diff --git a/.gitattributes b/.gitattributes
index bed0738c7eeb449bca98b5d2f33c89a1ee56349a..761b2ac7f224d343a6d0b3282035d083994f5d1b 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -58,3 +58,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+examples.html filter=lfs diff=lfs merge=lfs -text
+index.sqlite filter=lfs diff=lfs merge=lfs -text
diff --git a/README.md b/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..9f2ad8ed06597e1cf507db8ef36351d822cbbd3b
--- /dev/null
+++ b/README.md
@@ -0,0 +1,104 @@
+---
+license: other
+pretty_name: SFT Benchmark Trajectories
+task_categories:
+- text-generation
+tags:
+- agent-trajectories
+- sft
+- tool-use
+- software-engineering
+size_categories:
+- 100K<n<1M
+---
+
+# SFT Benchmark Trajectories
+
+A canonical, model-agnostic inventory of recorded expert demonstrations and agent completion trajectories. This is a **federated release**: the complete 155,278-trajectory metadata inventory, scores, lineage and source pointers are public, while canonical event payloads are hosted only where the source-specific redistribution audit supports it.
+
+## Release summary
+
+- Complete canonical inventory after exact-record deduplication: **155,278**
+- Canonical payloads hosted here: **136,373**
+- Metadata-only records recoverable from pinned upstream sources: **18,905**
+- Unique tasks: **67,610**
+- Successful / unsuccessful / unknown outcomes: **54,296 / 79,973 / 21,009**
+- Structurally eligible: **143,999**
+- Registry training-eligible before use-rights filtering: **110,208**
+- Native environment replays performed by this project: **0**
+- Scores rebuilt for this publication: **no** (persisted scores were copied)
+- Source code commit: `59e8576bfe3372b6a601a898c980b5db31eb1789`
+
+| Canonical source | Inventory | Public payload | Training permission | Transformed redistribution |
+|---|---:|---|---|---|
+| `agentsuite_tau2` | 8,340 | metadata_only | uncertain | uncertain |
+| `browsecomp_rollouts` | 5,063 | metadata_only | uncertain | uncertain |
+| `mind2web` | 1,009 | hosted | permitted_with_conditions | permitted |
+| `nvidia_open_swe_traces` | 51,626 | hosted | permitted | permitted |
+| `official_tau3` | 200 | hosted | permitted_with_conditions | permitted |
+| `swebench_multilingual_trajectories` | 129 | hosted | permitted_with_conditions | permitted_with_conditions |
+| `swebench_verified_trajectories` | 4,544 | hosted | permitted_with_conditions | permitted_with_conditions |
+| `swegym_openhands` | 491 | hosted | permitted | permitted |
+| `swegym_sampled` | 5,502 | metadata_only | uncertain | uncertain |
+| `swesmith_trajectories` | 24,100 | hosted | permitted_with_conditions | permitted_with_conditions |
+| `terminalbench_trajectories` | 34,263 | hosted | permitted_with_conditions | permitted_with_conditions |
+| `terminaltraj` | 20,000 | hosted | permitted | permitted_with_conditions |
+| `toolbench` | 11 | hosted | permitted_with_conditions | permitted_with_conditions |
+
+`permitted_with_conditions` is not the same as unrestricted use. Inspect `registry/permissions.json` for verification status, commercial-training status, conditions, evidence and pinned URLs. `THIRD_PARTY_NOTICES.md` contains attribution and notice guidance. This is an engineering audit, not legal advice.
+
+## Download and consume
+
+```bash
+pip install 'git+https://github.com/heisenger/nvidia_benchmark.git@59e8576bfe3372b6a601a898c980b5db31eb1789'
+pip install huggingface_hub
+huggingface-cli download Heisenger/sft-benchmark-trajectories   --repo-type dataset --local-dir ./sft-benchmark-trajectories
+```
+
+```python
+from agent_trajectories import CanonicalDatasetAdapter, DatasetQuery
+
+dataset = CanonicalDatasetAdapter("./sft-benchmark-trajectories")
+query = DatasetQuery.for_training(
+    splits=("train",),
+    content_availability=("hosted",),
+    training_permissions=("permitted", "permitted_with_conditions"),
+    training_verifications=("verified",),
+)
+for sample in dataset.iter_examples(query, view="full_trajectory"):
+    print(sample.trajectory_id, sample.events)
+    break
+```
+
+Normal iteration reads persisted canonical data and never contacts upstream sources. If a consumer is authorized to retrieve a metadata-only source, they can run the explicit pinned source adapter and combine the resulting canonical shard without changing the public snapshot:
+
+```python
+dataset = CanonicalDatasetAdapter(
+    "./sft-benchmark-trajectories",
+    content_overlays={"agentsuite_tau2": "./authorized/agentsuite_tau2"},
+)
+```
+
+The adapter rejects a payload query with a source-specific `MissingTrajectoryContentError` when only metadata is available. Metadata and permission queries remain available for the entire inventory.
+
+## Layout
+
+- `manifest.json`: portable relative paths, complete inventory counts, content availability and hashes.
+- `canonical/`: hosted canonical gzip shards.
+- `index.sqlite`: all 155,278 metadata/ranking/outcome records, including metadata-only sources.
+- `registry/permissions.json`: queryable multi-dimensional permission audit.
+- `rankings.jsonl.gz`, `task_splits.jsonl.gz`, outcome/eligibility indexes and `evaluation/`: persisted full-inventory derivations.
+- `examples.html`: human-readable trajectory examples.
+
+## Outcomes and scores
+
+Native rewards and success labels are preserved from upstream records; they were not regenerated. The project computed structural completeness, action/observation, known-outcome, verification-reliability and SFT-eligibility quality scores, plus deterministic within-task ranks. It did **not** replay upstream environments or compute model NLL. Historical labels are not scores for a new target model.
+
+## Licenses and limitations
+
+This aggregate uses source-specific terms and therefore has repository license `other`. Each record retains source ID, immutable revision and record locator. Payload hosting does not erase embedded repository, model-provider, website-content, attribution or benchmark-contamination obligations. Sources with unresolved transformed-redistribution rights remain metadata-only rather than being dropped or falsely labelled permissive.
+
+- Outcomes are predominantly source-reported, not independently replayed.
+- Some actions and observations are encoded in text rather than typed tool-call fields.
+- The inventory contains the imported subset at each pinned revision, not every upstream row.
