- FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.14/meta/catalog.parquet'
83
  WHERE kind = 'material_set' AND needs_auth = 'no' AND license.tier IN ('permissive', 'share-alike')
84
  ORDER BY collection, units DESC;
85
  ```
@@ -90,7 +90,7 @@ with `kind = 'corpus'` it returns corpora instead, counted in rows (`files`).
90
  Choose files in one corpus:
91
 
92
  ```sql
93
- SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.14/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.14/meta/catalog.parquet' WHERE source = 'if-archive'));
94
  SELECT uri, size_bytes, lower(hex(sha256)) AS sha256
95
  FROM read_parquet(getvariable('files'))
96
  WHERE source = 'if-archive' AND ext = 'pdf' AND size_bytes BETWEEN 100000 AND 5000000
@@ -112,15 +112,15 @@ A data file may hold the rows of several sources, so the query keeps this one's
112
  ### Task material: units by collection
113
 
114
  <picture class="not-prose">
115
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/material-narrow-light.svg">
116
- <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/material-wide-light.svg" alt="Units by collection, longest bar first: Spreadsheets 536; Browser tests 439; Tests and exercises 425; Model serving configurations 397; Media and documents 391; Science reference data 342; Rule calculations 329; Security detection 328; Hardware design (EDA) tests 258; Issue trackers 258; Solver instances 193; Disk images 179; Reproduction packages and runs 106; In no collection 9; SEC filing reviews 0.">
117
  </picture>
118
  <picture class="not-prose">
119
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/material-narrow-dark.svg">
120
- <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/material-wide-dark.svg" alt="Units by collection, longest bar first: Spreadsheets 536; Browser tests 439; Tests and exercises 425; Model serving configurations 397; Media and documents 391; Science reference data 342; Rule calculations 329; Security detection 328; Hardware design (EDA) tests 258; Issue trackers 258; Solver instances 193; Disk images 179; Reproduction packages and runs 106; In no collection 9; SEC filing reviews 0.">
121
  </picture>
122
 
123
- 1,805 material sets with 4,190 units, in 14 collections (1 of them with no material set yet); 1 of the material sets are in no collection.
124
 
125
  <details>
126
  <summary>Task material by collection, as a table</summary>
@@ -136,7 +136,7 @@ A data file may hold the rows of several sources, so the query keeps this one's
136
  | Hardware design (EDA) tests | 81 | 258 | 651 | 75 | 80.8 MB | 80.8 MB |
137
  | Issue trackers | 212 | 258 | 682 | 212 | 196 MB | 196 MB |
138
  | Media and documents | 276 | 391 | 1,088 | 269 | 569 MB | 569 MB |
139
- | Model serving configurations | 102 | 397 | 1,165 | 82 | 250 MB | 221 MB |
140
  | Rule calculations | 177 | 329 | 727 | 154 | 196 MB | 186 MB |
141
  | Security detection | 110 | 328 | 983 | 77 | 684 MB | 652 MB |
142
  | Science reference data | 161 | 342 | 671 | 157 | 63.8 MB | 63.8 MB |
@@ -150,12 +150,12 @@ Upstreams are the upstream works the material sets of a collection come from.
150
  ### Corpora: rows by main domain
151
 
152
  <picture class="not-prose">
153
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/corpora-narrow-light.svg">
154
- <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/corpora-wide-light.svg" alt="Corpus rows by main domain, longest bar first: Data science 750,815,212; System administration 185,826,217; Many domains 109,262,742; Software engineering 45,267,701; Science 44,633,675; Operations 39,404,642; Media 12,751,612; Machine learning 9,636,889; Hardware 8,716,520; Unclassified 5,284,162; Security 4,521,631; Misc 241,125.">
155
  </picture>
156
  <picture class="not-prose">
157
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/corpora-narrow-dark.svg">
158
- <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/corpora-wide-dark.svg" alt="Corpus rows by main domain, longest bar first: Data science 750,815,212; System administration 185,826,217; Many domains 109,262,742; Software engineering 45,267,701; Science 44,633,675; Operations 39,404,642; Media 12,751,612; Machine learning 9,636,889; Hardware 8,716,520; Unclassified 5,284,162; Security 4,521,631; Misc 241,125.">
159
  </picture>
160
 
161
  139 corpora with 1,216,362,128 rows.
@@ -188,12 +188,12 @@ Every material set and every corpus, with its counts, size, license tier and log
188
  ## How it is made
189
 
190
  <picture class="not-prose">
191
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/steps-narrow-light.svg">
192
- <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/steps-wide-light.svg" alt="How the index is made, in four steps. 1, collect: collectors read whole upstreams into corpora, one row per file or record. 2, admit: works the search found upstream and groups of corpus rows become material sets. 3, check: licenses, personal data, overlap with benchmarks and sizes are checked before release. 4, release: one commit with its own tag, release-2026-11-11.14.">
193
  </picture>
194
  <picture class="not-prose">
195
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/steps-narrow-dark.svg">
196
- <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/steps-wide-dark.svg" alt="How the index is made, in four steps. 1, collect: collectors read whole upstreams into corpora, one row per file or record. 2, admit: works the search found upstream and groups of corpus rows become material sets. 3, check: licenses, personal data, overlap with benchmarks and sizes are checked before release. 4, release: one commit with its own tag, release-2026-11-11.14.">
197
  </picture>
198
 
199
  The terms, each step and how every number is made are folded below.
@@ -247,9 +247,9 @@ only their sizes add up to the size of the whole index.
247
 
248
  | Level | Recorded size | Deduplicated size | Rows read | Rows compared | Rows that match | Rows read, not compared | Rows without a recorded size | Their estimated size |
249
  |---|---:|---:|---:|---:|---:|---:|---:|---:|
250
- | Whole index | 4.45 PB | 4.11 PB | 35,586 | 34,707 | 34,578 (99.6%) | 879: address unreachable 449, file gone 80, login needed 329, no size from the server 19, other error 2 | 312,300,338 | 202 TB, an estimate (95% range 171 TB to 245 TB), for 215,326,513 of them, from 14,161 rows read |
251
  | Corpora | 4.45 PB | 4.11 PB | 31,594 | 30,723 | 30,610 (99.6%) | 871: address unreachable 449, file gone 80, login needed 322, no size from the server 19, other error 1 | 312,300,338 | 202 TB, an estimate (95% range 171 TB to 245 TB), for 215,326,513 of them, from 14,161 rows read |
252
- | Task material | 2.88 GB | 2.81 GB | 3,992 | 3,984 | 3,968 (99.6%) | 8: login needed 7, other error 1 | 0 | none |
253
 
254
  Rows read are the sample of rows the check reads in `meta/sizes.parquet`;
255
  a row compared matched its recorded size or did not, and a row read but not compared never reached its file.
@@ -404,7 +404,7 @@ Every row field and every column of the catalog: [`docs/fields.md`](docs/fields.
404
  | `none_known` | 3 | 0 |
405
  | `present` | 14 | 0 |
406
  | `suspected` | 1 | 0 |
407
- | Not judged yet | 121 | 1,805 |
408
  | Sample files read per judged source | 9 to 30 | none |
409
  | Days the sample files were read | 2026-10-04 and 2026-10-05 | none |
410
 
@@ -415,10 +415,10 @@ Every row field and every column of the catalog: [`docs/fields.md`](docs/fields.
415
 
416
  | Sources | Corpora | Task material |
417
  |---|---:|---:|
418
- | With a recorded size for every row | 92 of 139 | 1,805 of 1,805 |
419
- | With a summary read from sample files | 0 of 139 | 0 of 1,805 |
420
- | With personal information read from sample files | 18 of 139 | 0 of 1,805 |
421
- | Still being collected | 25 of 139 | 0 of 1,805 |
422
 
423
  - Rows whose size the source does not publish are shown as unknown and left out of the size totals, never counted as 0.
424
  - Fields not judged yet say so;
@@ -444,9 +444,9 @@ Every row field and every column of the catalog: [`docs/fields.md`](docs/fields.
444
  <details>
445
  <summary>Releases</summary>
446
 
447
- - Latest release: `release-2026-11-11.14` (its newest CHANGELOG entry is dated 2026-11-11).
448
  - Releases are listed in [CHANGELOG.md](CHANGELOG.md);
449
- each one is a commit with its own release tag (this one is [`release-2026-11-11.14`](https://huggingface.co/datasets/Battam/training-data-foundation/tree/release-2026-11-11.14)),
450
  so any earlier state stays readable at its tag.
451
  - `meta/projection.json` records the release tag and the version of the program and templates that generated this card.
452
  - Croissant metadata: [generated by the Hub](https://huggingface.co/api/datasets/Battam/training-data-foundation/croissant);
@@ -479,7 +479,7 @@ A removal covers the current version of this repository (the main branch) and th
479
  @misc{training_data_foundation,
480
  title = {training-data-foundation: an index of public materials for building agent environments},
481
  howpublished = {https://huggingface.co/datasets/Battam/training-data-foundation},
482
- note = {Release release-2026-11-11.14},
483
  year = {2026}
484
  }
485
  ```
 
24
  <!-- artifact-pointer-layout: v2 -->
25
  <!-- Generated by artpool.index_projection from meta/, cards/*.yaml, the field registry and its templates. Every release regenerates it, so edits to this file do not last. -->
26
 
27
+ <img class="not-prose" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/banner.svg" alt="Public web, in two layers (task material in collections; corpora), one row per file (address, size, SHA-256, license), your selection, fetched and verified">
28
 
29
  This dataset is the public index of the **Training Data Foundation**: find public files to build agent tasks and environments from, and fetch them with one command.
30
 
31
  English | [简体中文](README.zh.md) · [Data page](https://huggingface.co/spaces/Battam/training-data-foundation) · [Usage video](https://huggingface.co/spaces/Battam/training-data-foundation#usage-video)
32
 
33
  <picture class="not-prose">
34
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/numbers-narrow-light.svg">
35
+ <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/numbers-wide-light.svg" alt="1,806 material sets with 4,191 units in 14 collections; 139 corpora with 1,216,362,128 rows; 4.45 PB pointed to, as the index holds addresses, not copies; 99.6% of the 34,485 rows compared match their recorded size.">
36
  </picture>
37
  <picture class="not-prose">
38
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/numbers-narrow-dark.svg">
39
+ <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/numbers-wide-dark.svg" alt="1,806 material sets with 4,191 units in 14 collections; 139 corpora with 1,216,362,128 rows; 4.45 PB pointed to, as the index holds addresses, not copies; 99.6% of the 34,485 rows compared match their recorded size.">
40
  </picture>
41
 
42
  ## Get data in one command
43
 
44
  ```sh
45
+ uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/tools/fetch.py --form spreadsheets --sample 3 --out ./data
46
  ```
47
 
48
  This fetches 3 units of every spreadsheet material set into `./data`, one folder per unit, and checks each file against the size and hashes its row records.
 
52
  <details>
53
  <summary>What the command takes, and in which order</summary>
54
 
55
+ The command and the queries below are pinned to the release tag `release-2026-11-11.15`.
56
  [uv](https://docs.astral.sh/uv/) reads the script's dependencies, at the versions published before the day of this release.
57
 
58
  The command returns units: 3 units from each material set whose form is `spreadsheets`, each in a folder of its own named after its `unit_id`.
 
79
 
80
  ```sql
81
  SELECT source, collection, title, units, files, top_types, license.tier AS license_tier, needs_auth
82
+ FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.15/meta/catalog.parquet'
83
  WHERE kind = 'material_set' AND needs_auth = 'no' AND license.tier IN ('permissive', 'share-alike')
84
  ORDER BY collection, units DESC;
85
  ```
 
90
  Choose files in one corpus:
91
 
92
  ```sql
93
+ SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.15/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.15/meta/catalog.parquet' WHERE source = 'if-archive'));
94
  SELECT uri, size_bytes, lower(hex(sha256)) AS sha256
95
  FROM read_parquet(getvariable('files'))
96
  WHERE source = 'if-archive' AND ext = 'pdf' AND size_bytes BETWEEN 100000 AND 5000000
 
112
  ### Task material: units by collection
113
 
114
  <picture class="not-prose">
115
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/material-narrow-light.svg">
116
+ <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/material-wide-light.svg" alt="Units by collection, longest bar first: Spreadsheets 536; Browser tests 439; Tests and exercises 425; Model serving configurations 398; Media and documents 391; Science reference data 342; Rule calculations 329; Security detection 328; Hardware design (EDA) tests 258; Issue trackers 258; Solver instances 193; Disk images 179; Reproduction packages and runs 106; In no collection 9; SEC filing reviews 0.">
117
  </picture>
118
  <picture class="not-prose">
119
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/material-narrow-dark.svg">
120
+ <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/material-wide-dark.svg" alt="Units by collection, longest bar first: Spreadsheets 536; Browser tests 439; Tests and exercises 425; Model serving configurations 398; Media and documents 391; Science reference data 342; Rule calculations 329; Security detection 328; Hardware design (EDA) tests 258; Issue trackers 258; Solver instances 193; Disk images 179; Reproduction packages and runs 106; In no collection 9; SEC filing reviews 0.">
121
  </picture>
122
 
123
+ 1,806 material sets with 4,191 units, in 14 collections (1 of them with no material set yet); 1 of the material sets are in no collection.
124
 
125
  <details>
126
  <summary>Task material by collection, as a table</summary>
 
136
  | Hardware design (EDA) tests | 81 | 258 | 651 | 75 | 80.8 MB | 80.8 MB |
137
  | Issue trackers | 212 | 258 | 682 | 212 | 196 MB | 196 MB |
138
  | Media and documents | 276 | 391 | 1,088 | 269 | 569 MB | 569 MB |
139
+ | Model serving configurations | 103 | 398 | 1,168 | 83 | 250 MB | 221 MB |
140
  | Rule calculations | 177 | 329 | 727 | 154 | 196 MB | 186 MB |
141
  | Security detection | 110 | 328 | 983 | 77 | 684 MB | 652 MB |
142
  | Science reference data | 161 | 342 | 671 | 157 | 63.8 MB | 63.8 MB |
 
150
  ### Corpora: rows by main domain
151
 
152
  <picture class="not-prose">
153
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/corpora-narrow-light.svg">
154
+ <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/corpora-wide-light.svg" alt="Corpus rows by main domain, longest bar first: Data science 750,815,212; System administration 185,826,217; Many domains 109,262,742; Software engineering 45,267,701; Science 44,633,675; Operations 39,404,642; Media 12,751,612; Machine learning 9,636,889; Hardware 8,716,520; Unclassified 5,284,162; Security 4,521,631; Misc 241,125.">
155
  </picture>
156
  <picture class="not-prose">
157
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/corpora-narrow-dark.svg">
158
+ <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/corpora-wide-dark.svg" alt="Corpus rows by main domain, longest bar first: Data science 750,815,212; System administration 185,826,217; Many domains 109,262,742; Software engineering 45,267,701; Science 44,633,675; Operations 39,404,642; Media 12,751,612; Machine learning 9,636,889; Hardware 8,716,520; Unclassified 5,284,162; Security 4,521,631; Misc 241,125.">
159
  </picture>
160
 
161
  139 corpora with 1,216,362,128 rows.
 
188
  ## How it is made
189
 
190
  <picture class="not-prose">
191
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/steps-narrow-light.svg">
192
+ <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/steps-wide-light.svg" alt="How the index is made, in four steps. 1, collect: collectors read whole upstreams into corpora, one row per file or record. 2, admit: works the search found upstream and groups of corpus rows become material sets. 3, check: licenses, personal data, overlap with benchmarks and sizes are checked before release. 4, release: one commit with its own tag, release-2026-11-11.15.">
193
  </picture>
194
  <picture class="not-prose">
195
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/steps-narrow-dark.svg">
196
+ <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/steps-wide-dark.svg" alt="How the index is made, in four steps. 1, collect: collectors read whole upstreams into corpora, one row per file or record. 2, admit: works the search found upstream and groups of corpus rows become material sets. 3, check: licenses, personal data, overlap with benchmarks and sizes are checked before release. 4, release: one commit with its own tag, release-2026-11-11.15.">
197
  </picture>
198
 
199
  The terms, each step and how every number is made are folded below.
 
247
 
248
  | Level | Recorded size | Deduplicated size | Rows read | Rows compared | Rows that match | Rows read, not compared | Rows without a recorded size | Their estimated size |
249
  |---|---:|---:|---:|---:|---:|---:|---:|---:|
250
+ | Whole index | 4.45 PB | 4.11 PB | 35,364 | 34,485 | 34,356 (99.6%) | 879: address unreachable 449, file gone 80, login needed 329, no size from the server 19, other error 2 | 312,300,338 | 202 TB, an estimate (95% range 171 TB to 245 TB), for 215,326,513 of them, from 14,161 rows read |
251
  | Corpora | 4.45 PB | 4.11 PB | 31,594 | 30,723 | 30,610 (99.6%) | 871: address unreachable 449, file gone 80, login needed 322, no size from the server 19, other error 1 | 312,300,338 | 202 TB, an estimate (95% range 171 TB to 245 TB), for 215,326,513 of them, from 14,161 rows read |
252
+ | Task material | 2.88 GB | 2.81 GB | 3,770 | 3,762 | 3,746 (99.6%) | 8: login needed 7, other error 1 | 0 | none |
253
 
254
  Rows read are the sample of rows the check reads in `meta/sizes.parquet`;
255
  a row compared matched its recorded size or did not, and a row read but not compared never reached its file.
 
404
  | `none_known` | 3 | 0 |
405
  | `present` | 14 | 0 |
406
  | `suspected` | 1 | 0 |
407
+ | Not judged yet | 121 | 1,806 |
408
  | Sample files read per judged source | 9 to 30 | none |
409
  | Days the sample files were read | 2026-10-04 and 2026-10-05 | none |
410
 
 
415
 
416
  | Sources | Corpora | Task material |
417
  |---|---:|---:|
418
+ | With a recorded size for every row | 92 of 139 | 1,806 of 1,806 |
419
+ | With a summary read from sample files | 0 of 139 | 0 of 1,806 |
420
+ | With personal information read from sample files | 18 of 139 | 0 of 1,806 |
421
+ | Still being collected | 25 of 139 | 0 of 1,806 |
422
 
423
  - Rows whose size the source does not publish are shown as unknown and left out of the size totals, never counted as 0.
424
  - Fields not judged yet say so;
 
444
  <details>
445
  <summary>Releases</summary>
446
 
447
+ - Latest release: `release-2026-11-11.15` (its newest CHANGELOG entry is dated 2026-11-11).
448
  - Releases are listed in [CHANGELOG.md](CHANGELOG.md);
449
+ each one is a commit with its own release tag (this one is [`release-2026-11-11.15`](https://huggingface.co/datasets/Battam/training-data-foundation/tree/release-2026-11-11.15)),
450
  so any earlier state stays readable at its tag.
451
  - `meta/projection.json` records the release tag and the version of the program and templates that generated this card.
452
  - Croissant metadata: [generated by the Hub](https://huggingface.co/api/datasets/Battam/training-data-foundation/croissant);
 
479
  @misc{training_data_foundation,
480
  title = {training-data-foundation: an index of public materials for building agent environments},
481
  howpublished = {https://huggingface.co/datasets/Battam/training-data-foundation},
482
+ note = {Release release-2026-11-11.15},
483
  year = {2026}
484
  }
485
  ```
README.zh.md CHANGED
@@ -11,25 +11,25 @@ tags:
11
  ---
12
  <!-- Generated by artpool.index_projection from meta/, cards/*.yaml, the field registry and its templates. Every release regenerates it, so edits to this file do not last. -->
13
 
14
- <img class="not-prose" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/banner.zh.svg" alt="公开网络,分两层(按合集的任务材料;语料库),每个文件一行(地址、大小、SHA-256、许可证),你的选择,下载并校验">
15
 
16
  本数据集是**数据基座**的公开索引:在这里找到可以用来构建智能体任务和环境的公开文件,并用一条命令把它们取下来。
17
 
18
  [English](README.md) | 简体中文 · [数据页](https://huggingface.co/spaces/Battam/training-data-foundation) · [使用示例视频](https://huggingface.co/spaces/Battam/training-data-foundation#usage-video)
19
 
20
  <picture class="not-prose">
21
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/numbers-narrow-light.zh.svg">
22
- <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/numbers-wide-light.zh.svg" alt="1,805 个材料集,共 4,190 个单元,分在 14 个合集;139 个语料库,共 1,216,362,128 行;指向的大小共 4.45 PB,索引存的是地址,不是文件副本;比对过的 34,707 行中,99.6% 与记录的大小一致。">
23
  </picture>
24
  <picture class="not-prose">
25
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/numbers-narrow-dark.zh.svg">
26
- <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/numbers-wide-dark.zh.svg" alt="1,805 个材料集,共 4,190 个单元,分在 14 个合集;139 个语料库,共 1,216,362,128 行;指向的大小共 4.45 PB,索引存的是地址,不是文件副本;比对过的 34,707 行中,99.6% 与记录的大小一致。">
27
  </picture>
28
 
29
  ## 一条命令取数据
30
 
31
  ```sh
32
- uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/tools/fetch.py --form spreadsheets --sample 3 --out ./data
33
  ```
34
 
35
  这条命令从每个电子表格材料集各取 3 个单元放进 `./data`,每个单元一个文件夹,并按每个文件所在行记录的大小和哈希值校验它。
@@ -39,7 +39,7 @@ uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/r
39
  <details>
40
  <summary>命令取什么、按什么顺序取</summary>
41
 
42
- 这条命令和下面的查询都固定在发布标签 `release-2026-11-11.14` 上。
43
  [uv](https://docs.astral.sh/uv/) 会读取脚本声明的依赖,取这个发布当天之前发布的版本。
44
 
45
  命令返回单元:形式为 `spreadsheets` 的每个材料集各 3 个单元,每个单元放在以它的 `unit_id` 命名的单独文件夹里。
@@ -66,7 +66,7 @@ uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/r
66
 
67
  ```sql
68
  SELECT source, collection, title, units, files, top_types, license.tier AS license_tier, needs_auth
69
- FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.14/meta/catalog.parquet'
70
  WHERE kind = 'material_set' AND needs_auth = 'no' AND license.tier IN ('permissive', 'share-alike')
71
  ORDER BY collection, units DESC;
72
  ```
@@ -77,7 +77,7 @@ ORDER BY collection, units DESC;
77
  在一个语料库里挑选文件:
78
 
79
  ```sql
80
- SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.14/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.14/meta/catalog.parquet' WHERE source = 'if-archive'));
81
  SELECT uri, size_bytes, lower(hex(sha256)) AS sha256
82
  FROM read_parquet(getvariable('files'))
83
  WHERE source = 'if-archive' AND ext = 'pdf' AND size_bytes BETWEEN 100000 AND 5000000
@@ -99,15 +99,15 @@ ORDER BY size_bytes LIMIT 20;
99
  ### 任务材料:按合集的单元数
100
 
101
  <picture class="not-prose">
102
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/material-narrow-light.zh.svg">
103
- <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/material-wide-light.zh.svg" alt="按合集的单元数,条形由长到短:电子表格 536;浏览器测试 439;测试和练习 425;模型服务配置 397;媒体和文档 391;科学参考数据 342;规则计算 329;安全检测 328;硬件设计(EDA)测试 258;问题跟踪系统 258;求解器实例 193;磁盘镜像 179;复现包和运行 106;不属于任何合集 9;SEC 申报审查 0。">
104
  </picture>
105
  <picture class="not-prose">
106
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/material-narrow-dark.zh.svg">
107
- <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/material-wide-dark.zh.svg" alt="按合集的单元数,条形由长到短:电子表格 536;浏览器测试 439;测试和练习 425;模型服务配置 397;媒体和文档 391;科学参考数据 342;规则计算 329;安全检测 328;硬件设计(EDA)测试 258;问题跟踪系统 258;求解器实例 193;磁盘镜像 179;复现包和运行 106;不属于任何合集 9;SEC 申报审查 0。">
108
  </picture>
109
 
110
- 1,805 个材料集,共 4,190 个单元,分在 14 个合集中(其中 1 个合集还没有材料集);其中 1 个材料集不属于任何合集。
111
 
112
  <details>
113
  <summary>按合集的任务材料(表格)</summary>
@@ -123,7 +123,7 @@ ORDER BY size_bytes LIMIT 20;
123
  | 硬件设计(EDA)测试 | 81 | 258 | 651 | 75 | 80.8 MB | 80.8 MB |
124
  | 问题跟踪系统 | 212 | 258 | 682 | 212 | 196 MB | 196 MB |
125
  | 媒体和文档 | 276 | 391 | 1,088 | 269 | 569 MB | 569 MB |
126
- | 模型服务配置 | 102 | 397 | 1,165 | 82 | 250 MB | 221 MB |
127
  | 规则计算 | 177 | 329 | 727 | 154 | 196 MB | 186 MB |
128
  | 安全检测 | 110 | 328 | 983 | 77 | 684 MB | 652 MB |
129
  | 科学参考数据 | 161 | 342 | 671 | 157 | 63.8 MB | 63.8 MB |
@@ -137,12 +137,12 @@ ORDER BY size_bytes LIMIT 20;
137
  ### 语料库:按主要领域的行数
138
 
139
  <picture class="not-prose">
140
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/corpora-narrow-light.zh.svg">
141
- <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/corpora-wide-light.zh.svg" alt="按主要领域的语料库行数,条形由长到短:数据科学 750,815,212;系统管理 185,826,217;多领域 109,262,742;软件工程 45,267,701;科学 44,633,675;运营 39,404,642;媒体 12,751,612;机器学习 9,636,889;硬件 8,716,520;未分类 5,284,162;安全 4,521,631;杂项 241,125。">
142
  </picture>
143
  <picture class="not-prose">
144
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/corpora-narrow-dark.zh.svg">
145
- <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/corpora-wide-dark.zh.svg" alt="按主要领域的语料库行数,条形由长到短:数据科学 750,815,212;系统管理 185,826,217;多领域 109,262,742;软件工程 45,267,701;科学 44,633,675;运营 39,404,642;媒体 12,751,612;机器学习 9,636,889;硬件 8,716,520;未分类 5,284,162;安全 4,521,631;杂项 241,125。">
146
  </picture>
147
 
148
  139 个语料库,共 1,216,362,128 行。
@@ -175,12 +175,12 @@ ORDER BY size_bytes LIMIT 20;
175
  ## 怎样做成
176
 
177
  <picture class="not-prose">
178
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/steps-narrow-light.zh.svg">
179
- <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/steps-wide-light.zh.svg" alt="索引分四步做成。1,采集:采集器整批读取完整上游,每个文件或每条记录记一行,成为语料库。2,收录:检索在上游找到的作品和成组的语料库行,收录为材料集。3,检查:发布前核对许可证、个人信息、与基准测试的重叠和大小。4,发布:一个带有自己标签的提交,release-2026-11-11.14。">
180
  </picture>
181
  <picture class="not-prose">
182
- <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/steps-narrow-dark.zh.svg">
183
- <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.14/docs/figures/steps-wide-dark.zh.svg" alt="索引分四步做成。1,采集:采集器整批读取完整上游,每个文件或每条记录记一行,成为语料库。2,收录:检索在上游找到的作品和成组的语料库行,收录为材料集。3,检查:发布前核对许可证、个人信息、与基准测试的重叠和大小。4,发布:一个带有自己标签的提交,release-2026-11-11.14。">
184
  </picture>
185
 
186
  术语、每一步,以及每个数字怎样得出,都折叠在下面。
@@ -234,9 +234,9 @@ ORDER BY size_bytes LIMIT 20;
234
 
