IoTProber commited on
Commit
b105aa6
·
verified ·
1 Parent(s): 5be8261

Addendum is dataset-mirror-only: code repository keeps paper code; fix dangling report references

Browse files
Files changed (1) hide show
  1. evaluation/REBUTTAL_ADDENDUM.md +8 -6
evaluation/REBUTTAL_ADDENDUM.md CHANGED
@@ -19,8 +19,11 @@ Everything below is reproducible from the pinned revision listed in §5.
19
 
20
  Commit `004c3f2` ("Remove evaluation directory") had deleted the tracked
21
  evaluation tooling, and several scripts referenced by the documentation were
22
- never committed. The following are now in the repository **and** the
23
- HuggingFace dataset (prefix `evaluation/`):
 
 
 
24
 
25
  | File | Role |
26
  |---|---|
@@ -28,9 +31,8 @@ HuggingFace dataset (prefix `evaluation/`):
28
  | `evaluation/drift/score_table10.py` | canonical, versioned Table-10 scorer (v1.1.0) with every matching rule explicit |
29
  | `evaluation/unseen/score_table8.py` | canonical, versioned Table-8 scorer (v1.1.0) with both vendor-matching conventions |
30
  | `evaluation/unseen/split_and_evaluate.py` | the loader/metric functions the Table-8 scorer imports (single source of truth) |
31
- | `evaluation/unseen/recompute_vendor_f1.py` | referenced by `VERIFICATION_v2.md` (also published) |
32
- | `evaluation/run_bench_100.py`, `evaluation/bench_100/{run_bench_v2,score_v2,finalize_report,dump_evidence,prewarm_local}.py` | bench harness referenced by `bench_100/IMPROVEMENT_ANALYSIS.md` (also published) |
33
- | `evaluation/bench_100/IMPROVEMENT_ANALYSIS.md`, `evaluation/drift/PACA_V2_REPORT.md`, `evaluation/unseen/VERIFICATION_v2.md` | the reports that reference the scripts above |
34
  | `evaluation/predict/{TYPE}_{type,vendor}_prediction.json` (22 files, all 11 types) | the decision-agent main predictions, incl. **ALARM/CONTROLLER** |
35
  | `evaluation/validation/type/predict/IoTProber/type_{ALARM,CONTROLLER}.csv`, `.../vendor/predict/IoTProber/vendor_{ALARM,CONTROLLER}.csv` | completes the 11-type prediction sets in the validation tree |
36
  | `evaluation/drift/CADE/SVM_validation/performance/{IoTProber,CADE}/val_metrics.json`, `val_classification_report.txt`, `{IoTProber,CADE}/detection_reasons.json`, `Ground/drift_reasons.json` | provenance of every Table-10 number (restored from `25f6a47`) |
@@ -286,7 +288,7 @@ python demo/unseen_detector_demo.py --skip-model # evidence + ranking only, any
286
  |---|---|
287
  | Version-of-record archive (code + data) | Zenodo DOI **10.5281/zenodo.23440077** (a new version of this record is being deposited with this addendum; the concept DOI stays citable) |
288
  | Dataset mirror, **pinned content revision** (every input file behind the numbers in §2–§3) | `IoTProber/raw_dataset @ f4dbbd53ceb2f3ad44eba79e81744dc78b1866cc` — https://huggingface.co/datasets/IoTProber/raw_dataset/tree/f4dbbd53ceb2f3ad44eba79e81744dc78b1866cc (this addendum and the manifest are published at the branch tip alongside it) |
289
- | Code mirror | https://github.com/IoTProber/IoTProber (commit that adds this addendum) |
290
  | File-level manifest | `evaluation/MANIFEST_f4dbbd53.txt` (next section) |
291
 
292
  HuggingFace commit hashes are immutable: every number in §2–§3 remains
 
19
 
20
  Commit `004c3f2` ("Remove evaluation directory") had deleted the tracked
21
  evaluation tooling, and several scripts referenced by the documentation were
22
+ never committed. Everything below is published in the HuggingFace dataset
23
+ (prefix `evaluation/`, via `hf download IoTProber/raw_dataset --repo-type
24
+ dataset --include "evaluation/*"`); the code repository intentionally keeps
25
+ only the paper code (`agent/`, `graph/`, `demo/`, `config/`), so none of it
26
+ is duplicated in git:
27
 
28
  | File | Role |
29
  |---|---|
 
31
  | `evaluation/drift/score_table10.py` | canonical, versioned Table-10 scorer (v1.1.0) with every matching rule explicit |
32
  | `evaluation/unseen/score_table8.py` | canonical, versioned Table-8 scorer (v1.1.0) with both vendor-matching conventions |
33
  | `evaluation/unseen/split_and_evaluate.py` | the loader/metric functions the Table-8 scorer imports (single source of truth) |
34
+ | `evaluation/unseen/recompute_vendor_f1.py` | unseen-adapter vendor-F1 recomputation |
35
+ | `evaluation/run_bench_100.py`, `evaluation/bench_100/{run_bench_v2,score_v2,finalize_report,dump_evidence,prewarm_local}.py` | bench-100 harness (per-IP decision records, dual-criterion scoring) |
 
36
  | `evaluation/predict/{TYPE}_{type,vendor}_prediction.json` (22 files, all 11 types) | the decision-agent main predictions, incl. **ALARM/CONTROLLER** |
37
  | `evaluation/validation/type/predict/IoTProber/type_{ALARM,CONTROLLER}.csv`, `.../vendor/predict/IoTProber/vendor_{ALARM,CONTROLLER}.csv` | completes the 11-type prediction sets in the validation tree |
38
  | `evaluation/drift/CADE/SVM_validation/performance/{IoTProber,CADE}/val_metrics.json`, `val_classification_report.txt`, `{IoTProber,CADE}/detection_reasons.json`, `Ground/drift_reasons.json` | provenance of every Table-10 number (restored from `25f6a47`) |
 
288
  |---|---|
289
  | Version-of-record archive (code + data) | Zenodo DOI **10.5281/zenodo.23440077** (a new version of this record is being deposited with this addendum; the concept DOI stays citable) |
290
  | Dataset mirror, **pinned content revision** (every input file behind the numbers in §2–§3) | `IoTProber/raw_dataset @ f4dbbd53ceb2f3ad44eba79e81744dc78b1866cc` — https://huggingface.co/datasets/IoTProber/raw_dataset/tree/f4dbbd53ceb2f3ad44eba79e81744dc78b1866cc (this addendum and the manifest are published at the branch tip alongside it) |
291
+ | Code mirror (paper code only: `agent/`, `graph/`, `demo/`, `config/`) | https://github.com/IoTProber/IoTProber |
292
  | File-level manifest | `evaluation/MANIFEST_f4dbbd53.txt` (next section) |
293
 
294
  HuggingFace commit hashes are immutable: every number in §2–§3 remains