+- `training_eligible` is a structural/source-quality gate; combine it with permission filters for a compliant subset.
diff --git a/THIRD_PARTY_NOTICES.md b/THIRD_PARTY_NOTICES.md
new file mode 100644
index 0000000000000000000000000000000000000000..557d4a1285edabb1af3e62f7a1b842b19416a3d4
--- /dev/null
+++ b/THIRD_PARTY_NOTICES.md
@@ -0,0 +1,33 @@
+# Third-party notices and attribution
+
+The canonical library is an aggregate. The package source code license does not
+replace source-dataset terms. Every record retains `source_id`, immutable
+revision and `record_locator`; the evidence and conditions used for publication
+are machine-readable in `data/registry/permissions.json`.
+
+Publicly hosted payloads use the following source terms:
+
+| Source | Terms and attribution |
+|---|---|
+| `swegym_openhands` | [SWE-Gym/OpenHands-SFT-Trajectories](https://huggingface.co/datasets/SWE-Gym/OpenHands-SFT-Trajectories/tree/4aaa5a4a4b5861f4799d2336908760c190ac3b17), MIT. Credit the SWE-Gym authors and retain the MIT notice. |
+| `nvidia_open_swe_traces` | [NVIDIA Open-SWE-Traces](https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586), CC-BY-4.0. Credit NVIDIA/Open-SWE-Traces, link CC-BY-4.0, identify canonical transformations, and retain the per-record repository license. |
+| `terminaltraj` | [m-a-p/TerminalTraj](https://huggingface.co/datasets/m-a-p/TerminalTraj/tree/5c1823f4a8b9ca0cf02c27d2db52c5b35b53a308), linked publisher project under Apache-2.0. Credit the TerminalTraj authors and honor embedded MIT/Apache/BSD repository terms. |
+| `official_tau3` | [Sierra Research tau2-bench](https://github.com/sierra-research/tau2-bench/tree/dd8757d4c8791dad7b3cdaf02a105a3d5f9d86bd), MIT. Copyright Sierra Research; retain the MIT notice. Training on benchmark tasks/rewards must be disclosed as evaluation overlap. |
+| `mind2web` | [OSU-NLP Mind2Web](https://huggingface.co/datasets/osunlp/Mind2Web/tree/17ece8eb89862368edc0cc806acee6fca5163474), CC-BY-4.0. Credit the Mind2Web authors, link CC-BY-4.0, identify transformations, and retain the research-purpose/harm disclaimer. |
+| `toolbench` | [OpenBMB ToolBench](https://github.com/OpenBMB/ToolBench/tree/d56fdd89faf8c91fa135090b212bb9057ee5cfc2), Apache-2.0 with research-and-education intended use. Credit OpenBMB/ToolBench and consider third-party API content rights. |
+| `swebench_verified_trajectories` | [tarsur385/swebench-verified-trajectories](https://huggingface.co/datasets/tarsur385/swebench-verified-trajectories/tree/773748a7c1222e8a642a7059821498e14293562a), MIT. Retain the dataset notice; task repositories retain their own terms. |
+| `swebench_multilingual_trajectories` | [AlienKevin/SWE-bench-Multilingual-trajectories](https://huggingface.co/datasets/AlienKevin/SWE-bench-Multilingual-trajectories/tree/7221c09c6cc0ed16d51c9cb9886a1ab484465670), MIT. Retain the dataset notice; task repositories retain their own terms. |
+| `terminalbench_trajectories` | [yoonholee/terminalbench-trajectories](https://huggingface.co/datasets/yoonholee/terminalbench-trajectories/tree/04e8940f5b6736a7ce8d22224fe2f2af74163ed2), Apache-2.0. Credit the dataset publisher and Terminal-Bench; consider the terms of model outputs represented in public leaderboard data. |
+| `swesmith_trajectories` | [SWE-bench/SWE-smith-trajectories](https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories/tree/08e109b4a59eaeebf80e4675cd125d42e7ac99a4), MIT. Credit SWE-smith and retain generated task repository terms. |
+
+License references: [MIT](https://opensource.org/license/mit),
+[Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0), and
+[CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/legalcode).
+
+The raw payloads for `agentsuite_tau2`, `swegym_sampled`, and
+`browsecomp_rollouts` are metadata-only in the public aggregate because the
+separate trajectory repositories do not establish sufficiently clear
+transformed-redistribution terms. Their absence is not a claim that access,
+research use, or every form of training is prohibited.
+
+This file is an engineering attribution aid, not legal advice.
diff --git a/canonical/mind2web/default--train--0000.jsonl.gz b/canonical/mind2web/default--train--0000.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..cd88221101357c6577ce796da8c37d706742ba9b
--- /dev/null
+++ b/canonical/mind2web/default--train--0000.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:654357077e6249cd030e516a74bc3ac7551e26553809581bbebb56a4f87e4c37
+size 60960835
diff --git a/canonical/mind2web/default--train--0001.jsonl.gz b/canonical/mind2web/default--train--0001.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..7444a939de1faa66214de08c86955a63526bf03d
--- /dev/null
+++ b/canonical/mind2web/default--train--0001.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea62d59ed99fa6632c59bb62cf339054fe72e717c38d36109526414bb9ce2f60
+size 61759916
diff --git a/canonical/mind2web/default--train--0002.jsonl.gz b/canonical/mind2web/default--train--0002.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..29de6595623849d7a0b57b59f816f8780b1097cf
--- /dev/null
+++ b/canonical/mind2web/default--train--0002.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77d3d9b0ef99cf5ff8300c7e0681847780752ce60e2c2e0887e247b491b48e22
+size 61958505
diff --git a/canonical/mind2web/default--train--0003.jsonl.gz b/canonical/mind2web/default--train--0003.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..ec3f9badef9c42c5531d27395056884aa7e9325a
--- /dev/null
+++ b/canonical/mind2web/default--train--0003.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3807d46087383372f43ca0e3ed95d1b14e9ceca917940c652d136eca5157d7c
+size 64515683
diff --git a/canonical/mind2web/default--train--0005.jsonl.gz b/canonical/mind2web/default--train--0005.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..54b015d43e043ac08ed6b7996366cdcf16eef644
--- /dev/null
+++ b/canonical/mind2web/default--train--0005.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd075e514c9006f2781cec84a0bbb816e1bc6ccdf22217885925e2ab5d02df8b
+size 59608293
diff --git a/canonical/mind2web/default--train--0007.jsonl.gz b/canonical/mind2web/default--train--0007.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..14b48886f96aa490d81efba3c8344982a6ba87d6