235
  | 层级 | 记录的大小 | 去重后大小 | 读过的行 | 比对过的行 | 一致的行 | 读过但未比对的行 | 没有记录大小的行 | 这些行的估计大小 |
236
  |---|---:|---:|---:|---:|---:|---:|---:|---:|
237
- | 整个索引 | 4.45 PB | 4.11 PB | 35,586 | 34,707 | 34,578(99.6%) | 879:地址无法访问 449 行、文件已不存在 80 行、需要登录 329 行、服务器未给出大小 19 行、其他错误 2 行 | 312,300,338 | 202 TB,这是估计值(95% 范围 171 TB 到 245 TB),针对其中 215,326,513 行,依据读过的 14,161 行 |
238
  | 语料库 | 4.45 PB | 4.11 PB | 31,594 | 30,723 | 30,610(99.6%) | 871:地址无法访问 449 行、文件已不存在 80 行、需要登录 322 行、服务器未给出大小 19 行、其他错误 1 行 | 312,300,338 | 202 TB,这是估计值(95% 范围 171 TB 到 245 TB),针对其中 215,326,513 行,依据读过的 14,161 行 |
239
- | 任务材料 | 2.88 GB | 2.81 GB | 3,992 | 3,984 | 3,968(99.6%) | 8:需要登录 7 行、其他错误 1 行 | 0 | 无 |
240
 
241
  读过的行是核对在 `meta/sizes.parquet` 中读取的抽样行;
242
  比对过的行与记录的大小一致或不一致,读过但未比对的行没有取到它的文件。
@@ -393,7 +393,7 @@ ORDER BY size_bytes LIMIT 20;
393
  | `none_known` | 3 | 0 |
394
  | `present` | 14 | 0 |
395
  | `suspected` | 1 | 0 |
396
- | 尚未判读 | 121 | 1,805 |
397
  | 每个已判读的来源读过的样本文件数 | 9 到 30 | 无 |
398
  | 读取样本文件的日期 | 2026-10-04 和 2026-10-05 | 无 |
399
 
@@ -404,10 +404,10 @@ ORDER BY size_bytes LIMIT 20;
404
 
405
  | 来源 | 语料库 | 任务材料 |
406
  |---|---:|---:|
407
- | 每一行都有记录的大小 | 139 个中的 92 个 | 1,805 个中的 1,805 个 |
408
- | 已从样本文件读出摘要 | 139 个中的 0 个 | 1,805 个中的 0 个 |
409
- | 已从样本文件读出个人信息 | 139 个中的 18 个 | 1,805 个中的 0 个 |
410
- | 仍在采集中 | 139 个中的 25 个 | 1,805 个中的 0 个 |
411
 
412
  - 来源没有公布大小的行显示为未知,不计入大小合计,也从不按 0 计。
413
  - 尚未判读的字段会注明;
@@ -433,9 +433,9 @@ ORDER BY size_bytes LIMIT 20;
433
  <details>
434
  <summary>发布</summary>
435
 
436
- - 最新发布:`release-2026-11-11.14`(其最新的 CHANGELOG 条目日期为 2026-11-11)。
437
  - 各次发布列在 [CHANGELOG.md](CHANGELOG.md) 中;
438
- 每次发布是一个带有自己发布标签的提交(本次为 [`release-2026-11-11.14`](https://huggingface.co/datasets/Battam/training-data-foundation/tree/release-2026-11-11.14)),
439
  因此任何较早的状态都可以在它的标签处读到。
440
  - `meta/projection.json` 记录发布标签,以及生成本卡片的程序和模板的版本。
441
  - Croissant 元数据:[由 Hub 生成](https://huggingface.co/api/datasets/Battam/training-data-foundation/croissant);
@@ -476,7 +476,7 @@ ORDER BY size_bytes LIMIT 20;
476
  @misc{training_data_foundation,
477
  title = {training-data-foundation: an index of public materials for building agent environments},
478
  howpublished = {https://huggingface.co/datasets/Battam/training-data-foundation},
479
- note = {Release release-2026-11-11.14},
480
  year = {2026}
481
  }
482
  ```
 
11
  ---
12
  <!-- Generated by artpool.index_projection from meta/, cards/*.yaml, the field registry and its templates. Every release regenerates it, so edits to this file do not last. -->
13
 
14
+ <img class="not-prose" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/banner.zh.svg" alt="公开网络,分两层(按合集的任务材料;语料库),每个文件一行(地址、大小、SHA-256、许可证),你的选择,下载并校验">
15
 
16
  本数据集是**数据基座**的公开索引:在这里找到可以用来构建智能体任务和环境的公开文件,并用一条命令把它们取下来。
17
 
18
  [English](README.md) | 简体中文 · [数据页](https://huggingface.co/spaces/Battam/training-data-foundation) · [使用示例视频](https://huggingface.co/spaces/Battam/training-data-foundation#usage-video)
19
 
20
  <picture class="not-prose">
21
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/numbers-narrow-light.zh.svg">
22
+ <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/numbers-wide-light.zh.svg" alt="1,806 个材料集,共 4,191 个单元,分在 14 个合集;139 个语料库,共 1,216,362,128 行;指向的大小共 4.45 PB,索引存的是地址,不是文件副本;比对过的 34,485 行中,99.6% 与记录的大小一致。">
23
  </picture>
24
  <picture class="not-prose">
25
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/numbers-narrow-dark.zh.svg">
26
+ <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/numbers-wide-dark.zh.svg" alt="1,806 个材料集,共 4,191 个单元,分在 14 个合集;139 个语料库,共 1,216,362,128 行;指向的大小共 4.45 PB,索引存的是地址,不是文件副本;比对过的 34,485 行中,99.6% 与记录的大小一致。">
27
  </picture>
28
 
29
  ## 一条命令取数据
30
 
31
  ```sh
32
+ uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/tools/fetch.py --form spreadsheets --sample 3 --out ./data
33
  ```
34
 
35
  这条命令从每个电子表格材料集各取 3 个单元放进 `./data`,每个单元一个文件夹,并按每个文件所在行记录的大小和哈希值校验它。
 
39
  <details>
40
  <summary>命令取什么、按什么顺序取</summary>
41
 
42
+ 这条命令和下面的查询都固定在发布标签 `release-2026-11-11.15` 上。
43
  [uv](https://docs.astral.sh/uv/) 会读取脚本声明的依赖,取这个发布当天之前发布的版本。
44
 
45
  命令返回单元:形式为 `spreadsheets` 的每个材料集各 3 个单元,每个单元放在以它的 `unit_id` 命名的单独文件夹里。
 
66
 
67
  ```sql
68
  SELECT source, collection, title, units, files, top_types, license.tier AS license_tier, needs_auth
69
+ FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.15/meta/catalog.parquet'
70
  WHERE kind = 'material_set' AND needs_auth = 'no' AND license.tier IN ('permissive', 'share-alike')
71
  ORDER BY collection, units DESC;
72
  ```
 
77
  在一个语料库里挑选文件:
78
 
79
  ```sql
80
+ SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.15/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.15/meta/catalog.parquet' WHERE source = 'if-archive'));
81
  SELECT uri, size_bytes, lower(hex(sha256)) AS sha256
82
  FROM read_parquet(getvariable('files'))
83
  WHERE source = 'if-archive' AND ext = 'pdf' AND size_bytes BETWEEN 100000 AND 5000000
 
99
  ### 任务材料:按合集的单元数
100
 
101
  <picture class="not-prose">
102
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/material-narrow-light.zh.svg">
103
+ <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/material-wide-light.zh.svg" alt="按合集的单元数,条形由长到短:电子表格 536;浏览器测试 439;测试和练习 425;模型服务配置 398;媒体和文档 391;科学参考数据 342;规则计算 329;安全检测 328;硬件设计(EDA)测试 258;问题跟踪系统 258;求解器实例 193;磁盘镜像 179;复现包和运行 106;不属于任何合集 9;SEC 申报审查 0。">
104
  </picture>
105
  <picture class="not-prose">
106
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/material-narrow-dark.zh.svg">
107
+ <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/material-wide-dark.zh.svg" alt="按合集的单元数,条形由长到短:电子表格 536;浏览器测试 439;测试和练习 425;模型服务配置 398;媒体和文档 391;科学参考数据 342;规则计算 329;安全检测 328;硬件设计(EDA)测试 258;问题跟踪系统 258;求解器实例 193;磁盘镜像 179;复现包和运行 106;不属于任何合集 9;SEC 申报审查 0。">
108
  </picture>
109
 
110
+ 1,806 个材料集,共 4,191 个单元,分在 14 个合集中(其中 1 个合集还没有材料集);其中 1 个材料集不属于任何合集。
111
 
112
  <details>
113
  <summary>按合集的任务材料(表格)</summary>
 
123
  | 硬件设计(EDA)测试 | 81 | 258 | 651 | 75 | 80.8 MB | 80.8 MB |
124
  | 问题跟踪系统 | 212 | 258 | 682 | 212 | 196 MB | 196 MB |
125
  | 媒体和文档 | 276 | 391 | 1,088 | 269 | 569 MB | 569 MB |
126
+ | 模型服务配置 | 103 | 398 | 1,168 | 83 | 250 MB | 221 MB |
127
  | 规则计算 | 177 | 329 | 727 | 154 | 196 MB | 186 MB |
128
  | 安全检测 | 110 | 328 | 983 | 77 | 684 MB | 652 MB |
129
  | 科学参考数据 | 161 | 342 | 671 | 157 | 63.8 MB | 63.8 MB |
 
137
  ### 语料库:按主要领域的行数
138
 
139
  <picture class="not-prose">
140
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/corpora-narrow-light.zh.svg">
141
+ <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/corpora-wide-light.zh.svg" alt="按主要领域的语料库行数,条形由长到短:数据科学 750,815,212;系统管理 185,826,217;多领域 109,262,742;软件工程 45,267,701;科学 44,633,675;运营 39,404,642;媒体 12,751,612;机器学习 9,636,889;硬件 8,716,520;未分类 5,284,162;安全 4,521,631;杂项 241,125。">
142
  </picture>
143
  <picture class="not-prose">
144
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/corpora-narrow-dark.zh.svg">
145
+ <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/corpora-wide-dark.zh.svg" alt="按主要领域的语料库行数,条形由长到短:数据科学 750,815,212;系统管理 185,826,217;多领域 109,262,742;软件工程 45,267,701;科学 44,633,675;运营 39,404,642;媒体 12,751,612;机器学习 9,636,889;硬件 8,716,520;未分类 5,284,162;安全 4,521,631;杂项 241,125。">
146
  </picture>
147
 
148
  139 个语料库,共 1,216,362,128 行。
 
175
  ## 怎样做成
176
 
177
  <picture class="not-prose">
178
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/steps-narrow-light.zh.svg">
179
+ <img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/steps-wide-light.zh.svg" alt="索引分四步做成。1,采集:采集器整批读取完整上游,每个文件或每条记录记一行,成为语料库。2,收录:检索在上游找到的作品和成组的语料库行,收录为材料集。3,检查:发布前核对许可证、个人信息、与基准测试的重叠和大小。4,发布:一个带有自己标签的提交,release-2026-11-11.15。">
180
  </picture>
181
  <picture class="not-prose">
182
+ <source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/steps-narrow-dark.zh.svg">
183
+ <img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.15/docs/figures/steps-wide-dark.zh.svg" alt="索引分四步做成。1,采集:采集器整批读取完整上游,每个文件或每条记录记一行,成为语料库。2,收录:检索在上游找到的作品和成组的语料库行,收录为材料集。3,检查:发布前核对许可证、个人信息、与基准测试的重叠和大小。4,发布:一个带有自己标签的提交,release-2026-11-11.15。">
184
  </picture>
185
 
186
  术语、每一步,以及每个数字怎样得出,都折叠在下面。
 
234
 
235
  | 层级 | 记录的大小 | 去重后大小 | 读过的行 | 比对过的行 | 一致的行 | 读过但未比对的行 | 没有记录大小的行 | 这些行的估计大小 |
236
  |---|---:|---:|---:|---:|---:|---:|---:|---:|
237
+ | 整个索引 | 4.45 PB | 4.11 PB | 35,364 | 34,485 | 34,356(99.6%) | 879:地址无法访问 449 行、文件已不存在 80 行、需要登录 329 行、服务器未给出大小 19 行、其他错误 2 行 | 312,300,338 | 202 TB,这是估计值(95% 范围 171 TB 到 245 TB),针对其中 215,326,513 行,依据读过的 14,161 行 |
238
  | 语料库 | 4.45 PB | 4.11 PB | 31,594 | 30,723 | 30,610(99.6%) | 871:地址无法访问 449 行、文件已不存在 80 行、需要登录 322 行、服务器未给出大小 19 行、其他错误 1 行 | 312,300,338 | 202 TB,这是估计值(95% 范围 171 TB 到 245 TB),针对其中 215,326,513 行,依据读过的 14,161 行 |
239
+ | 任务材料 | 2.88 GB | 2.81 GB | 3,770 | 3,762 | 3,746(99.6%) | 8:需要登录 7 行、其他错误 1 行 | 0 | 无 |
240
 
241
  读过的行是核对在 `meta/sizes.parquet` 中读取的抽样行;
242
  比对过的行与记录的大小一致或不一致,读过但未比对的行没有取到它的文件。
 
393
  | `none_known` | 3 | 0 |
394
  | `present` | 14 | 0 |
395
  | `suspected` | 1 | 0 |
396
+ | 尚未判读 | 121 | 1,806 |
397
  | 每个已判读的来源读过的样本文件数 | 9 到 30 | 无 |
398
  | 读取样本文件的日期 | 2026-10-04 和 2026-10-05 | 无 |
399
 
 
404
 
405
  | 来源 | 语料库 | 任务材料 |
406
  |---|---:|---:|
407
+ | 每一行都有记录的大小 | 139 个中的 92 个 | 1,806 个中的 1,806 个 |
408
+ | 已从样本文件读出摘要 | 139 个中的 0 个 | 1,806 个中的 0 个 |
409
+ | 已从样本文件读出个人信息 | 139 个中的 18 个 | 1,806 个中的 0 个 |
410
+ | 仍在采集中 | 139 个中的 25 个 | 1,806 个中的 0 个 |
411
 
412
  - 来源没有公布大小的行显示为未知,不计入大小合计,也从不按 0 计。
413
  - 尚未判读的字段会注明;
 
433
  <details>
434
  <summary>发布</summary>
435
 
436
+ - 最新发布:`release-2026-11-11.15`(其最新的 CHANGELOG 条目日期为 2026-11-11)。
437
  - 各次发布列在 [CHANGELOG.md](CHANGELOG.md) 中;
438
+ 每次发布是一个带有自己发布标签的提交(本次为 [`release-2026-11-11.15`](https://huggingface.co/datasets/Battam/training-data-foundation/tree/release-2026-11-11.15)),
439
  因此任何较早的状态都可以在它的标签处读到。
440
  - `meta/projection.json` 记录发布标签,以及生成本卡片的程序和模板的版本。
441
  - Croissant 元数据:[由 Hub 生成](https://huggingface.co/api/datasets/Battam/training-data-foundation/croissant);
 
476
  @misc{training_data_foundation,
477
  title = {training-data-foundation: an index of public materials for building agent environments},
478
  howpublished = {https://huggingface.co/datasets/Battam/training-data-foundation},
479
+ note = {Release release-2026-11-11.15},
480
  year = {2026}
481
  }
482
  ```
campaign_files/2026-10-a-spreadsheets-with-answers-2/montreal-open-data-workbooks/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** The file is self-contained.
10
 
11
- **Answers:** Each file with an answer has one answer file, at answer/<unit path>.json (the unit path plus .json). Reference result kind: key_formula_values.
12
-
13
  **Tools to open and check:** openpyxl (formulas and cached values); LibreOffice headless recalculation (soffice --headless --convert-to xlsx)
14
 
15
  **Licence (SPDX):** CC-BY-4.0
 
8
 
9
  **Self-contained:** The file is self-contained.
10
 
 
 
11
  **Tools to open and check:** openpyxl (formulas and cached values); LibreOffice headless recalculation (soffice --headless --convert-to xlsx)
12
 
13
  **Licence (SPDX):** CC-BY-4.0
campaign_files/2026-10-c-repros-2/moshenfeld-pld-accounting-repro/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** Each unit with an answer has one answer file, at answer/<package>/reported.json.
12
-
13
  **Tools to open and check:** Open with tools for these file types: .py.
14
 
15
  **Licence (SPDX):** MIT
 
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Open with tools for these file types: .py.
12
 
13
  **Licence (SPDX):** MIT
campaign_files/2026-10-c-repros-2/mw-rank12-exact-repro/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** Each unit with an answer has one answer file, at answer/<package>/reported.json.
12
-
13
  **Tools to open and check:** Open with tools for these file types: .py; .json; .txt.
14
 
15
  **Licence (SPDX):** NOASSERTION
 
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Open with tools for these file types: .py; .json; .txt.
12
 
13
  **Licence (SPDX):** NOASSERTION
campaign_files/2026-10-g-solver-instances-2/mmlib/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** The unit is self-contained. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: Optimal makespan (literature lower bound equals upper bound). Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Check: Program: check one mode per activity, precedence, renewable usage per period within availability and total non-renewable use within budget; pass when feasible and makespan equals 31.
14
 
15
  **Licence (SPDX):** NOASSERTION
@@ -22,8 +20,6 @@ Units:
22
  Read out of the archive: https://github.com/MarioVanhoucke/MMLIB-Multi-Mode-Resource-Constrained-Project-Library/releases/download/v2.1/MMLIB_v2_1.zip
23
  Format: PSPLIB style .mm text with modes, durations, renewable and non-renewable requests; one case = one instance; answer = mode choice and start times, and the makespan
24
  Origin: generated by Van Peteghem and Vanhoucke (2014) with RanGen. Standard benchmark suite or competition instance as published by the library, recorded tool_fixture.
25
- Reference result: optimal makespan (literature lower bound equals upper bound)
26
- Reference source: MMLIB (Parameters and BKS).xlsx, sheet MMLIB50, row 3 (D3 "J501_1.mm"): N3 LB-lit 31, O3 UB-lit 31
27
  Check: Program: check one mode per activity, precedence, renewable usage per period within availability and total non-renewable use within budget; pass when feasible and makespan equals 31
28
  Companions (not included): MMLIB_v2_1/MMLIB (Parameters and BKS).xlsx
29
  Dependencies: an xlsx reader to read the companion
 
8
 
9
  **Self-contained:** The unit is self-contained. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Program: check one mode per activity, precedence, renewable usage per period within availability and total non-renewable use within budget; pass when feasible and makespan equals 31.
12
 
13
  **Licence (SPDX):** NOASSERTION
 
20
  Read out of the archive: https://github.com/MarioVanhoucke/MMLIB-Multi-Mode-Resource-Constrained-Project-Library/releases/download/v2.1/MMLIB_v2_1.zip
21
  Format: PSPLIB style .mm text with modes, durations, renewable and non-renewable requests; one case = one instance; answer = mode choice and start times, and the makespan
22
  Origin: generated by Van Peteghem and Vanhoucke (2014) with RanGen. Standard benchmark suite or competition instance as published by the library, recorded tool_fixture.
 
 
23
  Check: Program: check one mode per activity, precedence, renewable usage per period within availability and total non-renewable use within budget; pass when feasible and makespan equals 31
24
  Companions (not included): MMLIB_v2_1/MMLIB (Parameters and BKS).xlsx
25
  Dependencies: an xlsx reader to read the companion
campaign_files/2026-10-gt-browser-tests-2/mutter-monitor-config-tests/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether each file needs other upstream files is not stated. 10 of 10 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: expected_display_configuration. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
14
 
15
  **Licence (SPDX):** GPL-2.0-or-later
@@ -21,8 +19,6 @@ Files:
21
  - `src/tests/monitor-configs/vertical.xml`: Two monitors stacked vertically: a 1024x768 primary on top at (0,0) and an 800x600 monitor below it at (0,768), logical layout mode, written as a GNOME monitors.xml display configuration.
22
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
23
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
24
- Reference result: expected_display_configuration
25
- Reference source: the expected configuration struct of meta_test_monitor_store_vertical in src/tests/monitor-store-unit-tests.c lines 286-355 (2 logical monitors; DP-1 1024x768 at (0,0) primary; DP-2 800x600 at (0,768) not primary; scale 1; refresh 60.000495910644531), checked field by field by check_monitor_store_configuration (check_monitor_store_configuration and check_monitor_store_configurations, lines 122-237: layout rectangle, scale, transform, is_primary, is_presentation, connector, vendor, product, serial, mode width, height, refresh rate and refresh rate mode, flags, underscanning, max bpc, rgb range); the reference file src/tests/monitor-configs/vertical.xml is the stored form of the same configuration
26
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
27
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
28
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
@@ -30,8 +26,6 @@ Files:
30
  - `src/tests/monitor-configs/primary.xml`: Two monitors side by side with the primary display set on the second (right-hand) monitor, written as a GNOME monitors.xml display configuration.
31
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
32
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
33
- Reference result: expected_display_configuration
34
- Reference source: the expected configuration struct of meta_test_monitor_store_primary in src/tests/monitor-store-unit-tests.c lines 358-427 (2 logical monitors; DP-1 1024x768 at (0,0) not primary; DP-2 800x600 at (1024,0) primary), checked field by field by check_monitor_store_configuration (check_monitor_store_configuration and check_monitor_store_configurations, lines 122-237: layout rectangle, scale, transform, is_primary, is_presentation, connector, vendor, product, serial, mode width, height, refresh rate and refresh rate mode, flags, underscanning, max bpc, rgb range); the reference file src/tests/monitor-configs/primary.xml is the stored form of the same configuration
35
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
36
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
37
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
@@ -39,8 +33,6 @@ Files:
39
  - `src/tests/monitor-configs/mirrored.xml`: Two monitors mirrored: one logical monitor at (0,0) that shows the same 800x600 picture on DP-1 and DP-2, written as a GNOME monitors.xml display configuration.
40
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
41
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
42
- Reference result: expected_display_configuration
43
- Reference source: the expected configuration struct of meta_test_monitor_store_mirrored in src/tests/monitor-store-unit-tests.c lines 802-856 (1 logical monitor 800x600 at (0,0), primary, holding both DP-1 and DP-2 at 800x600), checked field by field by check_monitor_store_configuration (check_monitor_store_configuration and check_monitor_store_configurations, lines 122-237: layout rectangle, scale, transform, is_primary, is_presentation, connector, vendor, product, serial, mode width, height, refresh rate and refresh rate mode, flags, underscanning, max bpc, rgb range); the reference file src/tests/monitor-configs/mirrored.xml is the stored form of the same configuration
44
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
45
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
46
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
@@ -48,8 +40,6 @@ Files:
48
  - `src/tests/monitor-configs/first-rotated.xml`: Two monitors with the first one rotated to portrait (rotation right) and the second, unrotated, placed to its right at x=768, written as a GNOME monitors.xml display configuration.
49
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
50
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
51
- Reference result: expected_display_configuration
52
- Reference source: the expected configuration struct of meta_test_monitor_store_first_rotated in src/tests/monitor-store-unit-tests.c lines 859-930 (DP-1 1024x768 at (0,0), rotation right (MTK_MONITOR_TRANSFORM_270), logical size 768x1024, primary; DP-2 1024x768 at (768,0) normal), checked field by field by check_monitor_store_configuration (check_monitor_store_configuration and check_monitor_store_configurations, lines 122-237: layout rectangle, scale, transform, is_primary, is_presentation, connector, vendor, product, serial, mode width, height, refresh rate and refresh rate mode, flags, underscanning, max bpc, rgb range); the reference file src/tests/monitor-configs/first-rotated.xml is the stored form of the same configuration
53
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
54
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
55
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
@@ -57,8 +47,6 @@ Files:
57
  - `src/tests/monitor-configs/fractional-scale.xml`: Single 1200x900 monitor with fractional display scaling set to 150 percent, written as a GNOME monitors.xml display configuration.
58
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
59
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
60
- Reference result: expected_display_configuration
61
- Reference source: the expected configuration struct of meta_test_monitor_store_fractional_scale in src/tests/monitor-store-unit-tests.c lines 710-753 (1 logical monitor at (0,0), scale 1.5, logical size 800x600, mode 1200x900, primary), checked field by field by check_monitor_store_configuration (check_monitor_store_configuration and check_monitor_store_configurations, lines 122-237: layout rectangle, scale, transform, is_primary, is_presentation, connector, vendor, product, serial, mode width, height, refresh rate and refresh rate mode, flags, underscanning, max bpc, rgb range); the reference file src/tests/monitor-configs/fractional-scale.xml is the stored form of the same configuration
62
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
63
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
64
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
@@ -66,8 +54,6 @@ Files:
66
  - `src/tests/monitor-configs/underscanning.xml`: Single 1024x768 monitor with underscanning (TV overscan compensation) turned on, written as a GNOME monitors.xml display configuration.
67
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
68
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
69
- Reference result: expected_display_configuration
70
- Reference source: the expected configuration struct of meta_test_monitor_store_underscanning in src/tests/monitor-store-unit-tests.c lines 430-474 (1 logical monitor 1024x768 at (0,0), primary, enable_underscanning TRUE), checked field by field by check_monitor_store_configuration (check_monitor_store_configuration and check_monitor_store_configurations, lines 122-237: layout rectangle, scale, transform, is_primary, is_presentation, connector, vendor, product, serial, mode width, height, refresh rate and refresh rate mode, flags, underscanning, max bpc, rgb range); the reference file src/tests/monitor-configs/underscanning.xml is the stored form of the same configuration
71
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
72
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
73
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
@@ -75,8 +61,6 @@ Files:
75
  - `src/tests/monitor-configs/refresh-rate-mode-variable.xml`: Single 1024x768 monitor with variable refresh rate (VRR) turned on, written as a GNOME monitors.xml display configuration.
76
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
77
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
78
- Reference result: expected_display_configuration
79
- Reference source: the expected configuration struct of meta_test_monitor_store_refresh_rate_mode_variable in src/tests/monitor-store-unit-tests.c lines 524-568 (1 logical monitor 1024x768 at (0,0), primary, refresh rate 60.000495910644531 with refresh_rate_mode variable), checked field by field by check_monitor_store_configuration (check_monitor_store_configuration and check_monitor_store_configurations, lines 122-237: layout rectangle, scale, transform, is_primary, is_presentation, connector, vendor, product, serial, mode width, height, refresh rate and refresh rate mode, flags, underscanning, max bpc, rgb range); the reference file src/tests/monitor-configs/refresh-rate-mode-variable.xml is the stored form of the same configuration
80
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
81
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
82
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
@@ -84,8 +68,6 @@ Files:
84
  - `src/tests/monitor-configs/rgb-range.xml`: Single 1024x768 monitor with the RGB range forced to limited, written as a GNOME monitors.xml display configuration.
85
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
86
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
87
- Reference result: expected_display_configuration
88
- Reference source: the expected configuration struct of meta_test_monitor_store_rgb_range in src/tests/monitor-store-unit-tests.c lines 618-661 (1 logical monitor 1024x768 at (0,0), primary, rgb_range limited), checked field by field by check_monitor_store_configuration (check_monitor_store_configuration and check_monitor_store_configurations, lines 122-237: layout rectangle, scale, transform, is_primary, is_presentation, connector, vendor, product, serial, mode width, height, refresh rate and refresh rate mode, flags, underscanning, max bpc, rgb range); the reference file src/tests/monitor-configs/rgb-range.xml is the stored form of the same configuration
89
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
90
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
91
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
@@ -93,8 +75,6 @@ Files:
93
  - `src/tests/monitor-configs/second-rotated.xml`: Two monitors with the second one rotated to portrait (rotation left) and placed to the right, the first one lowered to y=256, written as a GNOME monitors.xml display configuration.
94
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
95
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
96
- Reference result: expected_display_configuration
97
- Reference source: the expected configuration struct of meta_test_monitor_store_second_rotated in src/tests/monitor-store-unit-tests.c lines 933-1002 (DP-1 1024x768 at (0,256), normal, primary; DP-2 1024x768 at (1024,0), rotation left (MTK_MONITOR_TRANSFORM_90), logical size 768x1024), checked field by field by check_monitor_store_configuration (check_monitor_store_configuration and check_monitor_store_configurations, lines 122-237: layout rectangle, scale, transform, is_primary, is_presentation, connector, vendor, product, serial, mode width, height, refresh rate and refresh rate mode, flags, underscanning, max bpc, rgb range); the reference file src/tests/monitor-configs/second-rotated.xml is the stored form of the same configuration
98
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
99
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
100
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
@@ -102,8 +82,6 @@ Files:
102
  - `src/tests/monitor-configs/scale.xml`: Single 1920x1080 monitor with display scaling set to 200 percent, written as a GNOME monitors.xml display configuration.
103
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
104
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
105
- Reference result: expected_display_configuration
106
- Reference source: the expected configuration struct of meta_test_monitor_store_scale in src/tests/monitor-store-unit-tests.c lines 664-705 (1 logical monitor at (0,0), scale 2, logical size 960x540, mode 1920x1080, primary), checked field by field by check_monitor_store_configuration (check_monitor_store_configuration and check_monitor_store_configurations, lines 122-237: layout rectangle, scale, transform, is_primary, is_presentation, connector, vendor, product, serial, mode width, height, refresh rate and refresh rate mode, flags, underscanning, max bpc, rgb range); the reference file src/tests/monitor-configs/scale.xml is the stored form of the same configuration
107
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
108
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
109
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
 
8
 
9
  **Self-contained:** Whether each file needs other upstream files is not stated. 10 of 10 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
12
 
13
  **Licence (SPDX):** GPL-2.0-or-later
 
19
  - `src/tests/monitor-configs/vertical.xml`: Two monitors stacked vertically: a 1024x768 primary on top at (0,0) and an 800x600 monitor below it at (0,768), logical layout mode, written as a GNOME monitors.xml display configuration.
20
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
21
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
 
 
22
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
23
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
24
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
 
26
  - `src/tests/monitor-configs/primary.xml`: Two monitors side by side with the primary display set on the second (right-hand) monitor, written as a GNOME monitors.xml display configuration.
27
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
28
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
 
 
29
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
30
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
31
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
 
33
  - `src/tests/monitor-configs/mirrored.xml`: Two monitors mirrored: one logical monitor at (0,0) that shows the same 800x600 picture on DP-1 and DP-2, written as a GNOME monitors.xml display configuration.
34
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
35
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
 
 
36
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
37
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
38
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
 
40
  - `src/tests/monitor-configs/first-rotated.xml`: Two monitors with the first one rotated to portrait (rotation right) and the second, unrotated, placed to its right at x=768, written as a GNOME monitors.xml display configuration.
41
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
42
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
 
 
43
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
44
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
45
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
 
47
  - `src/tests/monitor-configs/fractional-scale.xml`: Single 1200x900 monitor with fractional display scaling set to 150 percent, written as a GNOME monitors.xml display configuration.
48
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
49
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
 
 
50
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
51
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
52
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
 
54
  - `src/tests/monitor-configs/underscanning.xml`: Single 1024x768 monitor with underscanning (TV overscan compensation) turned on, written as a GNOME monitors.xml display configuration.
55
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
56
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
 
 
57
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
58
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
59
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
 
61
  - `src/tests/monitor-configs/refresh-rate-mode-variable.xml`: Single 1024x768 monitor with variable refresh rate (VRR) turned on, written as a GNOME monitors.xml display configuration.
62
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
63
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
 
 
64
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
65
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
66
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
 
68
  - `src/tests/monitor-configs/rgb-range.xml`: Single 1024x768 monitor with the RGB range forced to limited, written as a GNOME monitors.xml display configuration.
69
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
70
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
 
 
71
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
72
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
73
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
 
75
  - `src/tests/monitor-configs/second-rotated.xml`: Two monitors with the second one rotated to portrait (rotation left) and placed to the right, the first one lowered to y=256, written as a GNOME monitors.xml display configuration.
76
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
77
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
 
 
78
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
79
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
80
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
 
82
  - `src/tests/monitor-configs/scale.xml`: Single 1920x1080 monitor with display scaling set to 200 percent, written as a GNOME monitors.xml display configuration.
83
  Format: GNOME Mutter monitors.xml (XML, <monitors version="2">, configuration, logicalmonitor, monitor, monitorspec, mode elements)
84
  Origin: monitors.xml fixture of Mutter's monitor config store unit tests (src/tests/monitor-configs), loaded by monitor-store-unit-tests.c through meta_set_custom_monitor_config (tool_fixture)
 
 
85
  Check: Final configuration state, no screenshot needed. Parse the monitors.xml the agent leaves (~/.config/monitors.xml or the path the task names) with an XML parser and compare it with the expected struct: the same number of configurations and logical monitors, and for each logical monitor the same x, y, scale, rotation and flip, primary and presentation flags, and for each monitor the same connector, vendor, product, serial, mode width, height, rate, rate mode, underscanning, max bpc and rgb range. All comparisons are exact (upstream uses g_assert_cmpint, g_assert_cmpstr and g_assert_cmpfloat with ==); omitted optional elements take Mutter's defaults as the struct shows (scale 1, rotation normal, rgb range auto). Element order and whitespace do not count.
86
  Tools: write ~/.config/monitors.xml directly with a text editor or a script (Python xml.etree.ElementTree or lxml), or apply the layout on a running GNOME session with gdctl (`gdctl set --logical-monitor --primary --monitor DP-1 ...`) or the org.gnome.Mutter.DisplayConfig D-Bus method ApplyMonitorsConfig (busctl or gdbus), which makes Mutter write monitors.xml
87
  Companions (not included): src/tests/monitor-store-unit-tests.c; COPYING
campaign_files/2026-10-m-issue-trackers-2/mongodb-jira/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** The file is self-contained.
10
 
11
- **Answers:** No answer files are included. Reference result: Facts computed from the ticket changelog and comments. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** curl the Jira REST API v2 with a fields= list and expand=changelog, read with jq or Python json; or the jira Python library.
14
 
15
  **Licence (SPDX):** NOASSERTION
@@ -21,7 +19,5 @@ Files:
21
  - `m027-04/SERVER-4401.json`: MongoDB Jira SERVER-4401: query optimizer cursor may return partial result set (Closed, Done, 2011-12-24).
22
  Format: Jira REST API v2 issue JSON restricted to summary, status, resolution, resolutiondate, created, updated, assignee, reporter, priority, issuetype, comment, description, components, fixVersions, versions, labels, with expand=changelog.
23
  Origin: The full response (no fields= filter) changed between two fetches only in volatile custom fields (MariaDB customfield_11700 and XWiki customfield_11170: development-status summaries with fresh object hashes; MongoDB customfield_28251: a float that keeps increasing). The pointer therefore names a fields-restricted URL (the standard fields plus expand=changelog), which came back byte-identical in two fetches about 10 minutes apart.
24
- Reference result: facts computed from the ticket changelog and comments
25
- Reference source: changelog.histories[] and fields.comment.comments[] of the pinned response.
26
  Check: Exact match against values read from the pinned JSON: changelog.histories[].items[] with field 'status' or 'assignee' (fromString, toString, created, author.displayName) and fields.comment.comments[] (author, created). Deterministic, no model judgement. Before answers are issued, the extraction program must pass on 2 hand-checked tickets of this tracker (spec general rule 5); hand-checking is pending for the delivery step.
27
  Tools: curl the Jira REST API v2 with a fields= list and expand=changelog, read with jq or Python json; or the jira Python library.
 
8
 
9
  **Self-contained:** The file is self-contained.
10
 
 
 
11
  **Tools to open and check:** curl the Jira REST API v2 with a fields= list and expand=changelog, read with jq or Python json; or the jira Python library.
12
 
13
  **Licence (SPDX):** NOASSERTION
 
19
  - `m027-04/SERVER-4401.json`: MongoDB Jira SERVER-4401: query optimizer cursor may return partial result set (Closed, Done, 2011-12-24).
20
  Format: Jira REST API v2 issue JSON restricted to summary, status, resolution, resolutiondate, created, updated, assignee, reporter, priority, issuetype, comment, description, components, fixVersions, versions, labels, with expand=changelog.
21
  Origin: The full response (no fields= filter) changed between two fetches only in volatile custom fields (MariaDB customfield_11700 and XWiki customfield_11170: development-status summaries with fresh object hashes; MongoDB customfield_28251: a float that keeps increasing). The pointer therefore names a fields-restricted URL (the standard fields plus expand=changelog), which came back byte-identical in two fetches about 10 minutes apart.
 
 
22
  Check: Exact match against values read from the pinned JSON: changelog.histories[].items[] with field 'status' or 'assignee' (fromString, toString, created, author.displayName) and fields.comment.comments[] (author, created). Deterministic, no model judgement. Before answers are issued, the extraction program must pass on 2 hand-checked tickets of this tracker (spec general rule 5); hand-checking is pending for the delivery step.
23
  Tools: curl the Jira REST API v2 with a fields= list and expand=changelog, read with jq or Python json; or the jira Python library.
campaign_files/2026-10-m-issue-trackers-2/morebus-passenger-gtfs/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** The unit is self-contained. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: Facts computed from the service calendar itself (active services on a date, removed or added dates, day counts). Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Read the CSV files with Python csv or pandas (or sqlite3 .import) and expand service days per GTFS rules; gtfs-kit or partridge compute active service_ids per date headless; dates are local to agency_timezone (Python zoneinfo).
14
 
15
  **Licence (SPDX):** OGL-UK-3.0
@@ -22,8 +20,6 @@ Units:
22
  Read out of the archive: https://s3-eu-west-1.amazonaws.com/passenger-sources/morebus/gtfs/morebus_1791388307.zip
23
  Format: text/csv (GTFS Schedule calendar.txt; companions calendar_dates.txt, agency.txt)
24
  Origin: Live static GTFS feed published by Morebus (published through Passenger, data.discoverpassenger.com) at https://s3-eu-west-1.amazonaws.com/passenger-sources/morebus/gtfs/morebus_1791388307.zip. calendar.txt and its companions were read by member out of the zip with HTTP Range requests (spec general rule 4: by member); kept as a snapshot pinned by sha256.
25
- Reference result: facts computed from the service calendar itself (active services on a date, removed or added dates, day counts)
26
- Reference source: calendar.txt weekday flags with start_date and end_date, calendar_dates.txt exception rows, agency_timezone of agency.txt, all from the pinned members.
27
  Check: deterministic, no model judgement. Hand-checking 2 samples per source is pending for the delivery step (spec general rule 5).
28
  Tools: Read the CSV files with Python csv or pandas (or sqlite3 .import) and expand service days per GTFS rules; gtfs-kit or partridge compute active service_ids per date headless; dates are local to agency_timezone (Python zoneinfo).
29
  Companions (not included): calendar_dates.txt; agency.txt
 
8
 
9
  **Self-contained:** The unit is self-contained. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Read the CSV files with Python csv or pandas (or sqlite3 .import) and expand service days per GTFS rules; gtfs-kit or partridge compute active service_ids per date headless; dates are local to agency_timezone (Python zoneinfo).
12
 
13
  **Licence (SPDX):** OGL-UK-3.0
 
20
  Read out of the archive: https://s3-eu-west-1.amazonaws.com/passenger-sources/morebus/gtfs/morebus_1791388307.zip
21
  Format: text/csv (GTFS Schedule calendar.txt; companions calendar_dates.txt, agency.txt)
22
  Origin: Live static GTFS feed published by Morebus (published through Passenger, data.discoverpassenger.com) at https://s3-eu-west-1.amazonaws.com/passenger-sources/morebus/gtfs/morebus_1791388307.zip. calendar.txt and its companions were read by member out of the zip with HTTP Range requests (spec general rule 4: by member); kept as a snapshot pinned by sha256.
 
 
23
  Check: deterministic, no model judgement. Hand-checking 2 samples per source is pending for the delivery step (spec general rule 5).
24
  Tools: Read the CSV files with Python csv or pandas (or sqlite3 .import) and expand service days per GTFS rules; gtfs-kit or partridge compute active service_ids per date headless; dates are local to agency_timezone (Python zoneinfo).
25
  Companions (not included): calendar_dates.txt; agency.txt
campaign_files/2026-10-m-issue-trackers-2/mozilla-holiday-calendars/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Each file is self-contained.
10
 
11
- **Answers:** No answer files are included. Reference result: Facts computed from the calendar itself (recurrence expansion, exclusions, overrides, time zone conversion, counts). Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Parse and expand the .ics with Python icalendar plus recurring-ical-events, or with libical tools; khal can import and list it headless (khal list); time zone conversion with Python zoneinfo.
14
 
15
  **Licence (SPDX):** NOASSERTION
@@ -21,21 +19,15 @@ Files:
21
  - `media/caldata/SriLankaHolidays.ics`: Sri Lanka holidays calendar (Thunderbird holiday calendars, caldata).
22
  Format: text/calendar (iCalendar, RFC 5545)
23
  Origin: Holiday calendar contributed to Mozilla's Thunderbird holiday-calendar collection (calendars.json credits named contributors); served on thunderbird.net. Event SUMMARY values are English; some LOCATION values are in Sinhala.
24
- Reference result: facts computed from the calendar itself (recurrence expansion, exclusions, overrides, time zone conversion, counts)
25
- Reference source: VEVENT properties of the pinned file: DTSTART/DTEND with TZID, RRULE, EXDATE, RECURRENCE-ID, SUMMARY, UID, and the VTIMEZONE definitions.
26