--- /dev/null
+++ b/canonical/mind2web/default--train--0007.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa20a9fa7022a87f523148839fbbccae8576eecd011eb8a8b7a12be2472c8e5e
+size 116176682
diff --git a/canonical/nvidia_open_swe_traces/v1.2--openhands--0002.jsonl.gz b/canonical/nvidia_open_swe_traces/v1.2--openhands--0002.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..af6f5c3855fbb016b3bd2699ce3f559a99b596da
--- /dev/null
+++ b/canonical/nvidia_open_swe_traces/v1.2--openhands--0002.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3067df96c93975319e290aa0b5f7b4be0ceecb6275f761cea33db9b6b3c54ffd
+size 305310880
diff --git a/canonical/nvidia_open_swe_traces/v1.2--openhands--0003.jsonl.gz b/canonical/nvidia_open_swe_traces/v1.2--openhands--0003.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..d571a6065c2bfae12c17ac6a954b7736d15e12a9
--- /dev/null
+++ b/canonical/nvidia_open_swe_traces/v1.2--openhands--0003.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b3324ce204a000df83ad9e207b4318af57ed19d36c11d7cbbae7bf5db90b453
+size 290056585
diff --git a/canonical/nvidia_open_swe_traces/v1.2--openhands--0005.jsonl.gz b/canonical/nvidia_open_swe_traces/v1.2--openhands--0005.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..9c1036003cbecb3015b68de3e626cce0e12215f2
--- /dev/null
+++ b/canonical/nvidia_open_swe_traces/v1.2--openhands--0005.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92b92e2aba80e79cbb07e9ba071db946fe4aa885deef4aefa9d934ae3bff193e
+size 5161644
diff --git a/canonical/nvidia_open_swe_traces/v1.2--openhands--0007.jsonl.gz b/canonical/nvidia_open_swe_traces/v1.2--openhands--0007.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..bfb210cc6215b6caab34396f084bce3e8c328874
--- /dev/null
+++ b/canonical/nvidia_open_swe_traces/v1.2--openhands--0007.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af1f7ed30b9037df06ebf7e7f0d146913ad813511e065e8f735a520b646b74bf
+size 371128455
diff --git a/canonical/nvidia_open_swe_traces/v1.2--openhands--0010.jsonl.gz b/canonical/nvidia_open_swe_traces/v1.2--openhands--0010.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..f44ea7a7a4ec8b8e284613bf3104d2fa6e932555
--- /dev/null
+++ b/canonical/nvidia_open_swe_traces/v1.2--openhands--0010.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2eff407239d58e2d9d525aa70402dcf01501ef3d3007d6407a69765c9c4936b5
+size 370224921
diff --git a/canonical/nvidia_open_swe_traces/v1.2--openhands--0011.jsonl.gz b/canonical/nvidia_open_swe_traces/v1.2--openhands--0011.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..44ddd18b37ca3de0edd9e869b95b17d335e98d36
--- /dev/null
+++ b/canonical/nvidia_open_swe_traces/v1.2--openhands--0011.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a68e3249212dbf0661db6e7d7a5b9afc6a2a4bf0172d0f32e9ad0cbf272abbc
+size 83589148
diff --git a/canonical/nvidia_open_swe_traces/v1.2--openhands--0012.jsonl.gz b/canonical/nvidia_open_swe_traces/v1.2--openhands--0012.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..f57df52a710a25780c484d183e819fa0e16f6a4c
--- /dev/null
+++ b/canonical/nvidia_open_swe_traces/v1.2--openhands--0012.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d54f3aebffe6bfdd1889488db25bf851210f000d6f6ca2fa2bebf83155fea40a
+size 243965767
diff --git a/canonical/nvidia_open_swe_traces/v1.2--openhands--0014.jsonl.gz b/canonical/nvidia_open_swe_traces/v1.2--openhands--0014.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..c73001027999983de33473a0d7c3aba658a3c952
--- /dev/null
+++ b/canonical/nvidia_open_swe_traces/v1.2--openhands--0014.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e964d62766afd638a8526c5e44fca9581e2cea0d3ec0ffd0a14beeefc1e16c6c
+size 249402952
diff --git a/canonical/nvidia_open_swe_traces/v1.2--openhands--0016.jsonl.gz b/canonical/nvidia_open_swe_traces/v1.2--openhands--0016.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..b1cec58993da5140b461837a846376a7a31d3401
--- /dev/null
+++ b/canonical/nvidia_open_swe_traces/v1.2--openhands--0016.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95b32f5071fe69c1949e66623413b9e02970973ea501040336955b56d6de0b16
+size 242546761
diff --git a/canonical/nvidia_open_swe_traces/v1.2--openhands--0017.jsonl.gz b/canonical/nvidia_open_swe_traces/v1.2--openhands--0017.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..a42f1d1e5c52bebcd456c7e4b67baf4ff14558bb
--- /dev/null
+++ b/canonical/nvidia_open_swe_traces/v1.2--openhands--0017.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c17ad85f0ef74b315314fc6929e3efe7875f5d88f319208bc07fbbbc621c296
+size 74960935
diff --git a/canonical/nvidia_open_swe_traces/v1.2--openhands--0018.jsonl.gz b/canonical/nvidia_open_swe_traces/v1.2--openhands--0018.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..be28e06c7ac71b56f1ab5b7db670610db60ac1c4
--- /dev/null
+++ b/canonical/nvidia_open_swe_traces/v1.2--openhands--0018.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:331b6e4d1611cc71c643b7a1af41744218d25aa5d6be33675f3ca2bcc20ab183
+size 373422381
diff --git a/canonical/swesmith_trajectories/default--tool--0000.jsonl.gz b/canonical/swesmith_trajectories/default--tool--0000.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..d8172d9db59a57ec2ef94c91f2d5e3e6cf26341a
--- /dev/null
+++ b/canonical/swesmith_trajectories/default--tool--0000.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:909cd6654ba6b253bf83874168fcfef57eb7ae700d94d39d4e10da15e2d4df22
+size 208736286
diff --git a/canonical/swesmith_trajectories/default--tool--0001.jsonl.gz b/canonical/swesmith_trajectories/default--tool--0001.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..7c9e9a67988080ffcacf7ff0b2ee81ca31a4c5b4
--- /dev/null
+++ b/canonical/swesmith_trajectories/default--tool--0001.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8efdc46253958b9bfdba85da004f1de435709881773472d627a21372f197ba78
+size 128599828
diff --git a/canonical/swesmith_trajectories/default--tool--0002.jsonl.gz b/canonical/swesmith_trajectories/default--tool--0002.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..c54aa1d444436ee98537bbcf8a8d23e7a024b366
--- /dev/null
+++ b/canonical/swesmith_trajectories/default--tool--0002.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80d7a0105bee1b9cde168524b9cd702005786014464a3d5adcb40dcb7a0603ff
+size 115440687
diff --git a/canonical/swesmith_trajectories/default--tool--0003.jsonl.gz b/canonical/swesmith_trajectories/default--tool--0003.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..cbcf801d9a685b0ed53d4911fd68db1b00f4914d
--- /dev/null
+++ b/canonical/swesmith_trajectories/default--tool--0003.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48df7f6939fd5619d1376c3e6909cfb71d88db56c11ce0511989b51620e44a9c
+size 147332694
diff --git a/canonical/swesmith_trajectories/default--tool--0004.jsonl.gz b/canonical/swesmith_trajectories/default--tool--0004.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..a3548e7e4fffa3220756f01e9660b24fe72a55c7
--- /dev/null
+++ b/canonical/swesmith_trajectories/default--tool--0004.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da2007c58f68ec2de3736cf7dbb6b1356e1d1657bfe8ef5f03ecb77b03a7e227
+size 93517927
diff --git a/canonical/swesmith_trajectories/default--tool--0005.jsonl.gz b/canonical/swesmith_trajectories/default--tool--0005.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..b08027330e6ae743f538458412da42c0ed468f94
--- /dev/null
+++ b/canonical/swesmith_trajectories/default--tool--0005.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:97e66d5f2c7e74a53d2d56528d9972ad3ad455ba7472900f8ba474e7bfecba9a
+size 147229136
diff --git a/canonical/swesmith_trajectories/default--tool--0006.jsonl.gz b/canonical/swesmith_trajectories/default--tool--0006.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..a60274737f25d0e71fbdf50554f1c4905c6edbcb
--- /dev/null
+++ b/canonical/swesmith_trajectories/default--tool--0006.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fcf23d8b7b0f907ce65e99fd25983b479e099e69506eaeef4c6794f6a663b1cd
+size 162574019
diff --git a/canonical/swesmith_trajectories/default--tool--0007.jsonl.gz b/canonical/swesmith_trajectories/default--tool--0007.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..7f0c2d26ed07ad8371f90b9d3171685670384f81
--- /dev/null
+++ b/canonical/swesmith_trajectories/default--tool--0007.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff5c4100eb4fa6610b91012d93d071989a7cb33510f9b5e273e19e41a05d7b39
+size 141721185
diff --git a/docs/dataset_adapter.md b/docs/dataset_adapter.md
new file mode 100644
index 0000000000000000000000000000000000000000..47b90d06be45db04b8beceb8d59f240b24f4442e
--- /dev/null
+++ b/docs/dataset_adapter.md
@@ -0,0 +1,352 @@
+# Consumer-facing dataset adapter
+
+`agent_trajectories` has two intentionally separate adapter layers:
+
+1. **Source adapters** acquire a pinned external release and normalize its
+   benchmark-specific rows into the canonical `Trajectory` schema. They run
+   during ingestion or an explicit refresh.
+2. **The consumer adapter** reads only the persisted canonical library and its
+   read-only metadata index. Training and evaluation jobs never import
+   benchmark parsers, contact upstream hosts, or rewrite dataset files.
+
+The stable entry point is `CanonicalDatasetAdapter`. Its API version is `1.1`;
+the package version is `0.3.0`.
+
+```python
+from agent_trajectories import CanonicalDatasetAdapter
+
+dataset = CanonicalDatasetAdapter("data/library")
+print(dataset.available_benchmarks())
+print(dataset.available_sources())
+```
+
+The adapter validates `manifest.json`, requires the persisted `index.sqlite`
+and available canonical shards, and opens SQLite in read-only mode. A federated
+release can contain full-inventory metadata while marking some payloads as
+metadata-only. It keeps only lightweight selected index rows in memory.
+Unsampled event payloads are parsed one trajectory at a time.
+
+## Query contract
+
+`DatasetQuery` fields use OR semantics within a field and AND semantics across
+fields:
+
+- `benchmarks`: stable families such as `swe-bench`, `swe-gym`, `swe-smith`,
+  `terminal-bench`, `tau2-bench`, `tau3-bench`, `browsecomp`, `mind2web`, and
+  `toolbench`;
+- `source_ids`: exact canonical sources such as `swegym_sampled` or
+  `agentsuite_tau2`;
+- `source_versions`: exact immutable source revisions;
+- `agent_models` and `agent_scaffolds`;
+- `trajectory_ids`, `task_keys`, and compatible `environment_keys`;
+- task-level library `splits`: `train`, `validation`, `test`, or
+  `protected_evaluation`;
+- `quality_tiers` and persisted task/environment `quality_ranks`;
+- dynamic `content_availability`: `hosted`, `overlay`, or `metadata_only`;
+- audited `research_permissions`, `training_permissions`,
+  `training_verifications`, `commercial_training_permissions`, and
+  transformed `redistribution_permissions`;
+- `structurally_eligible` and `training_eligible` gates;
+- `selection`: `all_attempts`, `all_successful`, or `best_verified`.
+
+The concise string aliases `all`, `successful`, and `best` are also accepted,
+as are `full`/`full-trajectory` and `turn`/`assistant-turn` view names. Enum
+members are preferred when static typing is available.
+
+`DatasetQuery.for_training(...)` enables both structural and training-source
+gates. This quality gate is not a legal-permission decision; combine it with
+the permission fields for the downstream project's policy. `best_verified`
+returns at most one successful source-evaluator or
+independently reproduced trajectory per selected task. Apply an
+`environment_keys` filter when alternatives from different environment
+versions must not compete. Quality rank itself is always calculated within a
+compatible task/environment group over the full canonical population; source
+filtering never silently rewrites the stored rank.
+
+## Full-trajectory SFT
+
+```python
+from agent_trajectories import (
+    CanonicalDatasetAdapter,
+    DatasetQuery,
+    TrainingView,
+    TrajectorySelection,
+)
+
+dataset = CanonicalDatasetAdapter("data/library")
+query = DatasetQuery.for_training(
+    benchmarks=("swe-gym", "swe-smith"),
+    splits=("train",),
+    selection=TrajectorySelection.ALL_SUCCESSFUL,
+)
+
+for example in dataset.iter_examples(query, view=TrainingView.FULL_TRAJECTORY):
+    row = example.to_dict()
+    model_input = row["model_input"]
+    loss_mask = row["loss_mask"]
+    lineage = row["metadata"]["lineage"]
+    # Apply a model-specific chat template/tokenizer downstream.