  Check: Exact match against values computed from the pinned file by an open-source RFC 5545 expander (Python icalendar plus recurring-ical-events, cross-checked with libical): occurrence dates and local start times, presence or absence on a date, and counts. Deterministic, no model judgement. Hand-checking 2 samples per source is pending for the delivery step (spec general rule 5).
27
  Tools: Parse and expand the .ics with Python icalendar plus recurring-ical-events, or with libical tools; khal can import and list it headless (khal list); time zone conversion with Python zoneinfo.
28
  - `media/caldata/ThaiHolidays.ics`: Thai holidays calendar (Thunderbird holiday calendars, caldata).
29
  Format: text/calendar (iCalendar, RFC 5545)
30
  Origin: Holiday calendar contributed to Mozilla's Thunderbird holiday-calendar collection (calendars.json credits named contributors); served on thunderbird.net. English.
31
- Reference result: facts computed from the calendar itself (recurrence expansion, exclusions, overrides, time zone conversion, counts)
32
- Reference source: VEVENT properties of the pinned file: DTSTART/DTEND with TZID, RRULE, EXDATE, RECURRENCE-ID, SUMMARY, UID, and the VTIMEZONE definitions.
33
  Check: Exact match against values computed from the pinned file by an open-source RFC 5545 expander (Python icalendar plus recurring-ical-events, cross-checked with libical): occurrence dates and local start times, presence or absence on a date, and counts. Deterministic, no model judgement. Hand-checking 2 samples per source is pending for the delivery step (spec general rule 5).
34
  Tools: Parse and expand the .ics with Python icalendar plus recurring-ical-events, or with libical tools; khal can import and list it headless (khal list); time zone conversion with Python zoneinfo.
35
  - `projects/calendar/caldata/MalaysiaHolidays.ics`: Malaysia holidays calendar, 2007 to 2008 edition (archived mozilla.org calendar project, caldata).
36
  Format: text/calendar (iCalendar, RFC 5545)
37
  Origin: Holiday calendar from the old mozilla.org Calendar project holiday collection, as archived in mozilla-it/www-archive.mozilla.org (commit 'moving content to GitHub', 2017). A different, older edition than the Thunderbird caldata Malaysia file (49 events, no EXDATE). English.
38
- Reference result: facts computed from the calendar itself (recurrence expansion, exclusions, overrides, time zone conversion, counts)
39
- Reference source: VEVENT properties of the pinned file: DTSTART/DTEND with TZID, RRULE, EXDATE, RECURRENCE-ID, SUMMARY, UID, and the VTIMEZONE definitions.
40
  Check: Exact match against values computed from the pinned file by an open-source RFC 5545 expander (Python icalendar plus recurring-ical-events, cross-checked with libical): occurrence dates and local start times, presence or absence on a date, and counts. Deterministic, no model judgement. Hand-checking 2 samples per source is pending for the delivery step (spec general rule 5).
41
  Tools: Parse and expand the .ics with Python icalendar plus recurring-ical-events, or with libical tools; khal can import and list it headless (khal list); time zone conversion with Python zoneinfo.
 
8
 
9
  **Self-contained:** Each file is self-contained.
10
 
 
 
11
  **Tools to open and check:** Parse and expand the .ics with Python icalendar plus recurring-ical-events, or with libical tools; khal can import and list it headless (khal list); time zone conversion with Python zoneinfo.
12
 
13
  **Licence (SPDX):** NOASSERTION
 
19
  - `media/caldata/SriLankaHolidays.ics`: Sri Lanka holidays calendar (Thunderbird holiday calendars, caldata).
20
  Format: text/calendar (iCalendar, RFC 5545)
21
  Origin: Holiday calendar contributed to Mozilla's Thunderbird holiday-calendar collection (calendars.json credits named contributors); served on thunderbird.net. Event SUMMARY values are English; some LOCATION values are in Sinhala.
 
 
22
  Check: Exact match against values computed from the pinned file by an open-source RFC 5545 expander (Python icalendar plus recurring-ical-events, cross-checked with libical): occurrence dates and local start times, presence or absence on a date, and counts. Deterministic, no model judgement. Hand-checking 2 samples per source is pending for the delivery step (spec general rule 5).
23
  Tools: Parse and expand the .ics with Python icalendar plus recurring-ical-events, or with libical tools; khal can import and list it headless (khal list); time zone conversion with Python zoneinfo.
24
  - `media/caldata/ThaiHolidays.ics`: Thai holidays calendar (Thunderbird holiday calendars, caldata).
25
  Format: text/calendar (iCalendar, RFC 5545)
26
  Origin: Holiday calendar contributed to Mozilla's Thunderbird holiday-calendar collection (calendars.json credits named contributors); served on thunderbird.net. English.
 
 
27
  Check: Exact match against values computed from the pinned file by an open-source RFC 5545 expander (Python icalendar plus recurring-ical-events, cross-checked with libical): occurrence dates and local start times, presence or absence on a date, and counts. Deterministic, no model judgement. Hand-checking 2 samples per source is pending for the delivery step (spec general rule 5).
28
  Tools: Parse and expand the .ics with Python icalendar plus recurring-ical-events, or with libical tools; khal can import and list it headless (khal list); time zone conversion with Python zoneinfo.
29
  - `projects/calendar/caldata/MalaysiaHolidays.ics`: Malaysia holidays calendar, 2007 to 2008 edition (archived mozilla.org calendar project, caldata).
30
  Format: text/calendar (iCalendar, RFC 5545)
31
  Origin: Holiday calendar from the old mozilla.org Calendar project holiday collection, as archived in mozilla-it/www-archive.mozilla.org (commit 'moving content to GitHub', 2017). A different, older edition than the Thunderbird caldata Malaysia file (49 events, no EXDATE). English.
 
 
32
  Check: Exact match against values computed from the pinned file by an open-source RFC 5545 expander (Python icalendar plus recurring-ical-events, cross-checked with libical): occurrence dates and local start times, presence or absence on a date, and counts. Deterministic, no model judgement. Hand-checking 2 samples per source is pending for the delivery step (spec general rule 5).
33
  Tools: Parse and expand the .ics with Python icalendar plus recurring-ical-events, or with libical tools; khal can import and list it headless (khal list); time zone conversion with Python zoneinfo.
campaign_files/2026-10-m-issue-trackers-2/mvgo-gtfs/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** The unit is self-contained. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: Facts computed from the service calendar itself (active services on a date, removed or added dates, day counts). Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Read the CSV files with Python csv or pandas (or sqlite3 .import) and expand service days per GTFS rules; gtfs-kit or partridge compute active service_ids per date headless; dates are local to agency_timezone (Python zoneinfo).
14
 
15
  **Licence (SPDX):** ODC-By-1.0
@@ -22,8 +20,6 @@ Units:
22
  Read out of the archive: https://gtfs.mvgo.org/gtfs.zip
23
  Format: text/csv (GTFS Schedule calendar.txt; companions calendar_dates.txt, agency.txt, feed_info.txt)
24
  Origin: Live static GTFS feed published by Mountain View Transportation Management Association (MVgo), feed produced by TripShot at https://gtfs.mvgo.org/gtfs.zip. calendar.txt and its companions were read by member out of the zip with HTTP Range requests (spec general rule 4: by member); kept as a snapshot pinned by sha256.
25
- Reference result: facts computed from the service calendar itself (active services on a date, removed or added dates, day counts)
26
- Reference source: calendar.txt weekday flags with start_date and end_date, calendar_dates.txt exception rows, agency_timezone of agency.txt, all from the pinned members.
27
  Check: deterministic, no model judgement. Hand-checking 2 samples per source is pending for the delivery step (spec general rule 5).
28
  Tools: Read the CSV files with Python csv or pandas (or sqlite3 .import) and expand service days per GTFS rules; gtfs-kit or partridge compute active service_ids per date headless; dates are local to agency_timezone (Python zoneinfo).
29
  Companions (not included): calendar_dates.txt; agency.txt; feed_info.txt
 
8
 
9
  **Self-contained:** The unit is self-contained. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Read the CSV files with Python csv or pandas (or sqlite3 .import) and expand service days per GTFS rules; gtfs-kit or partridge compute active service_ids per date headless; dates are local to agency_timezone (Python zoneinfo).
12
 
13
  **Licence (SPDX):** ODC-By-1.0
 
20
  Read out of the archive: https://gtfs.mvgo.org/gtfs.zip
21
  Format: text/csv (GTFS Schedule calendar.txt; companions calendar_dates.txt, agency.txt, feed_info.txt)
22
  Origin: Live static GTFS feed published by Mountain View Transportation Management Association (MVgo), feed produced by TripShot at https://gtfs.mvgo.org/gtfs.zip. calendar.txt and its companions were read by member out of the zip with HTTP Range requests (spec general rule 4: by member); kept as a snapshot pinned by sha256.
 
 
23
  Check: deterministic, no model judgement. Hand-checking 2 samples per source is pending for the delivery step (spec general rule 5).
24
  Tools: Read the CSV files with Python csv or pandas (or sqlite3 .import) and expand service days per GTFS rules; gtfs-kit or partridge compute active service_ids per date headless; dates are local to agency_timezone (Python zoneinfo).
25
  Companions (not included): calendar_dates.txt; agency.txt; feed_info.txt
campaign_files/2026-10-m-issue-trackers-2/mythtv-mythtv-users/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** The file is self-contained.
10
 
11
- **Answers:** No answer files are included. Reference result: Facts computed from the thread headers. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Read the archive as mbox with Python's mailbox and email modules (or mutt -f / notmuch on the decompressed file); rebuild the thread from Message-ID, In-Reply-To and References headers.
14
 
15
  **Licence (SPDX):** NOASSERTION
@@ -21,7 +19,5 @@ Files:
21
  - `m141-07/2019-March.txt.gz`: [mythtv-users] Slave backend reboot procedure (11 messages, 2019-03-23 to 2019-03-26).
22
  Format: Pipermail month archive, mbox text gzip-compressed: full headers (From with obfuscated address and name, Date, Subject, Message-ID, In-Reply-To, References) and plain-text bodies. Cut the thread by root Message-ID through the In-Reply-To closure (last References entry when In-Reply-To is missing).
23
  Origin: Real list traffic in a closed past month of the list's Pipermail archive. The unit is one thread, cut out of the month file by its root Message-ID; the month file is the pinned pointer and is delivered whole.
24
- Reference result: facts computed from the thread headers
25
- Reference source: Message-ID, In-Reply-To, References, From and Date headers of the messages in the thread, read from the pinned month file.
26
  Check: Exact match against values computed from the pinned month file: parse it as mbox, drop repeated copies of a Message-ID already seen (first copy kept), take the root Message-ID, collect every message whose In-Reply-To (or last References entry when In-Reply-To is missing) chains back to it, then count, take dates in UTC and compare sender names. Message-IDs are compared after removing angle brackets and double quotes. People are counted by person: From entries with the same address, or the same name after removing accents and letter case, are one person (the names in scale are the first spelling of each). The first-reply interval is the exact difference in whole seconds between the Date headers of the root and the earliest reply, converted to UTC; Date headers have one-second resolution, so no rounding is involved. Deterministic. The program was checked on 2 threads of this source by a second, independent parser (splits the raw text on mbox From lines and reads headers with regular expressions, no mailbox or email module): the chosen thread (also read message by message) and '[mythtv-users] CTRL-ALT-BACKSPACE' (44 messages, root <CAH8UtKM0onii=2uZMsJ4gzyN_aNOPJD03QfHf2gScjVhANt56Q@mail.gmail.com>); message count, people, last sender and time, first-reply seconds and direct replies agreed on both.
27
  Tools: Read the archive as mbox with Python's mailbox and email modules (or mutt -f / notmuch on the decompressed file); rebuild the thread from Message-ID, In-Reply-To and References headers.
 
8
 
9
  **Self-contained:** The file is self-contained.
10
 
 
 
11
  **Tools to open and check:** Read the archive as mbox with Python's mailbox and email modules (or mutt -f / notmuch on the decompressed file); rebuild the thread from Message-ID, In-Reply-To and References headers.
12
 
13
  **Licence (SPDX):** NOASSERTION
 
19
  - `m141-07/2019-March.txt.gz`: [mythtv-users] Slave backend reboot procedure (11 messages, 2019-03-23 to 2019-03-26).
20
  Format: Pipermail month archive, mbox text gzip-compressed: full headers (From with obfuscated address and name, Date, Subject, Message-ID, In-Reply-To, References) and plain-text bodies. Cut the thread by root Message-ID through the In-Reply-To closure (last References entry when In-Reply-To is missing).
21
  Origin: Real list traffic in a closed past month of the list's Pipermail archive. The unit is one thread, cut out of the month file by its root Message-ID; the month file is the pinned pointer and is delivered whole.
 
 
22
  Check: Exact match against values computed from the pinned month file: parse it as mbox, drop repeated copies of a Message-ID already seen (first copy kept), take the root Message-ID, collect every message whose In-Reply-To (or last References entry when In-Reply-To is missing) chains back to it, then count, take dates in UTC and compare sender names. Message-IDs are compared after removing angle brackets and double quotes. People are counted by person: From entries with the same address, or the same name after removing accents and letter case, are one person (the names in scale are the first spelling of each). The first-reply interval is the exact difference in whole seconds between the Date headers of the root and the earliest reply, converted to UTC; Date headers have one-second resolution, so no rounding is involved. Deterministic. The program was checked on 2 threads of this source by a second, independent parser (splits the raw text on mbox From lines and reads headers with regular expressions, no mailbox or email module): the chosen thread (also read message by message) and '[mythtv-users] CTRL-ALT-BACKSPACE' (44 messages, root <CAH8UtKM0onii=2uZMsJ4gzyN_aNOPJD03QfHf2gScjVhANt56Q@mail.gmail.com>); message count, people, last sender and time, first-reply seconds and direct replies agreed on both.
23
  Tools: Read the archive as mbox with Python's mailbox and email modules (or mutt -f / notmuch on the decompressed file); rebuild the thread from Message-ID, In-Reply-To and References headers.
campaign_files/2026-10-md-media-and-documents/mmark-testdata-table/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: Expected RFC 7991 XML fragment for one Markdown table. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** mmark command line program (mmark table.md, XML fragment output) or the Go package; the XML can be checked with xml2rfc
14
 
15
  **Licence (SPDX):** BSD-2-Clause
@@ -21,8 +19,6 @@ Files:
21
  - `testdata/table.md`: Mmark test: a Markdown table with caption converted to RFC 7991 XML.
22
  Format: Markdown input file and an XML file with the same base name
23
  Origin: test data written by the mmark author (Markdown input and the RFC 7991 XML mmark must produce); not copied from another suite
24
- Reference result: expected RFC 7991 XML fragment for one Markdown table
25
- Reference source: testdata/table.xml
26
  Check: show the agent the .md only; the agent writes the XML fragment as mmark does (xml.CommonFlags plus xml.XMLFragment, language en since the file name has no prefix); trim surrounding white space on both sides and pass on byte equality
27
  Tools: mmark command line program (mmark table.md, XML fragment output) or the Go package; the XML can be checked with xml2rfc
28
  Companions (not included): testdata/table.xml; mmark_test.go; LICENSE
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** mmark command line program (mmark table.md, XML fragment output) or the Go package; the XML can be checked with xml2rfc
12
 
13
  **Licence (SPDX):** BSD-2-Clause
 
19
  - `testdata/table.md`: Mmark test: a Markdown table with caption converted to RFC 7991 XML.
20
  Format: Markdown input file and an XML file with the same base name
21
  Origin: test data written by the mmark author (Markdown input and the RFC 7991 XML mmark must produce); not copied from another suite
 
 
22
  Check: show the agent the .md only; the agent writes the XML fragment as mmark does (xml.CommonFlags plus xml.XMLFragment, language en since the file name has no prefix); trim surrounding white space on both sides and pass on byte equality
23
  Tools: mmark command line program (mmark table.md, XML fragment output) or the Go package; the XML can be checked with xml2rfc
24
  Companions (not included): testdata/table.xml; mmark_test.go; LICENSE
campaign_files/2026-10-md-media-and-documents/myst-parser-renderer-fixtures-tables/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: Expected docutils pseudo-XML document tree for each table example. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Sphinx with the myst_parser extension (sphinx-build) or the myst-docutils command line myst-docutils-pseudoxml page.md; pandoc -f markdown -t native gives a comparable document tree
14
 
15
  **Licence (SPDX):** MIT
@@ -21,8 +19,6 @@ Files:
21
  - `tests/test_renderers/fixtures/tables.md`: MyST-Parser renderer fixtures "tables": Markdown tables parsed into the docutils document tree (pseudo-XML).
22
  Format: pytest-param-files fixture file: a title line, ".", the MyST Markdown input, ".", the expected pseudo-XML, "." per example (15 "." lines, 5 examples)
23
  Origin: renderer fixture written by the MyST-Parser maintainers for Markdown tables parsed into the docutils document tree
24
- Reference result: expected docutils pseudo-XML document tree for each table example
25
- Reference source: the part between the second and third "." of each example
26
  Check: show only the Markdown of each example (hide the pseudo-XML); the agent produces the docutils pseudo-XML that Sphinx with myst_parser yields for index.md; normalise boolean attributes "1"/"0" to "True"/"False" as normalize_doctree_xml does for docutils 0.22 and later; rstrip every line of both sides and the whole text, and pass when they are equal
27
  Tools: Sphinx with the myst_parser extension (sphinx-build) or the myst-docutils command line myst-docutils-pseudoxml page.md; pandoc -f markdown -t native gives a comparable document tree
28
  Companions (not included): tests/test_renderers/test_fixtures_sphinx.py; tests/conftest.py
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Sphinx with the myst_parser extension (sphinx-build) or the myst-docutils command line myst-docutils-pseudoxml page.md; pandoc -f markdown -t native gives a comparable document tree
12
 
13
  **Licence (SPDX):** MIT
 
19
  - `tests/test_renderers/fixtures/tables.md`: MyST-Parser renderer fixtures "tables": Markdown tables parsed into the docutils document tree (pseudo-XML).
20
  Format: pytest-param-files fixture file: a title line, ".", the MyST Markdown input, ".", the expected pseudo-XML, "." per example (15 "." lines, 5 examples)
21
  Origin: renderer fixture written by the MyST-Parser maintainers for Markdown tables parsed into the docutils document tree
 
 
22
  Check: show only the Markdown of each example (hide the pseudo-XML); the agent produces the docutils pseudo-XML that Sphinx with myst_parser yields for index.md; normalise boolean attributes "1"/"0" to "True"/"False" as normalize_doctree_xml does for docutils 0.22 and later; rstrip every line of both sides and the whole text, and pass when they are equal
23
  Tools: Sphinx with the myst_parser extension (sphinx-build) or the myst-docutils command line myst-docutils-pseudoxml page.md; pandoc -f markdown -t native gives a comparable document tree
24
  Companions (not included): tests/test_renderers/test_fixtures_sphinx.py; tests/conftest.py
campaign_files/2026-10-md-media-and-documents/myst-to-typst-tests/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: Expected Typst source for each mdast table case (4 cases). Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** myst-to-typst (Node.js, unified plugin) on the mdast tree, or myst build --typst from the MyST CLI; output compared by string equality
14
 
15
  **Licence (SPDX):** MIT
@@ -21,8 +19,6 @@ Files:
21
  - `packages/myst-to-typst/tests/tables.yml`: Myst-to-typst test cases: Markdown AST tables converted to Typst tablex source.
22
  Format: YAML file: title and a list of cases, each with title, mdast (JSON-like tree) and typst (expected string)
23
  Origin: YAML test cases written by the MyST maintainers: each case holds an mdast tree and the expected Typst source; run by run.spec.ts
24
- Reference result: expected Typst source for each mdast table case (4 cases)
25
- Reference source: the typst field of each case in tables.yml
26
  Check: give the agent the mdast tree of a case (or its Markdown equivalent) and ask for the Typst source myst-to-typst writes; compare with the typst field by exact string equality (vitest toEqual); one score per case
27
  Tools: myst-to-typst (Node.js, unified plugin) on the mdast tree, or myst build --typst from the MyST CLI; output compared by string equality
28
  Companions (not included): packages/myst-to-typst/tests/run.spec.ts; packages/myst-to-typst/package.json; LICENSE
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** myst-to-typst (Node.js, unified plugin) on the mdast tree, or myst build --typst from the MyST CLI; output compared by string equality
12
 
13
  **Licence (SPDX):** MIT
 
19
  - `packages/myst-to-typst/tests/tables.yml`: Myst-to-typst test cases: Markdown AST tables converted to Typst tablex source.
20
  Format: YAML file: title and a list of cases, each with title, mdast (JSON-like tree) and typst (expected string)
21
  Origin: YAML test cases written by the MyST maintainers: each case holds an mdast tree and the expected Typst source; run by run.spec.ts
 
 
22
  Check: give the agent the mdast tree of a case (or its Markdown equivalent) and ask for the Typst source myst-to-typst writes; compare with the typst field by exact string equality (vitest toEqual); one score per case
23
  Tools: myst-to-typst (Node.js, unified plugin) on the mdast tree, or myst build --typst from the MyST CLI; output compared by string equality
24
  Companions (not included): packages/myst-to-typst/tests/run.spec.ts; packages/myst-to-typst/package.json; LICENSE
campaign_files/2026-10-ml-serving-configs-2/mitre-menelaus/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether each file needs other upstream files is not stated. 2 of 2 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: labelled_drift_point. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: The agent streams the rows in order and returns the index of its first drift alarm (and optionally all alarms). Pass when the first alarm index lies in the labelled drift range 1000 to 1250 inclusive; an alarm before index 1000 is a false alarm and fails; no alarm at all fails; The agent streams the rows in order and returns the index of its first drift alarm. Pass when the first alarm index is 12000 or later; an alarm before index 12000 is a false alarm and fails; no alarm at all fails. Detection delay (alarm index minus 12000) is reported but not cut off.
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -21,16 +19,12 @@ Files:
21
  - `menelaus/datasets/dataCircleGSev3Sp3Train.csv`: Menelaus Circle drift data (dataCircleGSev3Sp3Train.csv): 2000 rows, two features var1,var2 and a 0/1 target y; labelled drift at index 1000 to 1250 affecting 66 percent of points.
22
  Format: CSV without header, columns var1,var2,y and a trailing empty field (read with usecols=[0, 1, 2], make_example_data.py line 128)
23
  Origin: synthetic Circle benchmark of Minku et al. 2010 (cited as minku2010 at make_example_data.py line 122), redistributed by MITRE in menelaus; the file name says Train; upstream generation terms not checked
24
- Reference result: labelled_drift_point
25
- Reference source: make_example_data.py lines 116 to 128 (fetch_circle_data: "Drift occurs from index 1000-1250 and affects 66% of the data points", line 120; Ref. minku2010, line 122)
26
  Check: The agent streams the rows in order and returns the index of its first drift alarm (and optionally all alarms). Pass when the first alarm index lies in the labelled drift range 1000 to 1250 inclusive; an alarm before index 1000 is a false alarm and fails; no alarm at all fails
27
  Companions (not included): menelaus/datasets/make_example_data.py; LICENSE.txt
28
  Dependencies: none to read the data (plain CSV); the drift label is text in the companion's docstring.
29
  - `menelaus/datasets/rainfall_data.csv`: Menelaus Rainfall drift data (rainfall_data.csv): 18159 daily rows from NOAA at Bellevue, Nebraska, eight standardised weather features and a 0/1 rain target; labelled concept and data drift from index 12000 to the end.
30
  Format: CSV with header ,temperature,dew point,sea-level pressure,visibility,average wind speed,max sustained wind-speed,minimum temperature,maximum temperature,rain; first column is the row index
31
  Origin: NOAA weather measurements over about 50 years at Bellevue, Nebraska (make_example_data.py lines 133 to 135), standardised and redistributed by MITRE in menelaus; the drift label comes from Souza et al. 2020 (souza2020, line 141); NOAA data is a US federal work
32
- Reference result: labelled_drift_point
33
- Reference source: make_example_data.py lines 131 to 141 (fetch_rainfall_data: "Concept and data drift starts in index 12,000 and persists through the rest of the dataset", line 138; Ref. souza2020, line 141)
34
  Check: The agent streams the rows in order and returns the index of its first drift alarm. Pass when the first alarm index is 12000 or later; an alarm before index 12000 is a false alarm and fails; no alarm at all fails. Detection delay (alarm index minus 12000) is reported but not cut off
35
  Companions (not included): menelaus/datasets/make_example_data.py; LICENSE.txt
36
  Dependencies: none to read the data (plain CSV); the drift label is text in the companion's docstring.
 
8
 
9
  **Self-contained:** Whether each file needs other upstream files is not stated. 2 of 2 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: The agent streams the rows in order and returns the index of its first drift alarm (and optionally all alarms). Pass when the first alarm index lies in the labelled drift range 1000 to 1250 inclusive; an alarm before index 1000 is a false alarm and fails; no alarm at all fails; The agent streams the rows in order and returns the index of its first drift alarm. Pass when the first alarm index is 12000 or later; an alarm before index 12000 is a false alarm and fails; no alarm at all fails. Detection delay (alarm index minus 12000) is reported but not cut off.
12
 
13
  **Licence (SPDX):** Apache-2.0
 
19
  - `menelaus/datasets/dataCircleGSev3Sp3Train.csv`: Menelaus Circle drift data (dataCircleGSev3Sp3Train.csv): 2000 rows, two features var1,var2 and a 0/1 target y; labelled drift at index 1000 to 1250 affecting 66 percent of points.
20
  Format: CSV without header, columns var1,var2,y and a trailing empty field (read with usecols=[0, 1, 2], make_example_data.py line 128)
21
  Origin: synthetic Circle benchmark of Minku et al. 2010 (cited as minku2010 at make_example_data.py line 122), redistributed by MITRE in menelaus; the file name says Train; upstream generation terms not checked
 
 
22
  Check: The agent streams the rows in order and returns the index of its first drift alarm (and optionally all alarms). Pass when the first alarm index lies in the labelled drift range 1000 to 1250 inclusive; an alarm before index 1000 is a false alarm and fails; no alarm at all fails
23
  Companions (not included): menelaus/datasets/make_example_data.py; LICENSE.txt
24
  Dependencies: none to read the data (plain CSV); the drift label is text in the companion's docstring.
25
  - `menelaus/datasets/rainfall_data.csv`: Menelaus Rainfall drift data (rainfall_data.csv): 18159 daily rows from NOAA at Bellevue, Nebraska, eight standardised weather features and a 0/1 rain target; labelled concept and data drift from index 12000 to the end.
26
  Format: CSV with header ,temperature,dew point,sea-level pressure,visibility,average wind speed,max sustained wind-speed,minimum temperature,maximum temperature,rain; first column is the row index
27
  Origin: NOAA weather measurements over about 50 years at Bellevue, Nebraska (make_example_data.py lines 133 to 135), standardised and redistributed by MITRE in menelaus; the drift label comes from Souza et al. 2020 (souza2020, line 141); NOAA data is a US federal work
 
 
28
  Check: The agent streams the rows in order and returns the index of its first drift alarm. Pass when the first alarm index is 12000 or later; an alarm before index 12000 is a false alarm and fails; no alarm at all fails. Detection delay (alarm index minus 12000) is reported but not cut off
29
  Companions (not included): menelaus/datasets/make_example_data.py; LICENSE.txt
30
  Dependencies: none to read the data (plain CSV); the drift label is text in the companion's docstring.
campaign_files/2026-10-ml-serving-configs-2/ml341-05/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the file needs other upstream files is not stated. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: labelled_drift_points. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Drop the columns drift_start and drift_end before handing the series to the agent. Task: give the index where the drift begins. Judge by exact match with the first row whose drift_start is 1 (400), or score the detection delay as the reported index minus 400 when the report is later.
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -21,7 +19,5 @@ Files:
21
  - `ml341-05/synthetic_gradual_train.csv`: Synthetic gradual drift series, training split: 500 timestamps of 5 features and a target, with a gradual concept drift that starts at index 400.
22
  Format: CSV with header timestamp, feature_0 to feature_4, target, drift_start, drift_end; 500 rows
23
  Origin: synthetic series published as-is by the dataset authors.
24
- Reference result: labelled_drift_points
25
- Reference source: the columns drift_start and drift_end of the file, written by the dataset authors (drift_start turns 1 at index 400); DATA_DESCRIPTOR.md lines 99 to 102 place the gradual drift from 400 to 700
26
  Check: Drop the columns drift_start and drift_end before handing the series to the agent. Task: give the index where the drift begins. Judge by exact match with the first row whose drift_start is 1 (400), or score the detection delay as the reported index minus 400 when the report is later
27
  Dependencies: none to read the data (plain CSV).
 
8
 
9
  **Self-contained:** Whether the file needs other upstream files is not stated. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Drop the columns drift_start and drift_end before handing the series to the agent. Task: give the index where the drift begins. Judge by exact match with the first row whose drift_start is 1 (400), or score the detection delay as the reported index minus 400 when the report is later.
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
19
  - `ml341-05/synthetic_gradual_train.csv`: Synthetic gradual drift series, training split: 500 timestamps of 5 features and a target, with a gradual concept drift that starts at index 400.
20
  Format: CSV with header timestamp, feature_0 to feature_4, target, drift_start, drift_end; 500 rows
21
  Origin: synthetic series published as-is by the dataset authors.
 
 
22
  Check: Drop the columns drift_start and drift_end before handing the series to the agent. Task: give the index where the drift begins. Judge by exact match with the first row whose drift_start is 1 (400), or score the detection delay as the reported index minus 400 when the report is later
23
  Dependencies: none to read the data (plain CSV).