+```
+
+Each full example has the complete event sequence, assistant-only
+`target_event_ids`, and an aligned `loss_mask`. Outcomes, verification evidence
+and ranks are in metadata, never in `model_input`.
+
+## Assistant-turn SFT
+
+```python
+query = DatasetQuery.for_training(
+    source_ids=("agentsuite_tau2",),
+    splits=("train",),
+)
+
+for turn in dataset.iter_examples(query, view=TrainingView.ASSISTANT_TURN):
+    row = turn.to_dict()
+    assert len(row["context"]) == row["target_index"]
+    assert row["target_event_id"] == row["metadata"]["lineage"]["turn_id"]
+    model_input = row["model_input"]       # causal prefix only
+    supervised_target = row["supervised_target"]
+```
+
+Full and turn views derive from the same canonical record, retain the same task
+split and source lineage, and have equivalent assistant targets. Turn context
+contains no target event, future observation, later action, outcome, or answer.
+
+## Best trajectory and multiple completions
+
+Select the best verified trajectory for every selected task:
+
+```python
+best = DatasetQuery(
+    benchmarks=("swe-bench",),
+    splits=("protected_evaluation",),
+    selection=TrajectorySelection.BEST_VERIFIED,
+)
+for record in dataset.iter_records(best):
+    print(record.metadata.task_key, record.metadata.quality_rank)
+```
+
+Request one exact canonical trajectory:
+
+```python
+record = dataset.get_trajectory("trajectory:...")
+print(record.trajectory.events)
+print(record.lineage())
+```
+
+Collapsed duplicate IDs are resolved to their retained canonical trajectory;
+`contributing_trajectory_ids` preserves every original ID.
+
+Enumerate all completions for a task, in quality order:
+
+```python
+first = next(dataset.metadata(DatasetQuery(source_ids=("agentsuite_tau2",))))
+attempts = dataset.alternatives(first.task_key)
+for attempt in attempts:
+    print(
+        attempt.metadata.trajectory_id,
+        attempt.metadata.quality_rank,
+        attempt.metadata.outcome_category,
+        attempt.metadata.agent_model,
+    )
+```
+
+`environment_key=` can restrict alternatives to one compatible environment.
+
+## Deterministic sampling
+
+```python
+from agent_trajectories import SamplingConfig, SamplingStrategy
+
+population = DatasetQuery.for_training(splits=("train",))
+
+# Uniform trajectory sampling.
+random_sample = SamplingConfig(count=1_000, seed=42)
+
+# Uniform task sampling; the best matching attempt represents each drawn task.
+task_sample = SamplingConfig(
+    count=1_000, seed=42, strategy=SamplingStrategy.TASK
+)
+
+# Round-robin sources, then sample within each source.
+balanced_sample = SamplingConfig(
+    count=1_000, seed=42, strategy=SamplingStrategy.SOURCE_BALANCED
+)
+
+for record in dataset.iter_records(population, sampling=balanced_sample):
+    print(record.metadata.trajectory_id, record.metadata.source_id)
+```
+
+Set `with_replacement=True` for repeated draws. The same persisted library,
+query, seed and sampling configuration produce the same ordered IDs.
+Task-level sampling prevents tasks with many recorded attempts from dominating.
+Source-balanced sampling distributes draws round-robin across non-empty sources
+and deterministically redistributes unused capacity when a source is exhausted.
+
+## Streaming and batching
+
+```python
+for batch in dataset.iter_batches(
+    8,
+    DatasetQuery.for_training(splits=("train",)),
+    view=TrainingView.ASSISTANT_TURN,
+):
+    rows = [example.to_dict() for example in batch]
+    print(len(rows), rows[0]["sample_id"])
+```
+
+These are bounded Python batches, not tensors. Canonical storage stays
+framework-independent and downstream code retains control of templates,
+tokenization, truncation, packing and collation.
+
+PyTorch is optional and imported lazily:
+
+```python
+torch_dataset = dataset.as_torch_iterable(
+    DatasetQuery.for_training(splits=("train",)),
+    view=TrainingView.ASSISTANT_TURN,
+)
+
+# Standard torch DataLoader worker processes are sharded by item index.
+from torch.utils.data import DataLoader
+loader = DataLoader(torch_dataset, batch_size=None, num_workers=2)
+```
+
+The core package has no ML-framework dependency. Install the optional extra
+only when needed: `pip install 'agent-trajectory-datasets[torch]'`.
+
+## Evaluation over all trajectories
+
+Task splits are assigned once by canonical `task_key`, then joined into every
+source and both derived views. A task cannot appear in more than one of
+`train`, `validation`, `test`, and `protected_evaluation`.
+
+```python
+evaluation = DatasetQuery(
+    benchmarks=("tau2-bench",),
+    splits=("validation", "test", "protected_evaluation"),
+    selection=TrajectorySelection.ALL_ATTEMPTS,
+)
+
+for record in dataset.iter_records(evaluation):
+    print(record.metadata.task_key, record.trajectory.outcome.success)
+```
+
+For all references, source verdicts and verification evidence associated with
+one task:
+
+```python
+task = next(dataset.metadata(evaluation)).task_key
+bundle = dataset.evaluation_bundle(task, query=evaluation)
+for reference in bundle.references:
+    print(
+        reference.trajectory_id,
+        reference.outcome.success,
+        reference.verification.method,
+        reference.verification.evidence,
+        reference.supervised_targets,
+    )
+```
+
+Historical source outcomes are evaluation references, not scores for a new
+model. Native environment success still requires running the corresponding
+benchmark evaluator. Unknown outcomes remain unknown rather than being counted
+as failures.
+
+## Metadata and lineage
+
+Every `DatasetRecord` carries `TrajectoryMetadata`, including:
+
+- canonical task, trajectory and environment IDs;
+- benchmark family, source ID, immutable revision, upstream record locator and
+  canonical/source split;
+- agent model and scaffold;
+- quality tier, numeric score and within-task quality rank;
+- structural/training eligibility;
+- current payload availability and the complete source permission audit,
+  including verification state, commercial-training status and conditions;
+- outcome category, reward, verification method/strength and independent-replay
+  status;
+- every duplicate source trajectory ID collapsed into the canonical record.