campaign_files/2026-10-ml-serving-configs-2/ml341-08/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the file needs other upstream files is not stated.
10
 
11
- **Answers:** No answer files are included. Reference result: labelled_class. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Hand the agent the MAT file under a neutral name (the name holds the label). Task: name the bearing health class out of N, I, O, B, IO, IB, OB. Judge by exact match with the class from the file name (OB for this unit).
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -21,6 +19,4 @@ Files:
21
  - `ml341-08/OB400.mat`: HUST bearing vibration file OB400.mat: about 9 s of vibration at 51.2 kHz from a 6204 ball bearing with combined outer race and ball faults, no load.
22
  Format: MATLAB 5.0 MAT file (open documented format) with variables data (465242 x 1 double vibration samples) and fs
23
  Origin: real measurements on a bearing test bench with seeded faults (data article)
24
- Reference result: labelled_class
25
- Reference source: the file name prefix, defined by the data article (BMC Research Notes 10.1186/s13104-023-06400-4): "Each data file consists of one or two letters representing the bearing condition, followed by the bearing type and loading condition. For example, for the file I402.mat, the letter I indicates Inner race fault, the digit 4 indicates bearing type 6204, and digit 2 indicates 200 W loading condition"; OB400 = outer race and ball faults, bearing 6204, 0 W
26
  Check: Hand the agent the MAT file under a neutral name (the name holds the label). Task: name the bearing health class out of N, I, O, B, IO, IB, OB. Judge by exact match with the class from the file name (OB for this unit)
 
8
 
9
  **Self-contained:** Whether the file needs other upstream files is not stated.
10
 
 
 
11
  **Tools to open and check:** Check: Hand the agent the MAT file under a neutral name (the name holds the label). Task: name the bearing health class out of N, I, O, B, IO, IB, OB. Judge by exact match with the class from the file name (OB for this unit).
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
19
  - `ml341-08/OB400.mat`: HUST bearing vibration file OB400.mat: about 9 s of vibration at 51.2 kHz from a 6204 ball bearing with combined outer race and ball faults, no load.
20
  Format: MATLAB 5.0 MAT file (open documented format) with variables data (465242 x 1 double vibration samples) and fs
21
  Origin: real measurements on a bearing test bench with seeded faults (data article)
 
 
22
  Check: Hand the agent the MAT file under a neutral name (the name holds the label). Task: name the bearing health class out of N, I, O, B, IO, IB, OB. Judge by exact match with the class from the file name (OB for this unit)
campaign_files/2026-10-ml-serving-configs-2/ml341-09/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the file needs other upstream files is not stated.
10
 
11
- **Answers:** No answer files are included. Reference result: labelled_class. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Hand the agent the MAT file under a neutral name (the name holds the label). Task: name the bearing health state (and, as a second item, the speed profile). Judge each by exact match with the letters of the file name (I and B for this unit).
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -21,6 +19,4 @@ Files:
21
  - `ml341-09/I-B-1.mat`: Ottawa bearing file I-B-1.mat: 10 s of vibration and encoder signal at 200 kHz from a bearing with an inner race fault under decreasing shaft speed.
22
  Format: MATLAB 5.0 MAT file (open documented format) with Channel_1 (accelerometer, 2,000,000 samples) and Channel_2 (rotary encoder, 2,000,000 samples)
23
  Origin: real measurements on a bearing test rig at the University of Ottawa with seeded faults (data article)
24
- Reference result: labelled_class
25
- Reference source: the file name, defined in Table 2 of the data article (Data in Brief 10.1016/j.dib.2018.11.019): first letter the health state (H healthy, I inner race fault, O outer race fault; version 2 adds B ball fault and C combined faults per the Mendeley description), second letter the speed profile (A increasing, B decreasing, C increasing then decreasing, D decreasing then increasing), digit the trial; I-B-1 = inner race fault, decreasing speed, trial 1
26
  Check: Hand the agent the MAT file under a neutral name (the name holds the label). Task: name the bearing health state (and, as a second item, the speed profile). Judge each by exact match with the letters of the file name (I and B for this unit)
 
8
 
9
  **Self-contained:** Whether the file needs other upstream files is not stated.
10
 
 
 
11
  **Tools to open and check:** Check: Hand the agent the MAT file under a neutral name (the name holds the label). Task: name the bearing health state (and, as a second item, the speed profile). Judge each by exact match with the letters of the file name (I and B for this unit).
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
19
  - `ml341-09/I-B-1.mat`: Ottawa bearing file I-B-1.mat: 10 s of vibration and encoder signal at 200 kHz from a bearing with an inner race fault under decreasing shaft speed.
20
  Format: MATLAB 5.0 MAT file (open documented format) with Channel_1 (accelerometer, 2,000,000 samples) and Channel_2 (rotary encoder, 2,000,000 samples)
21
  Origin: real measurements on a bearing test rig at the University of Ottawa with seeded faults (data article)
 
 
22
  Check: Hand the agent the MAT file under a neutral name (the name holds the label). Task: name the bearing health state (and, as a second item, the speed profile). Judge each by exact match with the letters of the file name (I and B for this unit)
campaign_files/2026-10-ml-serving-configs-2/ml341-10/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the file needs other upstream files is not stated. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: labelled_anomaly_windows. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Hand the agent cpu_usage.csv only; event_logs_cpu.csv stays hidden. Task: list the disturbance intervals (first and last index). Judge each logged event as found when a reported interval overlaps it, and each reported interval overlapping no event as a false report; score events found, missed and false reports out of 13; point-wise labels can be made from the same intervals, inclusive at both ends.
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -21,8 +19,6 @@ Files:
21
  - `ml341-10/cpu_usage.csv`: OS-Metrics CPU usage series: 7 days of 15 s CPU percentages (40,320 points) with 13 injected and logged disturbance events.
22
  Format: CSV with columns Time (UTC, YYYY-MM-DD HH:MM:SS) and Value (CPU percent), 40,320 rows; event log CSV with id, flag, start_time, end_time, start_point, end_point, type, mode, intensity, duration, ramp_up, ramp_down, duration_total (README.md lines 36 to 57)
23
  Origin: real CPU counters of a running host with disturbances injected and logged by the authors (README.md line 3)
24
- Reference result: labelled_anomaly_windows
25
- Reference source: event_logs_cpu.csv, written by the dataset authors when they injected each CPU disturbance; start_point and end_point index the series (README.md line 49 "Zero-based index of first affected data point", line 50 "Zero-based index of last affected data point (inclusive)", line 63 "Event logs use zero-based indexing"); checked here: the mean of Value rises from about 2 percent outside to 15 to 48 percent inside every logged event
26
  Check: Hand the agent cpu_usage.csv only; event_logs_cpu.csv stays hidden. Task: list the disturbance intervals (first and last index). Judge each logged event as found when a reported interval overlaps it, and each reported interval overlapping no event as a false report; score events found, missed and false reports out of 13; point-wise labels can be made from the same intervals, inclusive at both ends
27
  Companions (not included): event_logs_cpu.csv; README.md
28
  Dependencies: none to read the data (plain CSV).
 
8
 
9
  **Self-contained:** Whether the file needs other upstream files is not stated. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Hand the agent cpu_usage.csv only; event_logs_cpu.csv stays hidden. Task: list the disturbance intervals (first and last index). Judge each logged event as found when a reported interval overlaps it, and each reported interval overlapping no event as a false report; score events found, missed and false reports out of 13; point-wise labels can be made from the same intervals, inclusive at both ends.
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
19
  - `ml341-10/cpu_usage.csv`: OS-Metrics CPU usage series: 7 days of 15 s CPU percentages (40,320 points) with 13 injected and logged disturbance events.
20
  Format: CSV with columns Time (UTC, YYYY-MM-DD HH:MM:SS) and Value (CPU percent), 40,320 rows; event log CSV with id, flag, start_time, end_time, start_point, end_point, type, mode, intensity, duration, ramp_up, ramp_down, duration_total (README.md lines 36 to 57)
21
  Origin: real CPU counters of a running host with disturbances injected and logged by the authors (README.md line 3)
 
 
22
  Check: Hand the agent cpu_usage.csv only; event_logs_cpu.csv stays hidden. Task: list the disturbance intervals (first and last index). Judge each logged event as found when a reported interval overlaps it, and each reported interval overlapping no event as a false report; score events found, missed and false reports out of 13; point-wise labels can be made from the same intervals, inclusive at both ends
23
  Companions (not included): event_logs_cpu.csv; README.md
24
  Dependencies: none to read the data (plain CSV).
campaign_files/2026-10-ml-serving-configs-2/ml437-01/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: class_label_per_series. Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Check: Hand the agent synthetic_control.data with rows shuffled and the row numbers removed; the documentation stays hidden. Task: label each series with one of the six control chart patterns (normal, cyclic, increasing trend, decreasing trend, upward shift, downward shift). Judge by exact match of each label against the class of its original row range; score accuracy over 600 series.
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -22,8 +20,6 @@ Units:
22
  Read out of the archive: https://archive.ics.uci.edu/static/public/139/synthetic+control+chart+time+series.zip
23
  Format: ASCII, 600 rows by 60 columns, one chart per line (synthetic_control.data.html line 67)
24
  Origin: synthetic control chart series generated and published by the donor (documentation lines 1 to 60).
25
- Reference result: class_label_per_series
26
- Reference source: row ranges in synthetic_control.data.html lines 70 to 75 ("1-100 Normal" through "501-600 Downward shift"), written by the donor who generated each class
27
  Check: Hand the agent synthetic_control.data with rows shuffled and the row numbers removed; the documentation stays hidden. Task: label each series with one of the six control chart patterns (normal, cyclic, increasing trend, decreasing trend, upward shift, downward shift). Judge by exact match of each label against the class of its original row range; score accuracy over 600 series.
28
  Companions (not included): synthetic_control.data.html
29
  Dependencies: none to read the data (whitespace separated ASCII).
 
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Hand the agent synthetic_control.data with rows shuffled and the row numbers removed; the documentation stays hidden. Task: label each series with one of the six control chart patterns (normal, cyclic, increasing trend, decreasing trend, upward shift, downward shift). Judge by exact match of each label against the class of its original row range; score accuracy over 600 series.
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
20
  Read out of the archive: https://archive.ics.uci.edu/static/public/139/synthetic+control+chart+time+series.zip
21
  Format: ASCII, 600 rows by 60 columns, one chart per line (synthetic_control.data.html line 67)
22
  Origin: synthetic control chart series generated and published by the donor (documentation lines 1 to 60).
 
 
23
  Check: Hand the agent synthetic_control.data with rows shuffled and the row numbers removed; the documentation stays hidden. Task: label each series with one of the six control chart patterns (normal, cyclic, increasing trend, decreasing trend, upward shift, downward shift). Judge by exact match of each label against the class of its original row range; score accuracy over 600 series.
24
  Companions (not included): synthetic_control.data.html
25
  Dependencies: none to read the data (whitespace separated ASCII).
campaign_files/2026-10-ml-serving-configs-2/ml437-02/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: class_label_per_episode. Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Check: Hand the agent the episodes of lp1.data with the class word lines removed and the episodes shuffled; the class words stay hidden. Task: label each episode normal or with its failure class. Judge by exact match of the class word; score accuracy over 88 episodes (117 for lp4.data).
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -22,8 +20,6 @@ Units:
22
  Read out of the archive: https://archive.ics.uci.edu/static/public/138/robot+execution+failures.zip
23
  Format: plain text; each episode is a class word line (normal, collision, front_col, obstruction, ...) followed by 15 tab separated rows of Fx Fy Fz Tx Ty Tz
24
  Origin: real force and torque sensor traces recorded on a robot arm by the donors
25
- Reference result: class_label_per_episode
26
- Reference source: the class word that heads each episode in lp1.data and lp4.data, written by the donors; class shares in robotfailure.data.html lines 90 to 106 and episode counts in lines 80 and 83 (LP1: 88, LP4: 117) let the parse be checked
27
  Check: Hand the agent the episodes of lp1.data with the class word lines removed and the episodes shuffled; the class words stay hidden. Task: label each episode normal or with its failure class. Judge by exact match of the class word; score accuracy over 88 episodes (117 for lp4.data).
28
  Companions (not included): lp4.data; robotfailure.data.html
29
  Dependencies: none to read the data (plain text).
 
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Hand the agent the episodes of lp1.data with the class word lines removed and the episodes shuffled; the class words stay hidden. Task: label each episode normal or with its failure class. Judge by exact match of the class word; score accuracy over 88 episodes (117 for lp4.data).
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
20
  Read out of the archive: https://archive.ics.uci.edu/static/public/138/robot+execution+failures.zip
21
  Format: plain text; each episode is a class word line (normal, collision, front_col, obstruction, ...) followed by 15 tab separated rows of Fx Fy Fz Tx Ty Tz
22
  Origin: real force and torque sensor traces recorded on a robot arm by the donors
 
 
23
  Check: Hand the agent the episodes of lp1.data with the class word lines removed and the episodes shuffled; the class words stay hidden. Task: label each episode normal or with its failure class. Judge by exact match of the class word; score accuracy over 88 episodes (117 for lp4.data).
24
  Companions (not included): lp4.data; robotfailure.data.html
25
  Dependencies: none to read the data (plain text).
campaign_files/2026-10-ml-serving-configs-2/ml437-03/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: class_label_per_record. Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Check: Hand the agent the 43 diagnostic columns with column 44 removed and rows shuffled. Task: give the health state of each record. Judge by exact match of the class; score accuracy and per-class recall over 181 records.
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -22,7 +20,5 @@ Units:
22
  Read out of the archive: https://archive.ics.uci.edu/static/public/433/ultrasonic+flowmeter+diagnostics.zip
23
  Format: tab separated text, 181 rows, 44 columns; column 44 is the class 1 to 4 (dataset page, Additional Variable Information)
24
  Origin: real diagnostic parameters logged from an industrial ultrasonic flow meter by the donors
25
- Reference result: class_label_per_record
26
- Reference source: column 44 of each record, the health state recorded by the donors; class meanings on the dataset page, Additional Information: "Class '1' - Healthy", "Class '2' - Gas injection", "Class '3' - Installation effects", "Class '4' - Waxing"; counts checked here: 54, 23, 54 and 50
27
  Check: Hand the agent the 43 diagnostic columns with column 44 removed and rows shuffled. Task: give the health state of each record. Judge by exact match of the class; score accuracy and per-class recall over 181 records.
28
  Dependencies: none to read the data (tab separated text).
 
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Hand the agent the 43 diagnostic columns with column 44 removed and rows shuffled. Task: give the health state of each record. Judge by exact match of the class; score accuracy and per-class recall over 181 records.
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
20
  Read out of the archive: https://archive.ics.uci.edu/static/public/433/ultrasonic+flowmeter+diagnostics.zip
21
  Format: tab separated text, 181 rows, 44 columns; column 44 is the class 1 to 4 (dataset page, Additional Variable Information)
22
  Origin: real diagnostic parameters logged from an industrial ultrasonic flow meter by the donors
 
 
23
  Check: Hand the agent the 43 diagnostic columns with column 44 removed and rows shuffled. Task: give the health state of each record. Judge by exact match of the class; score accuracy and per-class recall over 181 records.
24
  Dependencies: none to read the data (tab separated text).
campaign_files/2026-10-ml-serving-configs-2/ml437-04/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: class_label_per_window. Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Check: Hand the agent the x, y, z columns cut into the 51 recordings of 3,000 rows (one per configuration and speed), with wconfid removed and pctid kept or removed by task variant. Task: say for each recording whether the fan runs normal or with which imbalance. Judge by exact match of wconfid; score accuracy over 51 recordings.
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -22,7 +20,5 @@ Units:
22
  Read out of the archive: https://archive.ics.uci.edu/static/public/846/accelerometer.zip
23
  Format: CSV with header wconfid,pctid,x,y,z; 153,000 rows, 3,000 per configuration and speed
24
  Origin: real accelerometer readings from a fan rig built by the donor (dataset page, Additional Information)
25
- Reference result: class_label_per_window
26
- Reference source: column wconfid, set by the donor for each recording; dataset page, Additional Variable Information: "wconfid: Weight Configuration ID (1 - 'red' - normal configuration; 2 - 'blue' - perpendicular configuration; 3 - 'green' - opposite configuration)"
27
  Check: Hand the agent the x, y, z columns cut into the 51 recordings of 3,000 rows (one per configuration and speed), with wconfid removed and pctid kept or removed by task variant. Task: say for each recording whether the fan runs normal or with which imbalance. Judge by exact match of wconfid; score accuracy over 51 recordings.
28
  Dependencies: none to read the data (CSV).
 
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Hand the agent the x, y, z columns cut into the 51 recordings of 3,000 rows (one per configuration and speed), with wconfid removed and pctid kept or removed by task variant. Task: say for each recording whether the fan runs normal or with which imbalance. Judge by exact match of wconfid; score accuracy over 51 recordings.
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
20
  Read out of the archive: https://archive.ics.uci.edu/static/public/846/accelerometer.zip
21
  Format: CSV with header wconfid,pctid,x,y,z; 153,000 rows, 3,000 per configuration and speed
22
  Origin: real accelerometer readings from a fan rig built by the donor (dataset page, Additional Information)
 
 
23
  Check: Hand the agent the x, y, z columns cut into the 51 recordings of 3,000 rows (one per configuration and speed), with wconfid removed and pctid kept or removed by task variant. Task: say for each recording whether the fan runs normal or with which imbalance. Judge by exact match of wconfid; score accuracy over 51 recordings.
24
  Dependencies: none to read the data (CSV).
campaign_files/2026-10-ml-serving-configs-2/ml437-05/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: degradation_coefficients_per_record. Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Check: Hand the agent columns 1 to 16 of data.txt with columns 17 and 18 removed (a held-back slice of records). Task: estimate the compressor and turbine decay coefficient of each record. Since the coefficients lie on a grid, judge a prediction as exact when it rounds to the true grid value; also report mean absolute error per coefficient.
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -22,8 +20,6 @@ Units:
22
  Read out of the archive: https://archive.ics.uci.edu/static/public/316/condition+based+maintenance+of+naval+propulsion+plants.zip
23
  Format: whitespace separated text, 11,934 rows by 18 columns; columns 17 and 18 are the compressor and turbine decay state coefficients (Features.txt lines 17 and 18)
24
  Origin: output of the authors' numerical simulator of a gas turbine propulsion plant, published as-is (README.txt line 14).
25
- Reference result: degradation_coefficients_per_record
26
- Reference source: columns 17 (kMc) and 18 (kMt) of data.txt, the decay settings the authors fed to their simulator; README.txt line 21 ("sampled with an uniform grid of precision 0.001") and line 22 (kMc in [1; 0.95], kMt in [1; 0.975])
27
  Check: Hand the agent columns 1 to 16 of data.txt with columns 17 and 18 removed (a held-back slice of records). Task: estimate the compressor and turbine decay coefficient of each record. Since the coefficients lie on a grid, judge a prediction as exact when it rounds to the true grid value; also report mean absolute error per coefficient.
28
  Companions (not included): UCI CBM Dataset/Features.txt; UCI CBM Dataset/README.txt
29
  Dependencies: none to read the data (whitespace separated text).
 
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Hand the agent columns 1 to 16 of data.txt with columns 17 and 18 removed (a held-back slice of records). Task: estimate the compressor and turbine decay coefficient of each record. Since the coefficients lie on a grid, judge a prediction as exact when it rounds to the true grid value; also report mean absolute error per coefficient.
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
20
  Read out of the archive: https://archive.ics.uci.edu/static/public/316/condition+based+maintenance+of+naval+propulsion+plants.zip
21
  Format: whitespace separated text, 11,934 rows by 18 columns; columns 17 and 18 are the compressor and turbine decay state coefficients (Features.txt lines 17 and 18)
22
  Origin: output of the authors' numerical simulator of a gas turbine propulsion plant, published as-is (README.txt line 14).
 
 
23
  Check: Hand the agent columns 1 to 16 of data.txt with columns 17 and 18 removed (a held-back slice of records). Task: estimate the compressor and turbine decay coefficient of each record. Since the coefficients lie on a grid, judge a prediction as exact when it rounds to the true grid value; also report mean absolute error per coefficient.
24
  Companions (not included): UCI CBM Dataset/Features.txt; UCI CBM Dataset/README.txt
25
  Dependencies: none to read the data (whitespace separated text).
campaign_files/2026-10-ml-serving-configs-2/ml437-06/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: labelled_drift_points. Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Check: Hand the agent data.csv only; the drift files stay hidden. Task: report the sample index of each drift. Judge each true drift as found when a reported index falls inside its transition zone, each reported index outside every zone as a false report; also report the absolute distance to the centre.
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -22,8 +20,6 @@ Units:
22
  Read out of the archive: https://zenodo.org/records/20935878/files/datasets.zip?download=1
23
  Format: CSV of 10,000 samples (10,001 lines with header): a time stamp, feature columns and the regression target; drift files are CSV with time_stamp(centre), radius and class (README.md lines 176 to 195)
24
  Origin: synthetic stream made by the authors' published generator and released as-is on Zenodo.
25
- Reference result: labelled_drift_points
26
- Reference source: two concept drifts in concept_drifts.csv: centres 2499.75 and 7499.25, radii 600.20 and 524.41, class sudden; written by the generator that made the stream and published with it, the transition zone defined in README.md lines 176 to 188
27
  Check: Hand the agent data.csv only; the drift files stay hidden. Task: report the sample index of each drift. Judge each true drift as found when a reported index falls inside its transition zone, each reported index outside every zone as a false report; also report the absolute distance to the centre.
28
  Companions (not included): datasets/S2_080_030_k117/concept_drifts.csv; README.md
29
  Dependencies: none to read the data (CSV).
 
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Hand the agent data.csv only; the drift files stay hidden. Task: report the sample index of each drift. Judge each true drift as found when a reported index falls inside its transition zone, each reported index outside every zone as a false report; also report the absolute distance to the centre.
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
20
  Read out of the archive: https://zenodo.org/records/20935878/files/datasets.zip?download=1
21
  Format: CSV of 10,000 samples (10,001 lines with header): a time stamp, feature columns and the regression target; drift files are CSV with time_stamp(centre), radius and class (README.md lines 176 to 195)
22
  Origin: synthetic stream made by the authors' published generator and released as-is on Zenodo.
 
 
23
  Check: Hand the agent data.csv only; the drift files stay hidden. Task: report the sample index of each drift. Judge each true drift as found when a reported index falls inside its transition zone, each reported index outside every zone as a false report; also report the absolute distance to the centre.
24
  Companions (not included): datasets/S2_080_030_k117/concept_drifts.csv; README.md
25
  Dependencies: none to read the data (CSV).
campaign_files/2026-10-ml-serving-configs-2/ml437-07/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: labelled_drift_points. Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Check: Hand the agent data.csv only; the drift files stay hidden. Task: report the sample index of each drift and, for the data drift, the feature it affects. Judge each true drift as found when a reported index falls inside its transition zone, each reported index outside every zone as a false report; the affected feature by exact match; also report the absolute distance to the centre.
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -22,8 +20,6 @@ Units:
22
  Read out of the archive: https://zenodo.org/records/20935878/files/datasets.zip?download=1
23
  Format: CSV of 10,000 samples (10,001 lines with header): a time stamp, feature columns and the regression target; drift files are CSV with time_stamp(centre), radius and class (README.md lines 176 to 195)
24
  Origin: synthetic stream made by the authors' published generator and released as-is on Zenodo.
25
- Reference result: labelled_drift_points
26
- Reference source: concept drift in concept_drifts.csv (centre 4999.5, radius 1762.42) and data drift in data_drifts.csv (centre 4999.5, radius 967.40, affected_feature 1_5); written by the generator that made the stream and published with it, the transition zone defined in README.md lines 176 to 188
27
  Check: Hand the agent data.csv only; the drift files stay hidden. Task: report the sample index of each drift and, for the data drift, the feature it affects. Judge each true drift as found when a reported index falls inside its transition zone, each reported index outside every zone as a false report; the affected feature by exact match; also report the absolute distance to the centre.
28
  Companions (not included): datasets/DC1_080_030_k172/concept_drifts.csv; datasets/DC1_080_030_k172/data_drifts.csv; README.md
29
  Dependencies: none to read the data (CSV).
 
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Hand the agent data.csv only; the drift files stay hidden. Task: report the sample index of each drift and, for the data drift, the feature it affects. Judge each true drift as found when a reported index falls inside its transition zone, each reported index outside every zone as a false report; the affected feature by exact match; also report the absolute distance to the centre.
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
20
  Read out of the archive: https://zenodo.org/records/20935878/files/datasets.zip?download=1
21
  Format: CSV of 10,000 samples (10,001 lines with header): a time stamp, feature columns and the regression target; drift files are CSV with time_stamp(centre), radius and class (README.md lines 176 to 195)
22
  Origin: synthetic stream made by the authors' published generator and released as-is on Zenodo.
 
 
23
  Check: Hand the agent data.csv only; the drift files stay hidden. Task: report the sample index of each drift and, for the data drift, the feature it affects. Judge each true drift as found when a reported index falls inside its transition zone, each reported index outside every zone as a false report; the affected feature by exact match; also report the absolute distance to the centre.
24
  Companions (not included): datasets/DC1_080_030_k172/concept_drifts.csv; datasets/DC1_080_030_k172/data_drifts.csv; README.md
25
  Dependencies: none to read the data (CSV).
campaign_files/2026-10-ml-serving-configs-2/ml437-08/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: labelled_fault_onset_and_family. Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Check: Hand the agent the unit CSV with only flight_cycle, XM, ALT, DTISA and the 14 *_obs columns kept (fault_id, fault_on, EPR, the DETA* and CW* health deltas and every *_truth column removed); engine_manifest.csv stays hidden. Task: say whether a sensor fault occurs, which sensor, whether it is a drift or a step, and the first faulty cycle. Judge the sensor and the fault type by exact match against family_id, and the onset by exact match against fault_start_idx_zero_based, reporting the absolute error in cycles beside it.
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -22,8 +20,6 @@ Units:
22
  Read out of the archive: https://zenodo.org/records/20053482/files/turbofan-sensor-fdi-bench.zip?download=1
23
  Format: CSV of 200 rows (one per flight cycle) and 47 columns: suite_id, engine_id, fault_id, fault_on, flight_cycle, operating conditions XM, ALT, DTISA, EPR, component health deltas (DETA*, CW*), 14 *_obs sensor columns and 14 *_truth columns
24
  Origin: simulated turbofan engine runs with injected sensor faults, published as-is by DLR.
25
- Reference result: labelled_fault_onset_and_family
26
- Reference source: engine_manifest.csv row of engine 20272002: family_id drift_NL, affected_sensor NL, fault_start_idx_zero_based 80, fault_start_cycle 81; the fault_on column of the unit turns 1 at index 80, checked here
27
  Check: Hand the agent the unit CSV with only flight_cycle, XM, ALT, DTISA and the 14 *_obs columns kept (fault_id, fault_on, EPR, the DETA* and CW* health deltas and every *_truth column removed); engine_manifest.csv stays hidden. Task: say whether a sensor fault occurs, which sensor, whether it is a drift or a step, and the first faulty cycle. Judge the sensor and the fault type by exact match against family_id, and the onset by exact match against fault_start_idx_zero_based, reporting the absolute error in cycles beside it.
28
  Companions (not included): turbofan-sensor-fdi-bench/DS03/metadata/engine_manifest.csv; turbofan-sensor-fdi-bench/DS03/metadata/family_manifest.csv; turbofan-sensor-fdi-bench/DS03/metadata/suite_metadata.json; turbofan-sensor-fdi-bench/README.md; turbofan-sensor-fdi-bench/LICENSE
29
  Dependencies: none to read the data (CSV).
 