+
+`record.lineage()` adds the complete `SourceRef`, producer provenance, source
+outcome and full verification evidence. Derived examples store that object at
+`example.metadata["lineage"]`; assistant-turn examples also store `turn_id` and
+full examples store all supervised `turn_ids`.
+
+## Federated payloads and authorized overlays
+
+`available_sources()` enumerates the complete indexed inventory;
+`content_sources()` lists sources whose event payload is currently readable.
+Inspect one permission record without loading trajectories:
+
+```python
+rights = dataset.permissions("agentsuite_tau2")
+print(rights.dimensions["model_training"])
+print(dataset.content_availability("agentsuite_tau2"))
+```
+
+When a project is separately authorized to retrieve a metadata-only source,
+run its explicit pinned source adapter and provide the resulting canonical
+file or directory as an overlay:
+
+```python
+dataset = CanonicalDatasetAdapter(
+    "./sft-benchmark-trajectories",
+    content_overlays={
+        "agentsuite_tau2": "./authorized/canonical/agentsuite_tau2",
+    },
+)
+
+query = DatasetQuery(
+    source_ids=("agentsuite_tau2",),
+    content_availability=("overlay",),
+)
+for record in dataset.iter_records(query):
+    ...
+```
+
+The overlay is read-only and is not copied or uploaded. Canonical IDs join it
+to the hosted full-inventory index. Selecting records whose payload is still
+absent raises `MissingTrajectoryContentError` with the missing source IDs;
+metadata queries never silently discard them.
+
+## Initial ingestion and explicit refresh
+
+Normal consumption performs no downloads. Initial ingestion is a separate,
+explicit operation. For a hosted release:
+
+```bash
+.venv/bin/python scripts/import_hf_release.py SOURCE_ID DATASET_ID \
+  --revision IMMUTABLE_SHA \
+  --output data/full/canonical/OUTPUT \
+  --cache data/cache/full/OUTPUT \
+  --config CONFIG --split SPLIT
+
+.venv/bin/python scripts/build_complete_library.py \
+  --input-dir data/full/canonical/OUTPUT \
+  --output data/library
+```
+
+The complete multi-source build command and source-specific revisions are in
+the root README and `docs/trajectory_sources.md`.
+
+A refresh is never implicit. Choose and audit a new upstream immutable SHA,
+write it to a new canonical output directory, then deliberately rebuild the
+library with the old or new directory selected:
+
+```bash
+.venv/bin/python scripts/import_hf_release.py SOURCE_ID DATASET_ID \
+  --revision NEW_IMMUTABLE_SHA \
+  --output data/full/canonical/OUTPUT_NEW_SHA \
+  --cache data/cache/full/OUTPUT_NEW_SHA \
+  --config CONFIG --split SPLIT
+
+# Review release-manifest.json and rejects before replacing the old input.
+.venv/bin/python scripts/build_complete_library.py \
+  --input-dir data/full/canonical/OUTPUT_NEW_SHA \
+  --output data/library-refreshed
+```
+
+This separation makes dataset changes reviewable and keeps training runs pinned
+to the manifest and shard hashes they actually consumed.
diff --git a/docs/licensing_and_federation.md b/docs/licensing_and_federation.md
new file mode 100644
index 0000000000000000000000000000000000000000..1255ccff17251e23e107077846b3475566072840
--- /dev/null
+++ b/docs/licensing_and_federation.md
@@ -0,0 +1,68 @@
+# Permission audit and federated content
+
+Permission is not a single Boolean. The registry records, for every canonical
+source, separate findings for public access, local research use, model
+training, commercial training, redistribution of original data, and
+redistribution of transformed canonical trajectories. Each finding has both a
+permission value and a verification value, plus source-specific conditions and
+evidence URLs.
+
+The audit vocabulary is:
+
+- permission: `permitted`, `permitted_with_conditions`, `uncertain`, or
+  `restricted`;
+- verification: `verified`, `inferred`, or `uncertain`;
+- public payload: `hosted` or `metadata_only`.
+
+The complete local inventory remains 155,278 trajectories. Publication does
+not change scores, outcomes, ranks, splits, lineage or canonical hashes.
+Instead, the public Hugging Face release contains the complete metadata index
+and the payloads supported by the audit. Metadata-only sources retain pinned
+upstream URLs and canonical shard hashes.
+
+The initial public payload decision is:
+
+- hosted: `swegym_openhands`, `nvidia_open_swe_traces`, `terminaltraj`,
+  `official_tau3`, `mind2web`, `toolbench`,
+  `swebench_verified_trajectories`,
+  `swebench_multilingual_trajectories`, `terminalbench_trajectories`, and
+  `swesmith_trajectories`;
+- metadata-only: `agentsuite_tau2`, `swegym_sampled`, and
+  `browsecomp_rollouts`.
+
+This is not a blanket declaration that hosted data is unrestricted or that
+metadata-only data cannot be used. For example, Mind2Web combines CC-BY-4.0
+metadata with a research-purpose statement; tau benchmark training creates
+evaluation contamination that must be disclosed; and software-engineering
+records can embed code governed by per-repository licenses.
+
+Downstream projects can select a policy directly:
+
+```python
+from agent_trajectories import CanonicalDatasetAdapter, DatasetQuery
+
+dataset = CanonicalDatasetAdapter("./sft-benchmark-trajectories")
+strict_research = DatasetQuery.for_training(
+    content_availability=("hosted",),
+    training_permissions=("permitted", "permitted_with_conditions"),
+    training_verifications=("verified",),
+    commercial_training_permissions=("permitted", "permitted_with_conditions"),
+)
+for metadata in dataset.metadata(strict_research):
+    print(metadata.source_id, metadata.source_permissions.to_dict())
+```
+
+An authorized user can explicitly ingest a metadata-only source at its pinned
+revision and overlay the canonical shards:
+
+```python
+dataset = CanonicalDatasetAdapter(
+    "./sft-benchmark-trajectories",
+    content_overlays={"agentsuite_tau2": "./canonical/agentsuite_tau2"},
+)
+```
+
+The overlay is local and is never uploaded by the adapter. Canonical trajectory
+IDs are matched against the persisted full-inventory index. Queries that select
+missing content fail with `MissingTrajectoryContentError` and the relevant
+source IDs instead of silently omitting records.
diff --git a/docs/nemotron_sft_demo.md b/docs/nemotron_sft_demo.md
new file mode 100644
index 0000000000000000000000000000000000000000..ab2d67ab92a63a60022aa1fb9246fdaae650593f
--- /dev/null
+++ b/docs/nemotron_sft_demo.md
@@ -0,0 +1,48 @@
+# Nemotron SFT tokenization demonstration
+
+`scripts/demo_nemotron_sft.py` reads a real hosted canonical record through the
+consumer adapter, derives full-trajectory and causal assistant-turn views,
+applies the pinned native Nemotron chat template, builds assistant-only token
+loss masks, and pads both samples into one framework-neutral model batch.