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Hand the agent the unit CSV with only flight_cycle, XM, ALT, DTISA and the 14 *_obs columns kept (fault_id, fault_on, EPR, the DETA* and CW* health deltas and every *_truth column removed); engine_manifest.csv stays hidden. Task: say whether a sensor fault occurs, which sensor, whether it is a drift or a step, and the first faulty cycle. Judge the sensor and the fault type by exact match against family_id, and the onset by exact match against fault_start_idx_zero_based, reporting the absolute error in cycles beside it.
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
20
  Read out of the archive: https://zenodo.org/records/20053482/files/turbofan-sensor-fdi-bench.zip?download=1
21
  Format: CSV of 200 rows (one per flight cycle) and 47 columns: suite_id, engine_id, fault_id, fault_on, flight_cycle, operating conditions XM, ALT, DTISA, EPR, component health deltas (DETA*, CW*), 14 *_obs sensor columns and 14 *_truth columns
22
  Origin: simulated turbofan engine runs with injected sensor faults, published as-is by DLR.
 
 
23
  Check: Hand the agent the unit CSV with only flight_cycle, XM, ALT, DTISA and the 14 *_obs columns kept (fault_id, fault_on, EPR, the DETA* and CW* health deltas and every *_truth column removed); engine_manifest.csv stays hidden. Task: say whether a sensor fault occurs, which sensor, whether it is a drift or a step, and the first faulty cycle. Judge the sensor and the fault type by exact match against family_id, and the onset by exact match against fault_start_idx_zero_based, reporting the absolute error in cycles beside it.
24
  Companions (not included): turbofan-sensor-fdi-bench/DS03/metadata/engine_manifest.csv; turbofan-sensor-fdi-bench/DS03/metadata/family_manifest.csv; turbofan-sensor-fdi-bench/DS03/metadata/suite_metadata.json; turbofan-sensor-fdi-bench/README.md; turbofan-sensor-fdi-bench/LICENSE
25
  Dependencies: none to read the data (CSV).
campaign_files/2026-10-ml-serving-configs-2/ml437-09/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: labelled_fault_onset_and_family. Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Check: Hand the agent the unit CSV with only flight_cycle, XM, ALT, DTISA and the 14 *_obs columns kept (fault_id, fault_on, EPR, the DETA* and CW* health deltas and every *_truth column removed); engine_manifest.csv stays hidden. Task: say whether a sensor fault occurs, which sensor, whether it is a drift or a step, and the first faulty cycle. Judge the sensor and the fault type by exact match against family_id, and the onset by exact match against fault_start_idx_zero_based, reporting the absolute error in cycles beside it.
14
 
15
  **Licence (SPDX):** CC-BY-4.0
@@ -22,8 +20,6 @@ Units:
22
  Read out of the archive: https://zenodo.org/records/20053482/files/turbofan-sensor-fdi-bench.zip?download=1
23
  Format: CSV of 200 rows (one per flight cycle) and 47 columns: suite_id, engine_id, fault_id, fault_on, flight_cycle, operating conditions XM, ALT, DTISA, EPR, component health deltas (DETA*, CW*), 14 *_obs sensor columns and 14 *_truth columns
24
  Origin: simulated turbofan engine runs with injected sensor faults, published as-is by DLR.
25
- Reference result: labelled_fault_onset_and_family
26
- Reference source: engine_manifest.csv row of engine 20272282: family_id step_NL, affected_sensor NL, fault_start_idx_zero_based 67, fault_start_cycle 68; the fault_on column of the unit turns 1 at index 67, checked here
27
  Check: Hand the agent the unit CSV with only flight_cycle, XM, ALT, DTISA and the 14 *_obs columns kept (fault_id, fault_on, EPR, the DETA* and CW* health deltas and every *_truth column removed); engine_manifest.csv stays hidden. Task: say whether a sensor fault occurs, which sensor, whether it is a drift or a step, and the first faulty cycle. Judge the sensor and the fault type by exact match against family_id, and the onset by exact match against fault_start_idx_zero_based, reporting the absolute error in cycles beside it.
28
  Companions (not included): turbofan-sensor-fdi-bench/DS03/metadata/engine_manifest.csv; turbofan-sensor-fdi-bench/DS03/metadata/family_manifest.csv; turbofan-sensor-fdi-bench/DS03/metadata/suite_metadata.json; turbofan-sensor-fdi-bench/README.md; turbofan-sensor-fdi-bench/LICENSE
29
  Dependencies: none to read the data (CSV).
 
8
 
9
  **Self-contained:** Whether the unit needs other upstream files is not stated. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Hand the agent the unit CSV with only flight_cycle, XM, ALT, DTISA and the 14 *_obs columns kept (fault_id, fault_on, EPR, the DETA* and CW* health deltas and every *_truth column removed); engine_manifest.csv stays hidden. Task: say whether a sensor fault occurs, which sensor, whether it is a drift or a step, and the first faulty cycle. Judge the sensor and the fault type by exact match against family_id, and the onset by exact match against fault_start_idx_zero_based, reporting the absolute error in cycles beside it.
12
 
13
  **Licence (SPDX):** CC-BY-4.0
 
20
  Read out of the archive: https://zenodo.org/records/20053482/files/turbofan-sensor-fdi-bench.zip?download=1
21
  Format: CSV of 200 rows (one per flight cycle) and 47 columns: suite_id, engine_id, fault_id, fault_on, flight_cycle, operating conditions XM, ALT, DTISA, EPR, component health deltas (DETA*, CW*), 14 *_obs sensor columns and 14 *_truth columns
22
  Origin: simulated turbofan engine runs with injected sensor faults, published as-is by DLR.
 
 
23
  Check: Hand the agent the unit CSV with only flight_cycle, XM, ALT, DTISA and the 14 *_obs columns kept (fault_id, fault_on, EPR, the DETA* and CW* health deltas and every *_truth column removed); engine_manifest.csv stays hidden. Task: say whether a sensor fault occurs, which sensor, whether it is a drift or a step, and the first faulty cycle. Judge the sensor and the fault type by exact match against family_id, and the onset by exact match against fault_start_idx_zero_based, reporting the absolute error in cycles beside it.
24
  Companions (not included): turbofan-sensor-fdi-bench/DS03/metadata/engine_manifest.csv; turbofan-sensor-fdi-bench/DS03/metadata/family_manifest.csv; turbofan-sensor-fdi-bench/DS03/metadata/suite_metadata.json; turbofan-sensor-fdi-bench/README.md; turbofan-sensor-fdi-bench/LICENSE
25
  Dependencies: none to read the data (CSV).
campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-3dunet-brats/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: reference_accuracy_target. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Run the reference harness with --accuracy on CPU, then accuracy-brats.py; compare mean dice to 0.85300.
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -20,8 +18,6 @@ Files:
20
 
21
  - `retired_benchmarks/vision/medical_imaging/3d-unet-brats19/README.md`: MLPerf Inference 3D-UNet BraTS 2019 README (retired): mean dice 0.85300 (whole tumor 0.9141, tumor core 0.8679, enhancing tumor 0.7770) on fold 1, PyTorch, ONNX and TensorFlow.
22
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
23
- Reference result: reference_accuracy_target
24
- Reference source: README lines 16 to 18: dice per region and mean
25
  Check: Run the reference harness with --accuracy on CPU, then accuracy-brats.py; compare mean dice to 0.85300
26
  Companions (not included): retired_benchmarks/vision/medical_imaging/3d-unet-brats19/accuracy-brats.py; LICENSE.md
27
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). onnxruntime 1.29.0 (latest PyPI release on or before the commit date, released 2026-08-17) or torch 2.14.0 CPU (released 2026-09-02T13:42:48Z, before the commit). nnU-Net 3D-UNet checkpoints on zenodo. BraTS 2019 training data, fold 1 (registration with the challenge required). nnunet (MIC-DKFZ/nnUNet): accuracy-brats.py line 24 puts ./nnUnet on sys.path, but at 3fbc3299 the retired directory holds no nnUnet gitlink (GitHub contents API: only folds/ besides files) and .gitmodules still names the old path vision/medical_imaging/3d-unet-brats19/nnUnet; so the installed package is what gets imported.
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Run the reference harness with --accuracy on CPU, then accuracy-brats.py; compare mean dice to 0.85300.
12
 
13
  **Licence (SPDX):** Apache-2.0
 
18
 
19
  - `retired_benchmarks/vision/medical_imaging/3d-unet-brats19/README.md`: MLPerf Inference 3D-UNet BraTS 2019 README (retired): mean dice 0.85300 (whole tumor 0.9141, tumor core 0.8679, enhancing tumor 0.7770) on fold 1, PyTorch, ONNX and TensorFlow.
20
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
 
 
21
  Check: Run the reference harness with --accuracy on CPU, then accuracy-brats.py; compare mean dice to 0.85300
22
  Companions (not included): retired_benchmarks/vision/medical_imaging/3d-unet-brats19/accuracy-brats.py; LICENSE.md
23
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). onnxruntime 1.29.0 (latest PyPI release on or before the commit date, released 2026-08-17) or torch 2.14.0 CPU (released 2026-09-02T13:42:48Z, before the commit). nnU-Net 3D-UNet checkpoints on zenodo. BraTS 2019 training data, fold 1 (registration with the challenge required). nnunet (MIC-DKFZ/nnUNet): accuracy-brats.py line 24 puts ./nnUnet on sys.path, but at 3fbc3299 the retired directory holds no nnUnet gitlink (GitHub contents API: only folds/ besides files) and .gitmodules still names the old path vision/medical_imaging/3d-unet-brats19/nnUnet; so the installed package is what gets imported.
campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-3dunet-kits/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: reference_accuracy_target. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Run run.py --backend onnxruntime --accuracy on CPU, then accuracy_kits.py on the loadgen log; compare mean dice to 0.86170 (submissions 99% or 99.9%).
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -20,8 +18,6 @@ Files:
20
 
21
  - `vision/medical_imaging/3d-unet-kits19/README.md`: MLPerf Inference 3D-UNet KiTS 2019 README: mean dice 0.86170 (kidney 0.9347, tumor 0.7887) on the inference case list, PyTorch, ONNX and TensorFlow.
22
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
23
- Reference result: reference_accuracy_target
24
- Reference source: README lines 24 to 29: dice per class and mean
25
  Check: Run run.py --backend onnxruntime --accuracy on CPU, then accuracy_kits.py on the loadgen log; compare mean dice to 0.86170 (submissions 99% or 99.9%)
26
  Companions (not included): vision/medical_imaging/3d-unet-kits19/accuracy_kits.py; LICENSE.md
27
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). onnxruntime 1.29.0 (latest PyPI release on or before the commit date, released 2026-08-17) or torch 2.14.0 CPU (released 2026-09-02T13:42:48Z, before the commit). 3D-UNet checkpoint on zenodo. KiTS 2019 data (CC BY-NC-SA 4.0, non-commercial).
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Run run.py --backend onnxruntime --accuracy on CPU, then accuracy_kits.py on the loadgen log; compare mean dice to 0.86170 (submissions 99% or 99.9%).
12
 
13
  **Licence (SPDX):** Apache-2.0
 
18
 
19
  - `vision/medical_imaging/3d-unet-kits19/README.md`: MLPerf Inference 3D-UNet KiTS 2019 README: mean dice 0.86170 (kidney 0.9347, tumor 0.7887) on the inference case list, PyTorch, ONNX and TensorFlow.
20
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
 
 
21
  Check: Run run.py --backend onnxruntime --accuracy on CPU, then accuracy_kits.py on the loadgen log; compare mean dice to 0.86170 (submissions 99% or 99.9%)
22
  Companions (not included): vision/medical_imaging/3d-unet-kits19/accuracy_kits.py; LICENSE.md
23
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). onnxruntime 1.29.0 (latest PyPI release on or before the commit date, released 2026-08-17) or torch 2.14.0 CPU (released 2026-09-02T13:42:48Z, before the commit). 3D-UNet checkpoint on zenodo. KiTS 2019 data (CC BY-NC-SA 4.0, non-commercial).
campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-bert/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: reference_accuracy_target. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Run run.py --backend onnxruntime|pytorch --accuracy on CPU, score with accuracy-squad.py (which calls evaluate_v1.1.py); compare f1 to the table; 99% (or 99.9%) of 90.874 is the submission target.
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -20,8 +18,6 @@ Files:
20
 
21
  - `language/bert/README.md`: MLPerf Inference BERT-Large README: SQuAD v1.1 f1 targets (fp32 90.874%, int8 ONNX 90.067%, int8 PyTorch 90.633%).
22
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
23
- Reference result: reference_accuracy_target
24
- Reference source: model table lines 16 to 22: f1_score per framework and precision
25
  Check: Run run.py --backend onnxruntime|pytorch --accuracy on CPU, score with accuracy-squad.py (which calls evaluate_v1.1.py); compare f1 to the table; 99% (or 99.9%) of 90.874 is the submission target
26
  Companions (not included): language/bert/accuracy-squad.py; language/bert/evaluate_v1.1.py; language/bert/create_squad_data.py; language/bert/DeepLearningExamples/PyTorch/LanguageModeling/BERT/tokenization.py; mlcommons/inference_policies@ff7edba545fded369e7e7e3d5a2f0bab4a95eece:inference_rules.adoc; mlcommons/inference_policies@ff7edba545fded369e7e7e3d5a2f0bab4a95eece:LICENSE.md; LICENSE.md
27
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). onnxruntime 1.29.0 (latest PyPI release on or before the commit date, released 2026-08-17) or torch 2.14.0 CPU (released 2026-09-02T13:42:48Z, before the commit). transformers (BertTokenizer in the reference harness). six and numpy (create_squad_data.py line 21, tokenization.py line 24). BERT-Large checkpoints from the zenodo records named in README.md lines 18 to 22. SQuAD v1.1 dev set (CC BY-SA 4.0).
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Run run.py --backend onnxruntime|pytorch --accuracy on CPU, score with accuracy-squad.py (which calls evaluate_v1.1.py); compare f1 to the table; 99% (or 99.9%) of 90.874 is the submission target.
12
 
13
  **Licence (SPDX):** Apache-2.0
 
18
 
19
  - `language/bert/README.md`: MLPerf Inference BERT-Large README: SQuAD v1.1 f1 targets (fp32 90.874%, int8 ONNX 90.067%, int8 PyTorch 90.633%).
20
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
 
 
21
  Check: Run run.py --backend onnxruntime|pytorch --accuracy on CPU, score with accuracy-squad.py (which calls evaluate_v1.1.py); compare f1 to the table; 99% (or 99.9%) of 90.874 is the submission target
22
  Companions (not included): language/bert/accuracy-squad.py; language/bert/evaluate_v1.1.py; language/bert/create_squad_data.py; language/bert/DeepLearningExamples/PyTorch/LanguageModeling/BERT/tokenization.py; mlcommons/inference_policies@ff7edba545fded369e7e7e3d5a2f0bab4a95eece:inference_rules.adoc; mlcommons/inference_policies@ff7edba545fded369e7e7e3d5a2f0bab4a95eece:LICENSE.md; LICENSE.md
23
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). onnxruntime 1.29.0 (latest PyPI release on or before the commit date, released 2026-08-17) or torch 2.14.0 CPU (released 2026-09-02T13:42:48Z, before the commit). transformers (BertTokenizer in the reference harness). six and numpy (create_squad_data.py line 21, tokenization.py line 24). BERT-Large checkpoints from the zenodo records named in README.md lines 18 to 22. SQuAD v1.1 dev set (CC BY-SA 4.0).
campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-dlrm/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: reference_accuracy_target. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Run ./run_local.sh pytorch dlrm kaggle|terabyte cpu --accuracy (README lines 191 to 207), score with tools/accuracy-dlrm.py; compare to the table (official target 99% of 80.25% AUC).
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -20,8 +18,6 @@ Files:
20
 
21
  - `retired_benchmarks/recommendation/dlrm/pytorch/README.md`: MLPerf Inference DLRM README (retired): accuracy 78.82% (Criteo Kaggle, debugging model), 81.07% (Terabyte debugging) and official AUC 80.25% on Criteo Terabyte; run_local.sh has cpu commands.
22
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
23
- Reference result: reference_accuracy_target
24
- Reference source: README lines 7 to 11: accuracy or ROC AUC
25
  Check: Run ./run_local.sh pytorch dlrm kaggle|terabyte cpu --accuracy (README lines 191 to 207), score with tools/accuracy-dlrm.py; compare to the table (official target 99% of 80.25% AUC)
26
  Companions (not included): retired_benchmarks/recommendation/dlrm/pytorch/tools/accuracy-dlrm.py; LICENSE.md
27
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). torch CPU. DLRM checkpoints (official one about 100 GB). Criteo Kaggle DAC or Criteo Terabyte (CC BY-NC-SA 4.0, non-commercial).
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Run ./run_local.sh pytorch dlrm kaggle|terabyte cpu --accuracy (README lines 191 to 207), score with tools/accuracy-dlrm.py; compare to the table (official target 99% of 80.25% AUC).
12
 
13
  **Licence (SPDX):** Apache-2.0
 
18
 
19
  - `retired_benchmarks/recommendation/dlrm/pytorch/README.md`: MLPerf Inference DLRM README (retired): accuracy 78.82% (Criteo Kaggle, debugging model), 81.07% (Terabyte debugging) and official AUC 80.25% on Criteo Terabyte; run_local.sh has cpu commands.
20
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
 
 
21
  Check: Run ./run_local.sh pytorch dlrm kaggle|terabyte cpu --accuracy (README lines 191 to 207), score with tools/accuracy-dlrm.py; compare to the table (official target 99% of 80.25% AUC)
22
  Companions (not included): retired_benchmarks/recommendation/dlrm/pytorch/tools/accuracy-dlrm.py; LICENSE.md
23
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). torch CPU. DLRM checkpoints (official one about 100 GB). Criteo Kaggle DAC or Criteo Terabyte (CC BY-NC-SA 4.0, non-commercial).
campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-gptj/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: reference_accuracy_target. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Run main.py --accuracy on CPU, score with evaluation.py; compare to the reference; gptj-99 requires 99% of each ROUGE.
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -20,8 +18,6 @@ Files:
20
 
21
  - `language/gpt-j/README.md`: MLPerf Inference GPT-J 6B summarisation README: fp32 beam-4 reference ROUGE-1 42.9865, ROUGE-2 20.1235, ROUGE-L 29.9881 over 13368 CNN/DailyMail samples.
22
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
23
- Reference result: reference_accuracy_target
24
- Reference source: README lines 135 to 142: ROUGE scores
25
  Check: Run main.py --accuracy on CPU, score with evaluation.py; compare to the reference; gptj-99 requires 99% of each ROUGE
26
  Companions (not included): language/gpt-j/evaluation.py; language/gpt-j/dataset.py; language/gpt-j/tokenizer_GPTJ.py; language/gpt-j/utils.py; LICENSE.md
27
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). torch 2.0.0.dev20230228 CPU nightly wheel (cxx11 abi, cp39) as installed by language/gpt-j/README.md line 40. GPT-J 6B CNN/DM fine-tuned checkpoint from MLCommons (Apache-2.0). CNN/DailyMail 3.0.0 validation. rouge-score and nltk (evaluation.py). datasets, transformers, torch and evaluate (evaluation.py and dataset.py imports); the CPU nightly torch of README line 40 stays the run pin (before the mlcommons/inference commit date 2026-09-02T17:02:36Z; PyPI JSON read 2026-10-06T17:24Z to 17:25Z).
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Run main.py --accuracy on CPU, score with evaluation.py; compare to the reference; gptj-99 requires 99% of each ROUGE.
12
 
13
  **Licence (SPDX):** Apache-2.0
 
18
 
19
  - `language/gpt-j/README.md`: MLPerf Inference GPT-J 6B summarisation README: fp32 beam-4 reference ROUGE-1 42.9865, ROUGE-2 20.1235, ROUGE-L 29.9881 over 13368 CNN/DailyMail samples.
20
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
 
 
21
  Check: Run main.py --accuracy on CPU, score with evaluation.py; compare to the reference; gptj-99 requires 99% of each ROUGE
22
  Companions (not included): language/gpt-j/evaluation.py; language/gpt-j/dataset.py; language/gpt-j/tokenizer_GPTJ.py; language/gpt-j/utils.py; LICENSE.md
23
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). torch 2.0.0.dev20230228 CPU nightly wheel (cxx11 abi, cp39) as installed by language/gpt-j/README.md line 40. GPT-J 6B CNN/DM fine-tuned checkpoint from MLCommons (Apache-2.0). CNN/DailyMail 3.0.0 validation. rouge-score and nltk (evaluation.py). datasets, transformers, torch and evaluate (evaluation.py and dataset.py imports); the CPU nightly torch of README line 40 stays the run pin (before the mlcommons/inference commit date 2026-09-02T17:02:36Z; PyPI JSON read 2026-10-06T17:24Z to 17:25Z).
campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-rnnt/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: reference_accuracy_target. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Run the reference harness with --accuracy on CPU and accuracy_eval.py on the loadgen log (it prints "Word Error Rate: {wer}%, accuracy={(1 - wer)}%", accuracy_eval.py lines 98 to 101); pass when the word accuracy 1 - WER is at least 99% of the reference 1 - 7.452253714852645%, i.e. WER <= 8.377731177704117%; the 15 printed decimals are not a tolerance.
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -20,8 +18,6 @@ Files:
20
 
21
  - `retired_benchmarks/speech_recognition/rnnt/README.md`: MLPerf Inference RNN-T README (retired): reference WER 7.452253714852645% on LibriSpeech dev-clean utterances under 15 s, greedy, FP32.
22
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
23
- Reference result: reference_accuracy_target
24
- Reference source: README line 116: WER over dev-clean (<15 s)
25
  Check: Run the reference harness with --accuracy on CPU and accuracy_eval.py on the loadgen log (it prints "Word Error Rate: {wer}%, accuracy={(1 - wer)}%", accuracy_eval.py lines 98 to 101); pass when the word accuracy 1 - WER is at least 99% of the reference 1 - 7.452253714852645%, i.e. WER <= 8.377731177704117%; the 15 printed decimals are not a tolerance
26
  Companions (not included): retired_benchmarks/speech_recognition/rnnt/accuracy_eval.py; retired_benchmarks/speech_recognition/rnnt/pytorch/parts/manifest.py; retired_benchmarks/speech_recognition/rnnt/pytorch/helpers.py; retired_benchmarks/speech_recognition/rnnt/pytorch/parts/text/__init__.py; retired_benchmarks/speech_recognition/rnnt/pytorch/metrics.py; retired_benchmarks/speech_recognition/rnnt/pytorch/parts/text/cleaners.py; retired_benchmarks/speech_recognition/rnnt/pytorch/parts/text/numbers.py; LICENSE.md; mlcommons/inference_policies@b04253d93958a50b5e776b82ad4dbae9665bb5fd:inference_rules.adoc
27
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). torch 2.14.0 CPU (released 2026-09-02T13:42:48Z, before the commit). RNN-T PyTorch checkpoint on zenodo as named in the README. LibriSpeech dev-clean (CC BY 4.0). unidecode and inflect (parts/text/cleaners.py and numbers.py).
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Run the reference harness with --accuracy on CPU and accuracy_eval.py on the loadgen log (it prints "Word Error Rate: {wer}%, accuracy={(1 - wer)}%", accuracy_eval.py lines 98 to 101); pass when the word accuracy 1 - WER is at least 99% of the reference 1 - 7.452253714852645%, i.e. WER <= 8.377731177704117%; the 15 printed decimals are not a tolerance.
12
 
13
  **Licence (SPDX):** Apache-2.0
 
18
 
19
  - `retired_benchmarks/speech_recognition/rnnt/README.md`: MLPerf Inference RNN-T README (retired): reference WER 7.452253714852645% on LibriSpeech dev-clean utterances under 15 s, greedy, FP32.
20
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
 
 
21
  Check: Run the reference harness with --accuracy on CPU and accuracy_eval.py on the loadgen log (it prints "Word Error Rate: {wer}%, accuracy={(1 - wer)}%", accuracy_eval.py lines 98 to 101); pass when the word accuracy 1 - WER is at least 99% of the reference 1 - 7.452253714852645%, i.e. WER <= 8.377731177704117%; the 15 printed decimals are not a tolerance
22
  Companions (not included): retired_benchmarks/speech_recognition/rnnt/accuracy_eval.py; retired_benchmarks/speech_recognition/rnnt/pytorch/parts/manifest.py; retired_benchmarks/speech_recognition/rnnt/pytorch/helpers.py; retired_benchmarks/speech_recognition/rnnt/pytorch/parts/text/__init__.py; retired_benchmarks/speech_recognition/rnnt/pytorch/metrics.py; retired_benchmarks/speech_recognition/rnnt/pytorch/parts/text/cleaners.py; retired_benchmarks/speech_recognition/rnnt/pytorch/parts/text/numbers.py; LICENSE.md; mlcommons/inference_policies@b04253d93958a50b5e776b82ad4dbae9665bb5fd:inference_rules.adoc
23
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). torch 2.14.0 CPU (released 2026-09-02T13:42:48Z, before the commit). RNN-T PyTorch checkpoint on zenodo as named in the README. LibriSpeech dev-clean (CC BY 4.0). unidecode and inflect (parts/text/cleaners.py and numbers.py).
campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-vision/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: reference_accuracy_target. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Run python/main.py with --accuracy on the CPU backend (onnxruntime or pytorch), then tools/accuracy-imagenet.py or accuracy-openimages.py on mlperf_log_accuracy.json; compare to the table; MLPerf submissions must reach 99% of the reference for resnet50 and retinanet.
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -20,8 +18,6 @@ Files:
20
 
21
  - `vision/classification_and_detection/README.md`: MLPerf Inference reference vision README: accuracy targets for resnet50-v1.5 (tf/onnx 76.456%, pytorch 76.014%, int8 75.790%, ncnn 76.14% top-1), retinanet 800x800 (mAP 0.3755 / 0.3757) and retired mobilenet and ssd rows.
22
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
23
- Reference result: reference_accuracy_target
24
- Reference source: model table lines 38 to 58: top-1 accuracy or mAP per model and framework (lines 40 to 46)
25
  Check: Run python/main.py with --accuracy on the CPU backend (onnxruntime or pytorch), then tools/accuracy-imagenet.py or accuracy-openimages.py on mlperf_log_accuracy.json; compare to the table; MLPerf submissions must reach 99% of the reference for resnet50 and retinanet
26
  Companions (not included): vision/classification_and_detection/tools/accuracy-imagenet.py; vision/classification_and_detection/tools/accuracy-openimages.py; mlcommons/inference_policies@ff7edba545fded369e7e7e3d5a2f0bab4a95eece:inference_rules.adoc; mlcommons/inference_policies@ff7edba545fded369e7e7e3d5a2f0bab4a95eece:LICENSE.md; LICENSE.md
27
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). onnxruntime 1.29.0 (latest PyPI release on or before the commit date, released 2026-08-17) or torch 2.14.0 CPU (released 2026-09-02T13:42:48Z, before the commit). pycocotools (accuracy-openimages.py imports it). ImageNet 2012 validation set (ImageNet terms of access; registration required). OpenImages MLPerf validation subset resized to 800x800 as built by the reference download scripts (vision/classification_and_detection/tools/openimages_mlperf.sh and tools/openimages.py at 3fbc3299; resize by the harness preprocessing); the 264 classes of openimages_mlperf.sh select 24781 validation image IDs (get_image_list, openimages.py lines 227 to 231; the IDs in first-appearance order, one per line with a trailing newline, hash to sha256 3d76037d6cbd080beac8fe92832c9695a1561f013648d9053be17011a9ee90b6), images taken unsigned from s3://open-images-dataset/validation/<ImageID>.jpg (line 54, 243); images not downloaded. model files from the zenodo records named in README.md lines 40 to 46.
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Run python/main.py with --accuracy on the CPU backend (onnxruntime or pytorch), then tools/accuracy-imagenet.py or accuracy-openimages.py on mlperf_log_accuracy.json; compare to the table; MLPerf submissions must reach 99% of the reference for resnet50 and retinanet.
12
 
13
  **Licence (SPDX):** Apache-2.0
 
18
 
19
  - `vision/classification_and_detection/README.md`: MLPerf Inference reference vision README: accuracy targets for resnet50-v1.5 (tf/onnx 76.456%, pytorch 76.014%, int8 75.790%, ncnn 76.14% top-1), retinanet 800x800 (mAP 0.3755 / 0.3757) and retired mobilenet and ssd rows.
20
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
 
 
21
  Check: Run python/main.py with --accuracy on the CPU backend (onnxruntime or pytorch), then tools/accuracy-imagenet.py or accuracy-openimages.py on mlperf_log_accuracy.json; compare to the table; MLPerf submissions must reach 99% of the reference for resnet50 and retinanet
22
  Companions (not included): vision/classification_and_detection/tools/accuracy-imagenet.py; vision/classification_and_detection/tools/accuracy-openimages.py; mlcommons/inference_policies@ff7edba545fded369e7e7e3d5a2f0bab4a95eece:inference_rules.adoc; mlcommons/inference_policies@ff7edba545fded369e7e7e3d5a2f0bab4a95eece:LICENSE.md; LICENSE.md
23
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). onnxruntime 1.29.0 (latest PyPI release on or before the commit date, released 2026-08-17) or torch 2.14.0 CPU (released 2026-09-02T13:42:48Z, before the commit). pycocotools (accuracy-openimages.py imports it). ImageNet 2012 validation set (ImageNet terms of access; registration required). OpenImages MLPerf validation subset resized to 800x800 as built by the reference download scripts (vision/classification_and_detection/tools/openimages_mlperf.sh and tools/openimages.py at 3fbc3299; resize by the harness preprocessing); the 264 classes of openimages_mlperf.sh select 24781 validation image IDs (get_image_list, openimages.py lines 227 to 231; the IDs in first-appearance order, one per line with a trailing newline, hash to sha256 3d76037d6cbd080beac8fe92832c9695a1561f013648d9053be17011a9ee90b6), images taken unsigned from s3://open-images-dataset/validation/<ImageID>.jpg (line 54, 243); images not downloaded. model files from the zenodo records named in README.md lines 40 to 46.
campaign_files/2026-10-ml-serving-configs-2/mlperf-inference-whisper/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: reference_accuracy_target. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Run the reference vLLM CPU harness in accuracy mode, score with accuracy_eval.py; pass when accuracy >= 0.99 * 97.9329%.
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -20,8 +18,6 @@ Files:
20
 
21
  - `speech2text/README.md`: MLPerf Inference speech2text (whisper-large-v3) README: reference WER 2.0671%, accuracy 97.9329%; submissions must reach 99% of it; reference install uses vLLM with VLLM_TARGET_DEVICE=cpu.
22
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
23
- Reference result: reference_accuracy_target
24
- Reference source: README lines 293 to 297: WER and accuracy
25
  Check: Run the reference vLLM CPU harness in accuracy mode, score with accuracy_eval.py; pass when accuracy >= 0.99 * 97.9329%
26
  Companions (not included): speech2text/accuracy_eval.py; speech2text/manifest.py; speech2text/legacy_helpers.py; speech2text/helpers.py; LICENSE.md
27
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). vLLM CPU build as installed by speech2text/README.md line 94. openai/whisper-large-v3 (Apache-2.0). LibriSpeech dev-clean and dev-other (CC BY 4.0) as prepared by the README (line 148). openai-whisper (accuracy_eval.py imports whisper.normalizers.EnglishTextNormalizer).
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Run the reference vLLM CPU harness in accuracy mode, score with accuracy_eval.py; pass when accuracy >= 0.99 * 97.9329%.
12
 
13
  **Licence (SPDX):** Apache-2.0
 
18
 
19
  - `speech2text/README.md`: MLPerf Inference speech2text (whisper-large-v3) README: reference WER 2.0671%, accuracy 97.9329%; submissions must reach 99% of it; reference install uses vLLM with VLLM_TARGET_DEVICE=cpu.
20
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
 
 
21
  Check: Run the reference vLLM CPU harness in accuracy mode, score with accuracy_eval.py; pass when accuracy >= 0.99 * 97.9329%
22
  Companions (not included): speech2text/accuracy_eval.py; speech2text/manifest.py; speech2text/legacy_helpers.py; speech2text/helpers.py; LICENSE.md
23
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). vLLM CPU build as installed by speech2text/README.md line 94. openai/whisper-large-v3 (Apache-2.0). LibriSpeech dev-clean and dev-other (CC BY 4.0) as prepared by the README (line 148). openai-whisper (accuracy_eval.py imports whisper.normalizers.EnglishTextNormalizer).