+
+The demonstrated tokenizer is
+`nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16` at immutable revision
+`a9904d24bcc1d289a1950fa9d2b978c47cf903b9`. No model weights or training are
+required.
+
+```bash
+pip install '.[tokenization]'
+HF_HOME=.cache/huggingface PYTHONPATH=src python \
+  scripts/demo_nemotron_sft.py \
+  --library ./sft-benchmark-trajectories \
+  --output examples/nemotron_sft_demo.json
+```
+
+The real successful tau trajectory
+`trajectory:f9cf8588a314fa9804d57402` has 13 events and six supervised assistant
+targets. Results from the pinned tokenizer are:
+
+| View | Total tokens | Supervised assistant tokens | Ignored tokens |
+|---|---:|---:|---:|
+| Full trajectory | 3,174 | 696 | 2,478 |
+| One causal assistant turn | 1,877 | 60 | 1,817 |
+
+For the turn view, 80 tokens belonging to prior assistant turns are explicitly
+masked out; only the selected target at event index 5 contributes to loss. The
+causal prefix is 1,817 tokens and does not include the target or any future
+observation.
+
+The resulting batch has fields `input_ids`, `attention_mask`, and `labels`,
+shape `[2, 3174]`, right padding, and label ignore index `-100`. Every context
+and padding label is `-100`; exactly 696 and 60 label positions are supervised
+in the two rows. Its deterministic JSON representation has SHA-256
+`d680ae7e4ea605b3db96370de6760d86ed227411dac901d74a511f72c3346ae8`.
+
+Nemotron's pinned template does not ship Transformers `{% generation %}` spans.
+The script makes no semantic template changes: it wraps the template's native
+assistant-rendering branch with those spans so Transformers can return exact
+token-aligned assistant masks. It disables historical-thinking truncation to
+keep causal prefixes byte-for-byte stable. The script checks mask/label
+alignment, target-prefix stability, non-empty supervision, and ignored padding
+before emitting the report. See `examples/nemotron_sft_demo.json` for the event
+preview, lineage, token previews, lengths, mask checks, and batch checksum.
diff --git a/duplicate_provenance.jsonl.gz b/duplicate_provenance.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..1de959f1d8c59787a2f228e43bc79511c89dee84
--- /dev/null
+++ b/duplicate_provenance.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea4cd16eac95b5fa34bc0413feb4a078347a214bcb2af15cfd8a03ea42482625
+size 76971
diff --git a/evaluation/model_imitation.json b/evaluation/model_imitation.json
new file mode 100644
index 0000000000000000000000000000000000000000..8a2f73eed4cdc4114c218a4425e839095b092618
--- /dev/null
+++ b/evaluation/model_imitation.json
@@ -0,0 +1,5 @@
+{
+  "implementation": "The canonical full and assistant-turn exports, exact structured-action scorer, and a pinned Nemotron tokenizer/mask/batch demonstration are available; NLL requires model weights and inference.",
+  "reason": "No trained language model weights or inference accelerator were used. Tokenizer-derived lengths and assistant-only masks are reported separately and are not NLL.",
+  "status": "not-run"
+}
diff --git a/evaluation/per_trajectory.jsonl.gz b/evaluation/per_trajectory.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..cf4887c9e7f79d63bfc2e83b2e47ea554d66b5a8
--- /dev/null
+++ b/evaluation/per_trajectory.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0dfc8d4e0517311818a245c86b83086557740cd6fce9cafc1acc2cde717042fe
+size 15275530
diff --git a/evaluation/run_config.json b/evaluation/run_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0cf0858ef3f6182e33af4598ade34bd6106f684c
--- /dev/null
+++ b/evaluation/run_config.json
@@ -0,0 +1,77 @@
+{
+  "evaluation_errors": [],
+  "evaluator_versions": {
+    "canonical_validation": "1.0",
+    "exact_record_deduplication": "1.0",
+    "quality_ranking": "1.0",
+    "structured_action_exact_match": "1.0",
+    "task_split": "trajectory-library-v1",
+    "trajectory_quality": "1.0"
+  },
+  "finalization_runtime_seconds": 25.363,
+  "inputs": [
+    "metadata-only:agentsuite_tau2",
+    "metadata-only:browsecomp_rollouts",
+    "metadata-only:browsecomp_rollouts",
+    "canonical/mind2web/default--train--0000.jsonl.gz",
+    "canonical/mind2web/default--train--0001.jsonl.gz",
+    "canonical/mind2web/default--train--0002.jsonl.gz",
+    "canonical/mind2web/default--train--0003.jsonl.gz",
+    "canonical/mind2web/default--train--0004.jsonl.gz",
+    "canonical/mind2web/default--train--0005.jsonl.gz",
+    "canonical/mind2web/default--train--0006.jsonl.gz",
+    "canonical/mind2web/default--train--0007.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0000.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0001.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0002.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0003.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0004.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0005.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0006.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0007.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0008.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0009.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0010.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0011.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0012.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0013.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0014.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0015.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0016.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0017.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0018.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0019.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0020.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0021.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0022.jsonl.gz",
+    "canonical/nvidia_open_swe_traces/v1.2--openhands--0023.jsonl.gz",
+    "canonical/official_tau3+toolbench/official_tau3_toolbench.jsonl.gz",
+    "canonical/swebench_multilingual_trajectories/default--Multi_SWE_agent_32B_Rust--0000.jsonl.gz",
+    "canonical/swebench_multilingual_trajectories/default--Qwen2.5_Coder_32B_Instruct--0001.jsonl.gz",
+    "canonical/swebench_multilingual_trajectories/default--SWE_agent_LM_32B--0002.jsonl.gz",
+    "canonical/swebench_verified_trajectories/all.jsonl.gz",
+    "canonical/swegym_openhands/default--train.success.oss--0000.jsonl.gz",
+    "metadata-only:swegym_sampled",
+    "metadata-only:swegym_sampled",
+    "metadata-only:swegym_sampled",