campaign_files/2026-10-ml-serving-configs-2/modelcloud-gptqmodel/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Whether each file needs other upstream files is not stated. 3 of 3 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: aggregate_metric_reference. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Quantize the model with GPTQ (bits 4, group_size 128, desc_act False, model_test.py lines 211-213), evaluate arc_challenge, and pass when value/expected*100 lies in [100*(1-floor_pct), 100*(1+ceil_pct)] (model_test.py lines 1886-1890).
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -21,24 +19,18 @@ Files:
21
  - `tests/models/test_gpt2.py`: GPTQModel model test gpt2 124M: expected arc_challenge acc 0.1903 and acc_norm 0.227 after 4-bit GPTQ quantization (fast-mode values 0.19368600682593856 and 0.23208191126279865).
22
  Format: Python (unittest subclass holding expected metric values)
23
  Origin: CI reference values kept in the framework's own test tree
24
- Reference result: aggregate_metric_reference
25
- Reference source: tests/models/test_gpt2.py lines 12-17: NATIVE_ARC_CHALLENGE_ACC 0.1903, NATIVE_ARC_CHALLENGE_ACC_NORM 0.227 (slow mode), _FAST 0.19368600682593856 and 0.23208191126279865 (fast mode, the default per model_test.py lines 292-296); compared against the quantized model's arc_challenge result in check_results (model_test.py lines 2040-2080)
26
  Check: Quantize the model with GPTQ (bits 4, group_size 128, desc_act False, model_test.py lines 211-213), evaluate arc_challenge, and pass when value/expected*100 lies in [100*(1-floor_pct), 100*(1+ceil_pct)] (model_test.py lines 1886-1890)
27
  Companions (not included): tests/models/model_test.py; LICENSE
28
  Dependencies: model openai-community/gpt2 (MIT); the test points at a local path /monster/data/model/... calibration data: local copy of a Neural Magic calibration set at /monster/data/model/dataset/nm-calibration (model_test.py line 1352), with a parquet fallback; licence not checked this round; dataset arc_challenge test (allenai/ai2_arc, CC-BY-SA-4.0) via lm-evaluation-harness, public evaluation split, flagged.
29
  - `tests/models/test_opt.py`: GPTQModel model test opt-125m: expected arc_challenge acc 0.192 and acc_norm 0.2253 after 4-bit GPTQ quantization (fast-mode values 0.18430034129692832 and 0.2226962457337884).
30
  Format: Python (unittest subclass holding expected metric values)
31
  Origin: CI reference values kept in the framework's own test tree
32
- Reference result: aggregate_metric_reference
33
- Reference source: tests/models/test_opt.py lines 12-17: NATIVE_ARC_CHALLENGE_ACC 0.192, NATIVE_ARC_CHALLENGE_ACC_NORM 0.2253 (slow mode), _FAST 0.18430034129692832 and 0.2226962457337884 (fast mode, the default per model_test.py lines 292-296); compared against the quantized model's arc_challenge result in check_results (model_test.py lines 2040-2080)
34
  Check: Quantize the model with GPTQ (bits 4, group_size 128, desc_act False, model_test.py lines 211-213), evaluate arc_challenge, and pass when value/expected*100 lies in [100*(1-floor_pct), 100*(1+ceil_pct)] (model_test.py lines 1886-1890)
35
  Companions (not included): tests/models/model_test.py; LICENSE
36
  Dependencies: model facebook/opt-125m (OPT-175B licence, non-commercial research use only); the test points at a local path /monster/data/model/... calibration data: local copy of a Neural Magic calibration set at /monster/data/model/dataset/nm-calibration (model_test.py line 1352), with a parquet fallback; licence not checked this round; dataset arc_challenge test (allenai/ai2_arc, CC-BY-SA-4.0) via lm-evaluation-harness, public evaluation split, flagged.
37
  - `tests/models/test_tinyllama.py`: GPTQModel model test TinyLlama 1.1B chat: expected arc_challenge acc 0.2995 and acc_norm 0.3268 after 4-bit GPTQ quantization (fast-mode values 0.30802047781569963 and 0.3250853242320819).
38
  Format: Python (unittest subclass holding expected metric values)
39
  Origin: CI reference values kept in the framework's own test tree
40
- Reference result: aggregate_metric_reference
41
- Reference source: tests/models/test_tinyllama.py lines 11-16: NATIVE_ARC_CHALLENGE_ACC 0.2995, NATIVE_ARC_CHALLENGE_ACC_NORM 0.3268 (slow mode), _FAST 0.30802047781569963 and 0.3250853242320819 (fast mode, the default per model_test.py lines 292-296); compared against the quantized model's arc_challenge result in check_results (model_test.py lines 2040-2080)
42
  Check: Quantize the model with GPTQ (bits 4, group_size 128, desc_act False, model_test.py lines 211-213), evaluate arc_challenge, and pass when value/expected*100 lies in [100*(1-floor_pct), 100*(1+ceil_pct)] (model_test.py lines 1886-1890)
43
  Companions (not included): tests/models/model_test.py; LICENSE
44
  Dependencies: model TinyLlama/TinyLlama-1.1B-Chat-v1.0 (Apache-2.0); the test points at a local path /monster/data/model/... calibration data: local copy of a Neural Magic calibration set at /monster/data/model/dataset/nm-calibration (model_test.py line 1352), with a parquet fallback; licence not checked this round; dataset arc_challenge test (allenai/ai2_arc, CC-BY-SA-4.0) via lm-evaluation-harness, public evaluation split, flagged.
 
8
 
9
  **Self-contained:** Whether each file needs other upstream files is not stated. 3 of 3 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Quantize the model with GPTQ (bits 4, group_size 128, desc_act False, model_test.py lines 211-213), evaluate arc_challenge, and pass when value/expected*100 lies in [100*(1-floor_pct), 100*(1+ceil_pct)] (model_test.py lines 1886-1890).
12
 
13
  **Licence (SPDX):** Apache-2.0
 
19
  - `tests/models/test_gpt2.py`: GPTQModel model test gpt2 124M: expected arc_challenge acc 0.1903 and acc_norm 0.227 after 4-bit GPTQ quantization (fast-mode values 0.19368600682593856 and 0.23208191126279865).
20
  Format: Python (unittest subclass holding expected metric values)
21
  Origin: CI reference values kept in the framework's own test tree
 
 
22
  Check: Quantize the model with GPTQ (bits 4, group_size 128, desc_act False, model_test.py lines 211-213), evaluate arc_challenge, and pass when value/expected*100 lies in [100*(1-floor_pct), 100*(1+ceil_pct)] (model_test.py lines 1886-1890)
23
  Companions (not included): tests/models/model_test.py; LICENSE
24
  Dependencies: model openai-community/gpt2 (MIT); the test points at a local path /monster/data/model/... calibration data: local copy of a Neural Magic calibration set at /monster/data/model/dataset/nm-calibration (model_test.py line 1352), with a parquet fallback; licence not checked this round; dataset arc_challenge test (allenai/ai2_arc, CC-BY-SA-4.0) via lm-evaluation-harness, public evaluation split, flagged.
25
  - `tests/models/test_opt.py`: GPTQModel model test opt-125m: expected arc_challenge acc 0.192 and acc_norm 0.2253 after 4-bit GPTQ quantization (fast-mode values 0.18430034129692832 and 0.2226962457337884).
26
  Format: Python (unittest subclass holding expected metric values)
27
  Origin: CI reference values kept in the framework's own test tree
 
 
28
  Check: Quantize the model with GPTQ (bits 4, group_size 128, desc_act False, model_test.py lines 211-213), evaluate arc_challenge, and pass when value/expected*100 lies in [100*(1-floor_pct), 100*(1+ceil_pct)] (model_test.py lines 1886-1890)
29
  Companions (not included): tests/models/model_test.py; LICENSE
30
  Dependencies: model facebook/opt-125m (OPT-175B licence, non-commercial research use only); the test points at a local path /monster/data/model/... calibration data: local copy of a Neural Magic calibration set at /monster/data/model/dataset/nm-calibration (model_test.py line 1352), with a parquet fallback; licence not checked this round; dataset arc_challenge test (allenai/ai2_arc, CC-BY-SA-4.0) via lm-evaluation-harness, public evaluation split, flagged.
31
  - `tests/models/test_tinyllama.py`: GPTQModel model test TinyLlama 1.1B chat: expected arc_challenge acc 0.2995 and acc_norm 0.3268 after 4-bit GPTQ quantization (fast-mode values 0.30802047781569963 and 0.3250853242320819).
32
  Format: Python (unittest subclass holding expected metric values)
33
  Origin: CI reference values kept in the framework's own test tree
 
 
34
  Check: Quantize the model with GPTQ (bits 4, group_size 128, desc_act False, model_test.py lines 211-213), evaluate arc_challenge, and pass when value/expected*100 lies in [100*(1-floor_pct), 100*(1+ceil_pct)] (model_test.py lines 1886-1890)
35
  Companions (not included): tests/models/model_test.py; LICENSE
36
  Dependencies: model TinyLlama/TinyLlama-1.1B-Chat-v1.0 (Apache-2.0); the test points at a local path /monster/data/model/... calibration data: local copy of a Neural Magic calibration set at /monster/data/model/dataset/nm-calibration (model_test.py line 1352), with a parquet fallback; licence not checked this round; dataset arc_challenge test (allenai/ai2_arc, CC-BY-SA-4.0) via lm-evaluation-harness, public evaluation split, flagged.
campaign_files/2026-10-ml-serving-configs-2/musl-libc-test-math/LICENSE ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ Licence: MIT
2
+
3
+ - MIT: https://repo.or.cz/libc-test.git/blob/7b95dfa5f5d5ca4d949221e0228ccc290bacc14e:/COPYRIGHT (publisher: libc-test (musl libc conformance and math accuracy tests))
campaign_files/2026-10-ml-serving-configs-2/musl-libc-test-math/README.md ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # musl-libc-test-math
2
+
3
+ **What this is:** 1 file published by libc-test (musl libc conformance and math accuracy tests). Collection theme: Operator reference implementations and test vectors (backend tests of operator specifications, reference implementations and test data of operator libraries).
4
+
5
+ **What one unit is:** One upstream file per unit, kept at its path in the upstream repository or site. Format: C source plus two C include files of T(rounding mode, x, y, dy, exceptions) rows with hex float literals.
6
+
7
+ **Where it comes from:** Upstream: repo.or.cz/libc-test.git. Kind: test or example files shipped with a tool. Accuracy tests of a C math library.
8
+
9
+ **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
+
11
+ **Tools to open and check:** Check: For each row with round-to-nearest, compute acosh(x), then ulperr(got, y, dy) of src/common/mtest.c lines 58-72 (difference in ulps of y plus dy); the row passes when checkulp holds (src/common/mtest.h lines 120-127: |d| < 1.5 in round to nearest, < 3.0 otherwise); acosh.c lines 37-41 count a failing row as an error only when |d| is at least 2.0 (smaller ones are printed as X); the raised exceptions must match the e column (checkexcept).
12
+
13
+ **Licence (SPDX):** MIT
14
+
15
+ **Provenance:** https://repo.or.cz/libc-test.git/blob/7b95dfa5f5d5ca4d949221e0228ccc290bacc14e:/src/math/acosh.c
16
+
17
+ Files:
18
+
19
+ - `src/math/acosh.c`: Libc-test math accuracy vectors for acosh: 34 rows of input, correctly rounded result, residual dy in ulps and expected floating point exceptions.
20
+ Format: C source plus two C include files of T(rounding mode, x, y, dy, exceptions) rows with hex float literals
21
+ Origin: accuracy tests of a C math library
22
+ Check: For each row with round-to-nearest, compute acosh(x), then ulperr(got, y, dy) of src/common/mtest.c lines 58-72 (difference in ulps of y plus dy); the row passes when checkulp holds (src/common/mtest.h lines 120-127: |d| < 1.5 in round to nearest, < 3.0 otherwise); acosh.c lines 37-41 count a failing row as an error only when |d| is at least 2.0 (smaller ones are printed as X); the raised exceptions must match the e column (checkexcept)
23
+ Companions (not included): src/math/sanity/acosh.h; src/math/special/acosh.h; src/common/mtest.h; src/common/mtest.c; COPYRIGHT
24
+ Dependencies: to run the original test: a C compiler and libc-test at commit 7b95dfa5.
campaign_files/2026-10-ml-serving-configs-2/nab-artificialwithanomaly/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: labelled_anomaly_windows. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: The agent emits one anomaly score per row (or a list of detection timestamps). Score with the NAB standard profile: a detection inside a window counts as a true positive weighted by the scaled sigmoid of its position (earlier is better), only the first detection per window counts, a detection outside every window is a false positive with weight fpWeight 0.11, a missed window costs fnWeight 1.0; rows in the probationary period (the first 15 percent of the file, capped at 750 rows; here 604 rows) are not scored; normalise to the NAB score where the null detector is 0 and the perfect detector is 100. A simpler acceptance check that needs no threshold sweep: the reported detection timestamps hit every labelled window and fall outside the windows at most as often as the expected answer file allows.
14
 
15
  **Licence (SPDX):** MIT
@@ -21,8 +19,6 @@ Files:
21
  - `data/artificialWithAnomaly/art_daily_jumpsup.csv`: NAB artificialWithAnomaly/art_daily_jumpsup.csv: synthetic daily-periodic series with an upward level jump, 1 labelled anomaly window; 4032 rows, 2014-04-01 00:00:00 to 2014-04-14 23:55:00, 5-minute steps.
22
  Format: CSV (timestamp,value) plus JSON label windows keyed by file name
23
  Origin: synthetic series shipped with the NAB benchmark (artificialWithAnomaly), made by Numenta to test detectors.
24
- Reference result: labelled_anomaly_windows
25
- Reference source: labels/combined_windows.json key "artificialWithAnomaly/art_daily_jumpsup.csv": 1 anomaly window(s) [["2014-04-10 16:15:00.000000", "2014-04-12 01:45:00.000000"]]; labels/combined_labels.json gives the anomaly timestamps the windows are centred on; labels made by Numenta with its labeling process (README.md lines 75 and 86: labeled anomalous periods)
26
  Check: The agent emits one anomaly score per row (or a list of detection timestamps). Score with the NAB standard profile: a detection inside a window counts as a true positive weighted by the scaled sigmoid of its position (earlier is better), only the first detection per window counts, a detection outside every window is a false positive with weight fpWeight 0.11, a missed window costs fnWeight 1.0; rows in the probationary period (the first 15 percent of the file, capped at 750 rows; here 604 rows) are not scored; normalise to the NAB score where the null detector is 0 and the perfect detector is 100. A simpler acceptance check that needs no threshold sweep: the reported detection timestamps hit every labelled window and fall outside the windows at most as often as the expected answer file allows
27
  Companions (not included): labels/combined_windows.json; labels/combined_labels.json; config/profiles.json; LICENSE.txt
28
  Dependencies: none to read the data (CSV with columns timestamp,value; JSON labels); scoring follows the NAB scorer at the pinned commit (nab/sweeper.py, nab/scorer.py, MIT, Python), which the judge may reimplement from the cited lines.
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: The agent emits one anomaly score per row (or a list of detection timestamps). Score with the NAB standard profile: a detection inside a window counts as a true positive weighted by the scaled sigmoid of its position (earlier is better), only the first detection per window counts, a detection outside every window is a false positive with weight fpWeight 0.11, a missed window costs fnWeight 1.0; rows in the probationary period (the first 15 percent of the file, capped at 750 rows; here 604 rows) are not scored; normalise to the NAB score where the null detector is 0 and the perfect detector is 100. A simpler acceptance check that needs no threshold sweep: the reported detection timestamps hit every labelled window and fall outside the windows at most as often as the expected answer file allows.
12
 
13
  **Licence (SPDX):** MIT
 
19
  - `data/artificialWithAnomaly/art_daily_jumpsup.csv`: NAB artificialWithAnomaly/art_daily_jumpsup.csv: synthetic daily-periodic series with an upward level jump, 1 labelled anomaly window; 4032 rows, 2014-04-01 00:00:00 to 2014-04-14 23:55:00, 5-minute steps.
20
  Format: CSV (timestamp,value) plus JSON label windows keyed by file name
21
  Origin: synthetic series shipped with the NAB benchmark (artificialWithAnomaly), made by Numenta to test detectors.
 
 
22
  Check: The agent emits one anomaly score per row (or a list of detection timestamps). Score with the NAB standard profile: a detection inside a window counts as a true positive weighted by the scaled sigmoid of its position (earlier is better), only the first detection per window counts, a detection outside every window is a false positive with weight fpWeight 0.11, a missed window costs fnWeight 1.0; rows in the probationary period (the first 15 percent of the file, capped at 750 rows; here 604 rows) are not scored; normalise to the NAB score where the null detector is 0 and the perfect detector is 100. A simpler acceptance check that needs no threshold sweep: the reported detection timestamps hit every labelled window and fall outside the windows at most as often as the expected answer file allows
23
  Companions (not included): labels/combined_windows.json; labels/combined_labels.json; config/profiles.json; LICENSE.txt
24
  Dependencies: none to read the data (CSV with columns timestamp,value; JSON labels); scoring follows the NAB scorer at the pinned commit (nab/sweeper.py, nab/scorer.py, MIT, Python), which the judge may reimplement from the cited lines.
campaign_files/2026-10-ml-serving-configs-2/nab-realawscloudwatch/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: labelled_anomaly_windows. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: The agent emits one anomaly score per row (or a list of detection timestamps). Score with the NAB standard profile: a detection inside a window counts as a true positive weighted by the scaled sigmoid of its position (earlier is better), only the first detection per window counts, a detection outside every window is a false positive with weight fpWeight 0.11, a missed window costs fnWeight 1.0; rows in the probationary period (the first 15 percent of the file, capped at 750 rows; here 604 rows) are not scored; normalise to the NAB score where the null detector is 0 and the perfect detector is 100. A simpler acceptance check that needs no threshold sweep: the reported detection timestamps hit every labelled window and fall outside the windows at most as often as the expected answer file allows.
14
 
15
  **Licence (SPDX):** MIT
@@ -21,8 +19,6 @@ Files:
21
  - `data/realAWSCloudwatch/ec2_cpu_utilization_825cc2.csv`: NAB realAWSCloudwatch/ec2_cpu_utilization_825cc2.csv: EC2 CPU utilisation monitoring metric, 1 labelled anomaly window; 4032 rows, 2014-04-10 00:04:00 to 2014-04-24 00:09:00, 5-minute steps.
22
  Format: CSV (timestamp,value) plus JSON label windows keyed by file name
23
  Origin: AWS Cloudwatch EC2 CPU utilisation metric published by Numenta in NAB
24
- Reference result: labelled_anomaly_windows
25
- Reference source: labels/combined_windows.json key "realAWSCloudwatch/ec2_cpu_utilization_825cc2.csv": 1 anomaly window(s) [["2014-04-15 07:24:00.000000", "2014-04-16 11:54:00.000000"]]; labels/combined_labels.json gives the anomaly timestamps the windows are centred on; labels made by Numenta with its labeling process (README.md lines 75 and 86: labeled anomalous periods)
26
  Check: The agent emits one anomaly score per row (or a list of detection timestamps). Score with the NAB standard profile: a detection inside a window counts as a true positive weighted by the scaled sigmoid of its position (earlier is better), only the first detection per window counts, a detection outside every window is a false positive with weight fpWeight 0.11, a missed window costs fnWeight 1.0; rows in the probationary period (the first 15 percent of the file, capped at 750 rows; here 604 rows) are not scored; normalise to the NAB score where the null detector is 0 and the perfect detector is 100. A simpler acceptance check that needs no threshold sweep: the reported detection timestamps hit every labelled window and fall outside the windows at most as often as the expected answer file allows
27
  Companions (not included): labels/combined_windows.json; labels/combined_labels.json; config/profiles.json; LICENSE.txt
28
  Dependencies: none to read the data (CSV with columns timestamp,value; JSON labels); scoring follows the NAB scorer at the pinned commit (nab/sweeper.py, nab/scorer.py, MIT, Python), which the judge may reimplement from the cited lines.
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: The agent emits one anomaly score per row (or a list of detection timestamps). Score with the NAB standard profile: a detection inside a window counts as a true positive weighted by the scaled sigmoid of its position (earlier is better), only the first detection per window counts, a detection outside every window is a false positive with weight fpWeight 0.11, a missed window costs fnWeight 1.0; rows in the probationary period (the first 15 percent of the file, capped at 750 rows; here 604 rows) are not scored; normalise to the NAB score where the null detector is 0 and the perfect detector is 100. A simpler acceptance check that needs no threshold sweep: the reported detection timestamps hit every labelled window and fall outside the windows at most as often as the expected answer file allows.
12
 
13
  **Licence (SPDX):** MIT
 
19
  - `data/realAWSCloudwatch/ec2_cpu_utilization_825cc2.csv`: NAB realAWSCloudwatch/ec2_cpu_utilization_825cc2.csv: EC2 CPU utilisation monitoring metric, 1 labelled anomaly window; 4032 rows, 2014-04-10 00:04:00 to 2014-04-24 00:09:00, 5-minute steps.
20
  Format: CSV (timestamp,value) plus JSON label windows keyed by file name
21
  Origin: AWS Cloudwatch EC2 CPU utilisation metric published by Numenta in NAB
 
 
22
  Check: The agent emits one anomaly score per row (or a list of detection timestamps). Score with the NAB standard profile: a detection inside a window counts as a true positive weighted by the scaled sigmoid of its position (earlier is better), only the first detection per window counts, a detection outside every window is a false positive with weight fpWeight 0.11, a missed window costs fnWeight 1.0; rows in the probationary period (the first 15 percent of the file, capped at 750 rows; here 604 rows) are not scored; normalise to the NAB score where the null detector is 0 and the perfect detector is 100. A simpler acceptance check that needs no threshold sweep: the reported detection timestamps hit every labelled window and fall outside the windows at most as often as the expected answer file allows
23
  Companions (not included): labels/combined_windows.json; labels/combined_labels.json; config/profiles.json; LICENSE.txt
24
  Dependencies: none to read the data (CSV with columns timestamp,value; JSON labels); scoring follows the NAB scorer at the pinned commit (nab/sweeper.py, nab/scorer.py, MIT, Python), which the judge may reimplement from the cited lines.
campaign_files/2026-10-ml-serving-configs/mlperf-inference-llama31-8b/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: reference_accuracy_target. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Run the reference vLLM harness in accuracy mode, score with evaluation.py; pass when each ROUGE >= 0.99 * target and gen_len >= 0.9 * target.
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -20,8 +18,6 @@ Files:
20
 
21
  - `language/llama3.1-8b/README.md`: MLPerf Inference Llama 3.1 8B summarisation README: datacenter BF16 targets rouge1 38.7792, rouge2 15.9075, rougeL 24.4957, rougeLsum 35.793, gen_len 8167644 over 13368 samples, plus edge targets; 99% for ROUGE, 90% for gen_len.
22
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
23
- Reference result: reference_accuracy_target
24
- Reference source: README lines 321 to 344: ROUGE scores and gen_len
25
  Check: Run the reference vLLM harness in accuracy mode, score with evaluation.py; pass when each ROUGE >= 0.99 * target and gen_len >= 0.9 * target
26
  Companions (not included): language/llama3.1-8b/evaluation.py; language/llama3.1-8b/dataset.py; LICENSE.md; tools/submission/submission_checker/constants.py
27
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). vLLM as used by the reference implementation. meta-llama/Llama-3.1-8B-Instruct (Llama 3.1 Community License; gated download on Hugging Face); Hugging Face model repo meta-llama/Llama-3.1-8B-Instruct. CNN/DailyMail 3.0.0 validation as preprocessed by MLCommons. rouge-score and nltk (evaluation.py). datasets, transformers, torch and evaluate (evaluation.py and dataset.py imports).
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Run the reference vLLM harness in accuracy mode, score with evaluation.py; pass when each ROUGE >= 0.99 * target and gen_len >= 0.9 * target.
12
 
13
  **Licence (SPDX):** Apache-2.0
 
18
 
19
  - `language/llama3.1-8b/README.md`: MLPerf Inference Llama 3.1 8B summarisation README: datacenter BF16 targets rouge1 38.7792, rouge2 15.9075, rougeL 24.4957, rougeLsum 35.793, gen_len 8167644 over 13368 samples, plus edge targets; 99% for ROUGE, 90% for gen_len.
20
  Format: Markdown README of an MLPerf Inference reference implementation with a model and accuracy table
 
 
21
  Check: Run the reference vLLM harness in accuracy mode, score with evaluation.py; pass when each ROUGE >= 0.99 * target and gen_len >= 0.9 * target
22
  Companions (not included): language/llama3.1-8b/evaluation.py; language/llama3.1-8b/dataset.py; LICENSE.md; tools/submission/submission_checker/constants.py
23
  Dependencies: mlcommons/inference reference harness at commit 3fbc329939999c13d0a7b5e67fb2092287e06047 (committed 2026-09-02T17:02:36Z; run scripts, SUT code). mlcommons-loadgen 6.0.17 (latest PyPI release on or before the commit date, released 2026-07-24). vLLM as used by the reference implementation. meta-llama/Llama-3.1-8B-Instruct (Llama 3.1 Community License; gated download on Hugging Face); Hugging Face model repo meta-llama/Llama-3.1-8B-Instruct. CNN/DailyMail 3.0.0 validation as preprocessed by MLCommons. rouge-score and nltk (evaluation.py). datasets, transformers, torch and evaluate (evaluation.py and dataset.py imports).
campaign_files/2026-10-s-security-detection-2/modsecurity-v3-regression-tests/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 3 of 3 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: expected_http_code_and_log; expected_http_code. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Task: write a rule that captures card numbers in request arguments, or extract the Luhn-valid number. Score: run with the ModSecurity v3 regression runner; HTTP code must equal 200 and the debug log must contain the expected line (regex search as the runner does); exact match; Task: write the rule chain that denies XSS arguments, or classify each request. Score: run with the ModSecurity v3 regression runner; each case passes when the HTTP code equals the expected code; exact match, all 9 cases; Task: write the configuration and rules that refuse the external-entity documents and accept the plain one, or classify each request. Score: run with the ModSecurity v3 regression runner; each case passes when the HTTP code equals the expected code (the runner also matches the expected debug_log line); exact match, all 3 cases.
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -21,24 +19,18 @@ Files:
21
  - `test/test-cases/regression/operator-verifycc.json`: ModSecurity @verifycc operator: find the Luhn-valid credit card number in a POST body and expect the capture in the debug log.
22
  Format: JSON list with 1 case (inline SecRules, request, response, expected debug_log and http_code)
23
  Origin: test cases written by hand by the rule maintainers to pin the expected verdict of their own rule (constructed, not recorded traffic)
24
- Reference result: expected_http_code_and_log
25
- Reference source: operator-verifycc.json: expected.debug_log "Added VerifyCC match TX.0: 5484605089158216" and expected.http_code 200; rule SecRule ARGS "@verifycc \\d{13,16}" in the rules list
26
  Check: Task: write a rule that captures card numbers in request arguments, or extract the Luhn-valid number. Score: run with the ModSecurity v3 regression runner; HTTP code must equal 200 and the debug log must contain the expected line (regex search as the runner does); exact match
27
  Companions (not included): LICENSE
28
  Dependencies: libmodsecurity v3 at this commit with its regression runner: needed only to rerun the case; version pinned when the pool is built; not packed
29
  - `test/test-cases/regression/operator-detectxss.json`: ModSecurity @detectXSS operator (libinjection): 9 cases, 7 XSS payloads must be denied with 403 and 2 benign inputs must pass with 200.
30
  Format: JSON list with 9 cases (inline SecRules, request, response, expected http_code), 10,575 B
31
  Origin: test cases written by hand by the maintainers to pin the expected verdict of their own rules (constructed, not recorded traffic)
32
- Reference result: expected_http_code
33
- Reference source: operator-detectxss.json: expected.http_code 403 at lines 40, 86, 132, 178, 224, 362, 408 (cases 1 to 5, 8, 9) and 200 at lines 270, 316 (cases 6 and 7, benign html and plain text)
34
  Check: Task: write the rule chain that denies XSS arguments, or classify each request. Score: run with the ModSecurity v3 regression runner; each case passes when the HTTP code equals the expected code; exact match, all 9 cases
35
  Companions (not included): LICENSE
36
  Dependencies: libmodsecurity v3 at this commit with its regression runner: needed only to rerun the cases; version pinned when the pool is built; not packed
37
  - `test/test-cases/regression/config-xml_external_entity.json`: ModSecurity XML external entity handling (SecXMLExternalEntity): 3 cases, an internal entity must parse and pass with 200, two documents declaring a SYSTEM entity for /etc/passwd must be denied with 403 by DTD validation.
38
  Format: JSON list with 3 cases (inline SecRules with SecXMLExternalEntity Off or On and an @validateDTD rule, XML request body, expected http_code and debug_log), 4,787 B; DTD test/test-cases/data/SoapEnvelope.dtd
39
  Origin: test cases written by hand by the maintainers to pin the expected verdict of their own rules (constructed, not recorded traffic)
40
- Reference result: expected_http_code
41
- Reference source: config-xml_external_entity.json: expected.http_code 200 at line 47 (case 1, internal entity "Jo Smith", debug_log target value at line 46), 403 at line 102 (case 2, XXE Off, debug_log "Failed to load DTD" at line 101) and 403 at line 158 (case 3, XXE On, debug_log "No declaration for element bookstore" at line 157)
42
  Check: Task: write the configuration and rules that refuse the external-entity documents and accept the plain one, or classify each request. Score: run with the ModSecurity v3 regression runner; each case passes when the HTTP code equals the expected code (the runner also matches the expected debug_log line); exact match, all 3 cases
43
  Companions (not included): test/test-cases/data/SoapEnvelope.dtd; LICENSE
44
  Dependencies: libmodsecurity v3 built with libxml2 at this commit with its regression runner: needed only to rerun the cases; version pinned when the pool is built; not packed
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 3 of 3 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Task: write a rule that captures card numbers in request arguments, or extract the Luhn-valid number. Score: run with the ModSecurity v3 regression runner; HTTP code must equal 200 and the debug log must contain the expected line (regex search as the runner does); exact match; Task: write the rule chain that denies XSS arguments, or classify each request. Score: run with the ModSecurity v3 regression runner; each case passes when the HTTP code equals the expected code; exact match, all 9 cases; Task: write the configuration and rules that refuse the external-entity documents and accept the plain one, or classify each request. Score: run with the ModSecurity v3 regression runner; each case passes when the HTTP code equals the expected code (the runner also matches the expected debug_log line); exact match, all 3 cases.
12
 
13
  **Licence (SPDX):** Apache-2.0
 
19
  - `test/test-cases/regression/operator-verifycc.json`: ModSecurity @verifycc operator: find the Luhn-valid credit card number in a POST body and expect the capture in the debug log.
20
  Format: JSON list with 1 case (inline SecRules, request, response, expected debug_log and http_code)
21
  Origin: test cases written by hand by the rule maintainers to pin the expected verdict of their own rule (constructed, not recorded traffic)
 
 
22
  Check: Task: write a rule that captures card numbers in request arguments, or extract the Luhn-valid number. Score: run with the ModSecurity v3 regression runner; HTTP code must equal 200 and the debug log must contain the expected line (regex search as the runner does); exact match
23
  Companions (not included): LICENSE
24
  Dependencies: libmodsecurity v3 at this commit with its regression runner: needed only to rerun the case; version pinned when the pool is built; not packed
25
  - `test/test-cases/regression/operator-detectxss.json`: ModSecurity @detectXSS operator (libinjection): 9 cases, 7 XSS payloads must be denied with 403 and 2 benign inputs must pass with 200.
26
  Format: JSON list with 9 cases (inline SecRules, request, response, expected http_code), 10,575 B
27
  Origin: test cases written by hand by the maintainers to pin the expected verdict of their own rules (constructed, not recorded traffic)
 
 
28
  Check: Task: write the rule chain that denies XSS arguments, or classify each request. Score: run with the ModSecurity v3 regression runner; each case passes when the HTTP code equals the expected code; exact match, all 9 cases
29
  Companions (not included): LICENSE
30
  Dependencies: libmodsecurity v3 at this commit with its regression runner: needed only to rerun the cases; version pinned when the pool is built; not packed
31
  - `test/test-cases/regression/config-xml_external_entity.json`: ModSecurity XML external entity handling (SecXMLExternalEntity): 3 cases, an internal entity must parse and pass with 200, two documents declaring a SYSTEM entity for /etc/passwd must be denied with 403 by DTD validation.
32
  Format: JSON list with 3 cases (inline SecRules with SecXMLExternalEntity Off or On and an @validateDTD rule, XML request body, expected http_code and debug_log), 4,787 B; DTD test/test-cases/data/SoapEnvelope.dtd
33
  Origin: test cases written by hand by the maintainers to pin the expected verdict of their own rules (constructed, not recorded traffic)
 
 
34
  Check: Task: write the configuration and rules that refuse the external-entity documents and accept the plain one, or classify each request. Score: run with the ModSecurity v3 regression runner; each case passes when the HTTP code equals the expected code (the runner also matches the expected debug_log line); exact match, all 3 cases
35
  Companions (not included): test/test-cases/data/SoapEnvelope.dtd; LICENSE
36
  Dependencies: libmodsecurity v3 built with libxml2 at this commit with its regression runner: needed only to rerun the cases; version pinned when the pool is built; not packed
campaign_files/2026-10-s-security-detection-2/mq-hackmac-2021-public-challenges/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Each unit is self-contained. Dependencies named upstream are listed per unit in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: flag_string. Where the reference result comes from and how a solution is checked is written per unit in the notes.
12
-
13
  **Tools to open and check:** Check: Exact string match of the submitted flag against the challenge yaml flags value; no tolerance.
14
 
15
  **Licence (SPDX):** NOASSERTION
@@ -21,35 +19,25 @@ Units:
21
  - `forensics/cathex/publish/CAT.jpg`: MQ HackMac 2021 forensics/cathex: recover the flag hidden by steganography in a single JPEG image.
22
  Format: one JPEG image CAT.jpg (3,483,943 B, 2456x3000), a painted artwork of cats in Roman dress; its sha256 matches the value declared in ctfd-challenge.yaml
23
  Origin: official competition instance built by the organisers for MQ HackMac 2021, a public CTF, and released with its flag and official solution (fixture of a competition, not recorded production telemetry)
24
- Reference result: flag_string
25
- Reference source: ctfd-challenge.yaml line 32 gives HACKMAC{c4TC43s4RsTyl3}; WRITEUP.md is the official solution. The declared sha256 matches the fetched CAT.jpg.
26
  Check: Exact string match of the submitted flag against the challenge yaml flags value; no tolerance
27
  Dependencies: the challenge yaml / FLAG.txt / writeup hold the flag and the official solution path (hidden from the solver); solver tools are free choices (for example an archive, image, pcap or office-format library), version fixed when the pool is built
28
  - `forensics/flagflagflagflag/publish/flag`: MQ HackMac 2021 forensics/flagflagflagflag: repair a PNG whose signature and chunk names are corrupted so that it renders the flag.
29
  Format: one file "flag" (29,865 B), a PNG with a deliberately corrupted signature and chunk names (for example 89 65 4E 34..., "ciDR"); sha256 matches the yaml
30
  Origin: official competition instance built by the organisers for MQ HackMac 2021, a public CTF, and released with its flag and official solution (fixture of a competition, not recorded production telemetry)
31
- Reference result: flag_string
32
- Reference source: ctfd-challenge.yaml line 33 gives HACKMAC{flag_FLA9_FL4G_F1_AG_!=LAG}; the declared sha256 of publish/flag matches the fetched file.
33
  Check: Exact string match of the submitted flag against the challenge yaml flags value; no tolerance
34
  Dependencies: the challenge yaml / FLAG.txt / writeup hold the flag and the official solution path (hidden from the solver); solver tools are free choices (for example an archive, image, pcap or office-format library), version fixed when the pool is built
35
  - `forensics/Crimson/` (4 files): MQ HackMac 2021 forensics/Crimson Society: recover the flag from data hidden across a set of four carrier images.
36
  Format: four JPEG carrier images illuminate1-4.jpg (stylised Illuminati eye-in-triangle graphics); the challenge/ directory works as one unit
37
  Origin: official competition instance built by the organisers for MQ HackMac 2021, a public CTF, and released with its flag and official solution (fixture of a competition, not recorded production telemetry)
38
- Reference result: flag_string
39
- Reference source: FLAG.txt line 1 and ctfd-challenge.yaml line 35 give HACKMAC{Masquerade}; WRITEUP.md is the official solution. (embedded-files/ holds the extracted solution files and is not part of the unit.)
40
  Check: Exact string match of the submitted flag against the challenge yaml flags value; no tolerance
41
  Dependencies: the challenge yaml / FLAG.txt / writeup hold the flag and the official solution path (hidden from the solver); solver tools are free choices (for example an archive, image, pcap or office-format library), version fixed when the pool is built
42
  - `forensics/Lost_but_not_forgotten/` (10 files): MQ HackMac 2021 forensics/Lost_but_not_forgotten: reconstruct the flag from a Volatility memory-forensics walkthrough given as ten screenshots.
43
  Format: ten PNG screenshots (1.PNG..10.PNG) of a Volatility 2.6 memory-forensics session on Kali Linux against a Windows memory image; the images/ directory works as one unit
44
  Origin: official competition instance built by the organisers for MQ HackMac 2021, a public CTF, and released with its flag and official solution (fixture of a competition, not recorded production telemetry)
45
- Reference result: flag_string
46
- Reference source: FLAG.txt line 1 and ctfd-challenge.yaml line 34 give HACKMAC{5h4rK5_aR3_aP3X_pR3DaT0r5}.
47
  Check: Exact string match of the submitted flag against the challenge yaml flags value; no tolerance
48
  Dependencies: the challenge yaml / FLAG.txt / writeup hold the flag and the official solution path (hidden from the solver); solver tools are free choices (for example an archive, image, pcap or office-format library), version fixed when the pool is built
49
  - `forensics/I_recognise_this/publish/flag`: MQ HackMac 2021 forensics/I_recognise_this: identify nested compression layers (XZ then ZIP then TAR) and peel them to reach the flag.
50
  Format: one file "flag" (896 B): XZ containing a ZIP (member "flag", 10,240 B, not encrypted) containing a TAR; three layers, none encrypted
51
  Origin: official competition instance built by the organisers for MQ HackMac 2021, a public CTF, and released with its flag and official solution (fixture of a competition, not recorded production telemetry)
52
- Reference result: flag_string
53
- Reference source: ctfd-challenge.yaml line 33 and FLAG.txt give HACKMAC{EnCrYpT10n_c0nV3rSi0n_&_c0mPr3s51On}; the declared sha256 of publish/flag matches; solve/solve.py cited only.
54
  Check: Exact string match of the submitted flag against the challenge yaml flags value; no tolerance
55
  Dependencies: the challenge yaml / FLAG.txt / writeup hold the flag and the official solution path (hidden from the solver); solver tools are free choices (for example an archive, image, pcap or office-format library), version fixed when the pool is built
 
8
 
9
  **Self-contained:** Each unit is self-contained. Dependencies named upstream are listed per unit in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Exact string match of the submitted flag against the challenge yaml flags value; no tolerance.
12
 
13
  **Licence (SPDX):** NOASSERTION
 
19
  - `forensics/cathex/publish/CAT.jpg`: MQ HackMac 2021 forensics/cathex: recover the flag hidden by steganography in a single JPEG image.
20
  Format: one JPEG image CAT.jpg (3,483,943 B, 2456x3000), a painted artwork of cats in Roman dress; its sha256 matches the value declared in ctfd-challenge.yaml
21
  Origin: official competition instance built by the organisers for MQ HackMac 2021, a public CTF, and released with its flag and official solution (fixture of a competition, not recorded production telemetry)
 
 
22
  Check: Exact string match of the submitted flag against the challenge yaml flags value; no tolerance
23
  Dependencies: the challenge yaml / FLAG.txt / writeup hold the flag and the official solution path (hidden from the solver); solver tools are free choices (for example an archive, image, pcap or office-format library), version fixed when the pool is built
24
  - `forensics/flagflagflagflag/publish/flag`: MQ HackMac 2021 forensics/flagflagflagflag: repair a PNG whose signature and chunk names are corrupted so that it renders the flag.
25
  Format: one file "flag" (29,865 B), a PNG with a deliberately corrupted signature and chunk names (for example 89 65 4E 34..., "ciDR"); sha256 matches the yaml
26
  Origin: official competition instance built by the organisers for MQ HackMac 2021, a public CTF, and released with its flag and official solution (fixture of a competition, not recorded production telemetry)
 
 
27
  Check: Exact string match of the submitted flag against the challenge yaml flags value; no tolerance
28
  Dependencies: the challenge yaml / FLAG.txt / writeup hold the flag and the official solution path (hidden from the solver); solver tools are free choices (for example an archive, image, pcap or office-format library), version fixed when the pool is built
29
  - `forensics/Crimson/` (4 files): MQ HackMac 2021 forensics/Crimson Society: recover the flag from data hidden across a set of four carrier images.
30
  Format: four JPEG carrier images illuminate1-4.jpg (stylised Illuminati eye-in-triangle graphics); the challenge/ directory works as one unit
31
  Origin: official competition instance built by the organisers for MQ HackMac 2021, a public CTF, and released with its flag and official solution (fixture of a competition, not recorded production telemetry)
 
 
32
  Check: Exact string match of the submitted flag against the challenge yaml flags value; no tolerance
33
  Dependencies: the challenge yaml / FLAG.txt / writeup hold the flag and the official solution path (hidden from the solver); solver tools are free choices (for example an archive, image, pcap or office-format library), version fixed when the pool is built
34
  - `forensics/Lost_but_not_forgotten/` (10 files): MQ HackMac 2021 forensics/Lost_but_not_forgotten: reconstruct the flag from a Volatility memory-forensics walkthrough given as ten screenshots.
35
  Format: ten PNG screenshots (1.PNG..10.PNG) of a Volatility 2.6 memory-forensics session on Kali Linux against a Windows memory image; the images/ directory works as one unit
36
  Origin: official competition instance built by the organisers for MQ HackMac 2021, a public CTF, and released with its flag and official solution (fixture of a competition, not recorded production telemetry)
 
 
37
  Check: Exact string match of the submitted flag against the challenge yaml flags value; no tolerance
38
  Dependencies: the challenge yaml / FLAG.txt / writeup hold the flag and the official solution path (hidden from the solver); solver tools are free choices (for example an archive, image, pcap or office-format library), version fixed when the pool is built
39
  - `forensics/I_recognise_this/publish/flag`: MQ HackMac 2021 forensics/I_recognise_this: identify nested compression layers (XZ then ZIP then TAR) and peel them to reach the flag.
40
  Format: one file "flag" (896 B): XZ containing a ZIP (member "flag", 10,240 B, not encrypted) containing a TAR; three layers, none encrypted
41
  Origin: official competition instance built by the organisers for MQ HackMac 2021, a public CTF, and released with its flag and official solution (fixture of a competition, not recorded production telemetry)
 
 
42
  Check: Exact string match of the submitted flag against the challenge yaml flags value; no tolerance
43
  Dependencies: the challenge yaml / FLAG.txt / writeup hold the flag and the official solution path (hidden from the solver); solver tools are free choices (for example an archive, image, pcap or office-format library), version fixed when the pool is built
campaign_files/2026-10-s-security-detection/modsecurity-v3-regression/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** The file is marked self-contained upstream.
10
 
11
- **Answers:** No answer files are included. Reference result: expected_http_code_and_log. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Run each case with the ModSecurity v3 regression runner (or an agent-written rule set under libmodsecurity); HTTP code must match and debug_log regex must be found.
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -21,6 +19,4 @@ Files:
21
  - `test/test-cases/regression/operator-detectsqli.json`: ModSecurity v3 regression cases for @detectSQLi (10 cases: rules, request, expected HTTP code and debug log).
22
  Format: JSON list of test cases with inline SecRules
23
  Origin: test cases written by hand for the detection project's test suite (was constructed)
24
- Reference result: expected_http_code_and_log
25
- Reference source: expected.http_code (e.g. 403) and expected.debug_log regex per case
26
  Check: Run each case with the ModSecurity v3 regression runner (or an agent-written rule set under libmodsecurity); HTTP code must match and debug_log regex must be found.
 
8
 
9
  **Self-contained:** The file is marked self-contained upstream.
10
 
 
 
11
  **Tools to open and check:** Check: Run each case with the ModSecurity v3 regression runner (or an agent-written rule set under libmodsecurity); HTTP code must match and debug_log regex must be found.
12
 
13
  **Licence (SPDX):** Apache-2.0
 
19
  - `test/test-cases/regression/operator-detectsqli.json`: ModSecurity v3 regression cases for @detectSQLi (10 cases: rules, request, expected HTTP code and debug log).
20
  Format: JSON list of test cases with inline SecRules
21
  Origin: test cases written by hand for the detection project's test suite (was constructed)
 
 
22
  Check: Run each case with the ModSecurity v3 regression runner (or an agent-written rule set under libmodsecurity); HTTP code must match and debug_log regex must be found.
campaign_files/2026-10-sc-science-reference-data-2/modern-robotics-library-docstring-examples/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** The file is self-contained. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: homogeneous_transform_and_jacobian_matrices. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Four cases from the Example Input blocks: (1) FKinBody with home configuration M, body screw axes Blist and thetalist = [pi/2, 3, pi] (lines 572 to 580); (2) FKinSpace with the same M, space screw axes Slist and thetalist (lines 606 to 614); (3) JacobianBody and (4) JacobianSpace of a 4-joint chain with thetalist = [0.2, 1.1, 0.1, 1.2] (lines 641 to 646 and 673 to 678); the agent reports the 4x4 end-effector transform for (1) and (2) and the 6x4 Jacobian for (3) and (4), in the book's convention (screw axis = [omega; v], angular part first); each element passes when it differs from the printed Output value by at most half a unit of its last printed digit (C2 rounding; 8 decimals give 5e-9; integer-printed entries such as 0, 1, -5 are judged to 5e-9 as well, since the same arrays print other entries to 8 decimals).
14
 
15
  **Licence (SPDX):** MIT
@@ -21,7 +19,5 @@ Files:
21
  - `packages/Python/modern_robotics/core.py`: Forward kinematics and space and body Jacobians of open-chain robots by the product of exponentials (Modern Robotics core.py docstring examples).
22
  Format: Python module (78788 bytes); the row uses the docstring examples of FKinBody (lines 572 to 591), FKinSpace (lines 606 to 629), JacobianBody (lines 641 to 661) and JacobianSpace (lines 673 to 697)
23
  Origin: documented examples (Example Input / Output in the docstrings) of the Modern Robotics textbook library (Lynch and Park); the outputs are the library's own printed results
24
- Reference result: homogeneous_transform_and_jacobian_matrices
25
- Reference source: the Output blocks of the four docstrings at the pinned commit (lines 582 to 585, 616 to 619, 648 to 653, 680 to 685), printed to 8 decimals
26
  Check: four cases from the Example Input blocks: (1) FKinBody with home configuration M, body screw axes Blist and thetalist = [pi/2, 3, pi] (lines 572 to 580); (2) FKinSpace with the same M, space screw axes Slist and thetalist (lines 606 to 614); (3) JacobianBody and (4) JacobianSpace of a 4-joint chain with thetalist = [0.2, 1.1, 0.1, 1.2] (lines 641 to 646 and 673 to 678); the agent reports the 4x4 end-effector transform for (1) and (2) and the 6x4 Jacobian for (3) and (4), in the book's convention (screw axis = [omega; v], angular part first); each element passes when it differs from the printed Output value by at most half a unit of its last printed digit (C2 rounding; 8 decimals give 5e-9; integer-printed entries such as 0, 1, -5 are judged to 5e-9 as well, since the same arrays print other entries to 8 decimals)
27
  Dependencies: modern_robotics at cbd41fdf0bf75dbd5e986d832ae14226964ff0ac; numpy
 
8
 
9
  **Self-contained:** The file is self-contained. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Four cases from the Example Input blocks: (1) FKinBody with home configuration M, body screw axes Blist and thetalist = [pi/2, 3, pi] (lines 572 to 580); (2) FKinSpace with the same M, space screw axes Slist and thetalist (lines 606 to 614); (3) JacobianBody and (4) JacobianSpace of a 4-joint chain with thetalist = [0.2, 1.1, 0.1, 1.2] (lines 641 to 646 and 673 to 678); the agent reports the 4x4 end-effector transform for (1) and (2) and the 6x4 Jacobian for (3) and (4), in the book's convention (screw axis = [omega; v], angular part first); each element passes when it differs from the printed Output value by at most half a unit of its last printed digit (C2 rounding; 8 decimals give 5e-9; integer-printed entries such as 0, 1, -5 are judged to 5e-9 as well, since the same arrays print other entries to 8 decimals).
12
 
13
  **Licence (SPDX):** MIT
 
19
  - `packages/Python/modern_robotics/core.py`: Forward kinematics and space and body Jacobians of open-chain robots by the product of exponentials (Modern Robotics core.py docstring examples).
20
  Format: Python module (78788 bytes); the row uses the docstring examples of FKinBody (lines 572 to 591), FKinSpace (lines 606 to 629), JacobianBody (lines 641 to 661) and JacobianSpace (lines 673 to 697)
21
  Origin: documented examples (Example Input / Output in the docstrings) of the Modern Robotics textbook library (Lynch and Park); the outputs are the library's own printed results
 
 
22
  Check: four cases from the Example Input blocks: (1) FKinBody with home configuration M, body screw axes Blist and thetalist = [pi/2, 3, pi] (lines 572 to 580); (2) FKinSpace with the same M, space screw axes Slist and thetalist (lines 606 to 614); (3) JacobianBody and (4) JacobianSpace of a 4-joint chain with thetalist = [0.2, 1.1, 0.1, 1.2] (lines 641 to 646 and 673 to 678); the agent reports the 4x4 end-effector transform for (1) and (2) and the 6x4 Jacobian for (3) and (4), in the book's convention (screw axis = [omega; v], angular part first); each element passes when it differs from the printed Output value by at most half a unit of its last printed digit (C2 rounding; 8 decimals give 5e-9; integer-printed entries such as 0, 1, -5 are judged to 5e-9 as well, since the same arrays print other entries to 8 decimals)
23
  Dependencies: modern_robotics at cbd41fdf0bf75dbd5e986d832ae14226964ff0ac; numpy
campaign_files/2026-10-sc-science-reference-data-2/molgw-testsuite/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: reference_numbers_with_tolerance. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Give the agent h2_pbe_gw.in and ask for the listed quantities; each passes when the absolute difference from its reference is below the tolerance on its own line of the testsuite file (1e-7 for total energies, 1e-4 for orbital and excitation energies).
14
 
15
  **Licence (SPDX):** GPL-3.0-only
@@ -21,8 +19,6 @@ Files:
21
  - `tests/inputs/h2_pbe_gw.in`: MOLGW h2_pbe_gw: PBE then G0W0 of the H2 molecule with cc-pVDZ; compute SCF total energy, Kohn-Sham and GW HOMO and LUMO energies, RPA correlation and the lowest neutral excitation.
22
  Format: MOLGW namelist input (.in, text); reference list testsuite (text, comma-separated key, value, position, tolerance)
23
  Origin: regression test of the MOLGW test suite (tool_fixture): inputs written by the developers, reference values produced by MOLGW and committed
24
- Reference result: reference_numbers_with_tolerance
25
- Reference source: tests/inputs/testsuite lines 420-428: line 421 'SCF Total Energy , -1.0727602939 , 0 , 1.0e-7', 422 'gKS HOMO energy , -8.191536 , 0 , 1.0e-4', 423 'gKS LUMO energy , -2.674684', 424 'RPA correlation energy , -0.0838669374 , 0 , 1.0e-7', 425 'Lowest neutral excitation , 9.996355', 426 'Total EXX Energy , -1.0186704037', 427 'GW HOMO energy , -12.857222 , 0 , 1.0e-4', 428 'GW LUMO energy , 1.474561 , 0 , 1.0e-4' (energies in Hartree, orbital and excitation energies in eV)
26
  Check: give the agent h2_pbe_gw.in and ask for the listed quantities; each passes when the absolute difference from its reference is below the tolerance on its own line of the testsuite file (1e-7 for total energies, 1e-4 for orbital and excitation energies)
27
  Companions (not included): tests/inputs/testsuite; tests/run_testsuite.py
28
  Dependencies: MOLGW; MOLGW basis library basis/
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Give the agent h2_pbe_gw.in and ask for the listed quantities; each passes when the absolute difference from its reference is below the tolerance on its own line of the testsuite file (1e-7 for total energies, 1e-4 for orbital and excitation energies).
12
 
13
  **Licence (SPDX):** GPL-3.0-only
 
19
  - `tests/inputs/h2_pbe_gw.in`: MOLGW h2_pbe_gw: PBE then G0W0 of the H2 molecule with cc-pVDZ; compute SCF total energy, Kohn-Sham and GW HOMO and LUMO energies, RPA correlation and the lowest neutral excitation.
20
  Format: MOLGW namelist input (.in, text); reference list testsuite (text, comma-separated key, value, position, tolerance)
21
  Origin: regression test of the MOLGW test suite (tool_fixture): inputs written by the developers, reference values produced by MOLGW and committed
 
 
22
  Check: give the agent h2_pbe_gw.in and ask for the listed quantities; each passes when the absolute difference from its reference is below the tolerance on its own line of the testsuite file (1e-7 for total energies, 1e-4 for orbital and excitation energies)
23
  Companions (not included): tests/inputs/testsuite; tests/run_testsuite.py
24
  Dependencies: MOLGW; MOLGW basis library basis/
campaign_files/2026-10-sc-science-reference-data-2/mopac-tests/README.md CHANGED
@@ -8,8 +8,6 @@
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 2 of 2 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
11
- **Answers:** No answer files are included. Reference result: reference_output_with_tolerance. Where the reference result comes from and how a solution is checked is written per file in the notes.
12
-
13
  **Tools to open and check:** Check: Give the agent Locate_TS.mop with the two .arc geometries and ask for the transition-state structure and its heat of formation; as the test locate_ts does (tests/CMakeLists.txt line 61), compare the agent's .out against Locate_TS.out with compare_output.py: heats of formation must agree within HOF_ERROR 0.01 kcal/mol (tests/CMakeLists.txt line 18, asserted in compare_output.py line 244), strings exactly (line 234), eigenvector subspaces within EIGVEC_THRESHOLD (compare_output.py line 31, asserted at line 272); other numbers beyond NUMERIC_THRESHOLD (compare_output.py line 28) only warn (line 239); give the agent CISD.mop and ask for the configuration-interaction states and heats of formation; as the test key-CISD does (tests/keywords/CMakeLists.txt line 43), compare the agent's .out against CISD.out with compare_output.py: heats within HOF_ERROR 0.01 kcal/mol (tests/CMakeLists.txt line 18, asserted in compare_output.py line 244), strings such as state symmetries exactly (line 234); other numbers beyond NUMERIC_THRESHOLD (compare_output.py line 28) only warn (line 239).
14
 
15
  **Licence (SPDX):** Apache-2.0
@@ -21,16 +19,12 @@ Files:
21
  - `tests/Locate_TS.out`: MOPAC Locate_TS: locate the transition state between two cyclobutadiene geometries (LOCATE-TS) and report its heat of formation.
22
  Format: reference MOPAC output (.out, text, final heat of formation at line 523); input Locate_TS.mop with geo_ref and geo_dat pointing at Cyclobutadiene_2.arc and Cyclobutadiene_1.arc
23
  Origin: regression test of MOPAC (tool_fixture): input and end-point geometries written by the developers, reference output produced by MOPAC v23.1.2
24
- Reference result: reference_output_with_tolerance
25
- Reference source: tests/Locate_TS.out, the reference run_test.py compares with; key value line 523 FINAL HEAT OF FORMATION = 71.59418 KCAL/MOL
26
  Check: give the agent Locate_TS.mop with the two .arc geometries and ask for the transition-state structure and its heat of formation; as the test locate_ts does (tests/CMakeLists.txt line 61), compare the agent's .out against Locate_TS.out with compare_output.py: heats of formation must agree within HOF_ERROR 0.01 kcal/mol (tests/CMakeLists.txt line 18, asserted in compare_output.py line 244), strings exactly (line 234), eigenvector subspaces within EIGVEC_THRESHOLD (compare_output.py line 31, asserted at line 272); other numbers beyond NUMERIC_THRESHOLD (compare_output.py line 28) only warn (line 239)
27
  Companions (not included): LICENSE; tests/CMakeLists.txt; tests/Cyclobutadiene_1.arc; tests/Cyclobutadiene_2.arc; tests/Locate_TS.mop; tests/compare_output.py; tests/run_test.py
28
  Dependencies: MOPAC v23.1.2; Python 3 with numpy for run_test.py and compare_output.py
29
  - `tests/keywords/CISD.out`: MOPAC keyword test CISD: ethylene with a 4-orbital CISD configuration interaction (C.I.=4, MECI), report heats of formation and the excited-state table.
30
  Format: reference MOPAC output (.out, text; heats of formation at lines 78 and 401, state tables at lines 283 and 579); input CISD.mop (keywords 1SCF C.I.=4 MECI)
31
  Origin: keyword regression test of MOPAC (tool_fixture): input written by the developers, reference output produced by MOPAC
32
- Reference result: reference_output_with_tolerance
33
- Reference source: tests/keywords/CISD.out (heats of formation 7.84782 at line 78 and 7.85711 kcal/mol at line 401; state energies, spins and symmetries in the tables at lines 283 and 579)
34
  Check: give the agent CISD.mop and ask for the configuration-interaction states and heats of formation; as the test key-CISD does (tests/keywords/CMakeLists.txt line 43), compare the agent's .out against CISD.out with compare_output.py: heats within HOF_ERROR 0.01 kcal/mol (tests/CMakeLists.txt line 18, asserted in compare_output.py line 244), strings such as state symmetries exactly (line 234); other numbers beyond NUMERIC_THRESHOLD (compare_output.py line 28) only warn (line 239)
35
  Companions (not included): LICENSE; tests/CMakeLists.txt; tests/compare_output.py; tests/keywords/CISD.mop; tests/keywords/CMakeLists.txt; tests/run_test.py
36
  Dependencies: MOPAC v23.1.2; Python 3 with numpy for run_test.py and compare_output.py
 
8
 
9
  **Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 2 of 2 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
10
 
 
 
11
  **Tools to open and check:** Check: Give the agent Locate_TS.mop with the two .arc geometries and ask for the transition-state structure and its heat of formation; as the test locate_ts does (tests/CMakeLists.txt line 61), compare the agent's .out against Locate_TS.out with compare_output.py: heats of formation must agree within HOF_ERROR 0.01 kcal/mol (tests/CMakeLists.txt line 18, asserted in compare_output.py line 244), strings exactly (line 234), eigenvector subspaces within EIGVEC_THRESHOLD (compare_output.py line 31, asserted at line 272); other numbers beyond NUMERIC_THRESHOLD (compare_output.py line 28) only warn (line 239); give the agent CISD.mop and ask for the configuration-interaction states and heats of formation; as the test key-CISD does (tests/keywords/CMakeLists.txt line 43), compare the agent's .out against CISD.out with compare_output.py: heats within HOF_ERROR 0.01 kcal/mol (tests/CMakeLists.txt line 18, asserted in compare_output.py line 244), strings such as state symmetries exactly (line 234); other numbers beyond NUMERIC_THRESHOLD (compare_output.py line 28) only warn (line 239).
12
 
13
  **Licence (SPDX):** Apache-2.0
 
19
  - `tests/Locate_TS.out`: MOPAC Locate_TS: locate the transition state between two cyclobutadiene geometries (LOCATE-TS) and report its heat of formation.
20
  Format: reference MOPAC output (.out, text, final heat of formation at line 523); input Locate_TS.mop with geo_ref and geo_dat pointing at Cyclobutadiene_2.arc and Cyclobutadiene_1.arc
21
  Origin: regression test of MOPAC (tool_fixture): input and end-point geometries written by the developers, reference output produced by MOPAC v23.1.2
 
 
22
  Check: give the agent Locate_TS.mop with the two .arc geometries and ask for the transition-state structure and its heat of formation; as the test locate_ts does (tests/CMakeLists.txt line 61), compare the agent's .out against Locate_TS.out with compare_output.py: heats of formation must agree within HOF_ERROR 0.01 kcal/mol (tests/CMakeLists.txt line 18, asserted in compare_output.py line 244), strings exactly (line 234), eigenvector subspaces within EIGVEC_THRESHOLD (compare_output.py line 31, asserted at line 272); other numbers beyond NUMERIC_THRESHOLD (compare_output.py line 28) only warn (line 239)
23
  Companions (not included): LICENSE; tests/CMakeLists.txt; tests/Cyclobutadiene_1.arc; tests/Cyclobutadiene_2.arc; tests/Locate_TS.mop; tests/compare_output.py; tests/run_test.py
24
  Dependencies: MOPAC v23.1.2; Python 3 with numpy for run_test.py and compare_output.py
25
  - `tests/keywords/CISD.out`: MOPAC keyword test CISD: ethylene with a 4-orbital CISD configuration interaction (C.I.=4, MECI), report heats of formation and the excited-state table.
26
  Format: reference MOPAC output (.out, text; heats of formation at lines 78 and 401, state tables at lines 283 and 579); input CISD.mop (keywords 1SCF C.I.=4 MECI)
27
  Origin: keyword regression test of MOPAC (tool_fixture): input written by the developers, reference output produced by MOPAC
 
 
28
  Check: give the agent CISD.mop and ask for the configuration-interaction states and heats of formation; as the test key-CISD does (tests/keywords/CMakeLists.txt line 43), compare the agent's .out against CISD.out with compare_output.py: heats within HOF_ERROR 0.01 kcal/mol (tests/CMakeLists.txt line 18, asserted in compare_output.py line 244), strings such as state symmetries exactly (line 234); other numbers beyond NUMERIC_THRESHOLD (compare_output.py line 28) only warn (line 239)
29
  Companions (not included): LICENSE; tests/CMakeLists.txt; tests/compare_output.py; tests/keywords/CISD.mop; tests/keywords/CMakeLists.txt; tests/run_test.py
30
  Dependencies: MOPAC v23.1.2; Python 3 with numpy for run_test.py and compare_output.py