+    "canonical/swesmith_trajectories/default--tool--0000.jsonl.gz",
+    "canonical/swesmith_trajectories/default--tool--0001.jsonl.gz",
+    "canonical/swesmith_trajectories/default--tool--0002.jsonl.gz",
+    "canonical/swesmith_trajectories/default--tool--0003.jsonl.gz",
+    "canonical/swesmith_trajectories/default--tool--0004.jsonl.gz",
+    "canonical/swesmith_trajectories/default--tool--0005.jsonl.gz",
+    "canonical/swesmith_trajectories/default--tool--0006.jsonl.gz",
+    "canonical/swesmith_trajectories/default--tool--0007.jsonl.gz",
+    "canonical/terminalbench_trajectories/default--train--0000.jsonl.gz",
+    "canonical/terminalbench_trajectories/default--train--0001.jsonl.gz",
+    "canonical/terminaltraj/default--train--0000.jsonl.gz",
+    "canonical/terminaltraj/default--train--0001.jsonl.gz"
+  ],
+  "model_imitation_status": "not-run",
+  "native_environment_replays": 0,
+  "platform": "Linux-6.18.44-x86_64-with-glibc2.41",
+  "python": "3.12.14 (main, Aug 25 2026, 14:00:49) [Clang 22.1.3 ]",
+  "registry": "registry/sources.json",
+  "runtime_seconds": 953.287,
+  "schema_version": 1
+}
diff --git a/evaluation/summary.json b/evaluation/summary.json
new file mode 100644
index 0000000000000000000000000000000000000000..d1773c8d5b2ae5ed521b074415374b1f7693778b
--- /dev/null
+++ b/evaluation/summary.json
@@ -0,0 +1,2901 @@
+{
+  "by_agent_scaffold": {
+    "AgentSuite tau2 benchmark rollout": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 4965,
+      "input_token_samples": 0,
+      "mean_actions": 17.801798561151077,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 45.433093525179856,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5750.385491606715,
+      "output_token_samples": 0,
+      "structurally_eligible": 8337,
+      "success": 3375,
+      "success_rate_known_outcomes": 0.40467625899280574,
+      "training_eligible": 8337,
+      "trajectories": 8340,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "Mind2Web human annotator": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 0,
+      "input_token_samples": 0,
+      "mean_actions": 7.705649157581764,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 16.411298315163528,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 256.4281466798811,
+      "output_token_samples": 0,
+      "structurally_eligible": 1009,
+      "success": 0,
+      "success_rate_known_outcomes": null,
+      "training_eligible": 1009,
+      "trajectories": 1009,
+      "unique_tasks": 1009,
+      "unknown": 1009
+    },
+    "NVIDIA Open-SWE trace generator": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 26883,
+      "input_token_samples": 0,
+      "mean_actions": 58.35106341765777,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 124.11902917134778,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 4650.015089296091,
+      "output_token_samples": 0,
+      "structurally_eligible": 51626,
+      "success": 24743,
+      "success_rate_known_outcomes": 0.4792740092201604,
+      "training_eligible": 51626,
+      "trajectories": 51626,
+      "unique_tasks": 33444,
+      "unknown": 0
+    },
+    "OpenHands": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 5011,
+      "input_token_samples": 0,
+      "mean_actions": 20.382406397673574,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 42.287168302435475,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 7133.47528171574,
+      "output_token_samples": 0,
+      "structurally_eligible": 5502,
+      "success": 491,
+      "success_rate_known_outcomes": 0.08924027626317703,
+      "training_eligible": 5502,
+      "trajectories": 5502,
+      "unique_tasks": 2438,
+      "unknown": 0
+    },
+    "SWE-agent": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 14673,
+      "input_token_samples": 0,
+      "mean_actions": 31.585477178423236,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 64.17095435684648,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 17664.133236514524,
+      "output_token_samples": 0,
+      "structurally_eligible": 24100,
+      "success": 9427,
+      "success_rate_known_outcomes": 0.39116182572614105,
+      "training_eligible": 24100,
+      "trajectories": 24100,
+      "unique_tasks": 10637,
+      "unknown": 0
+    },
+    "SWE-agent-compatible scaffold": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 121,
+      "input_token_samples": 0,
+      "mean_actions": 47.37984496124031,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 96.75968992248062,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 36004.64341085271,
+      "output_token_samples": 0,
+      "structurally_eligible": 110,
+      "success": 8,
+      "success_rate_known_outcomes": 0.06201550387596899,
+      "training_eligible": 0,
+      "trajectories": 129,
+      "unique_tasks": 43,
+      "unknown": 0
+    },
+    "ToolBench answer generation": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 4,
+      "input_token_samples": 0,
+      "mean_actions": 3.8181818181818183,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 9.090909090909092,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 834.3636363636364,
+      "output_token_samples": 0,
+      "structurally_eligible": 0,
+      "success": 7,
+      "success_rate_known_outcomes": 0.6363636363636364,
+      "training_eligible": 0,
+      "trajectories": 11,
+      "unique_tasks": 11,
+      "unknown": 0
+    },
+    "claude-code": {
+      "cost_samples": 2517,
+      "duration_samples": 2517,
+      "failure": 1418,
+      "input_token_samples": 1204,
+      "mean_actions": 54.15454906634883,
+      "mean_cost_source_native_units": 66.30842242871526,
+      "mean_duration_seconds": 545.8310086178694,
+      "mean_events": 89.69288835915772,
+      "mean_input_tokens": 2248965.17358804,
+      "mean_output_tokens": 2596.156146179402,
+      "mean_supervised_assistant_characters": 9204.064759634486,
+      "output_token_samples": 1204,
+      "structurally_eligible": 21,
+      "success": 1099,
+      "success_rate_known_outcomes": 0.43663090981326974,
+      "training_eligible": 0,
+      "trajectories": 2517,
+      "unique_tasks": 88,
+      "unknown": 0
+    },
+    "codex": {
+      "cost_samples": 1385,
+      "duration_samples": 1385,
+      "failure": 908,
+      "input_token_samples": 1383,
+      "mean_actions": 17.857761732851987,
+      "mean_cost_source_native_units": 9.600910829986692,
+      "mean_duration_seconds": 377.06245802180433,
+      "mean_events": 34.25270758122743,
+      "mean_input_tokens": 365547.114967462,
+      "mean_output_tokens": 9098.387563268258,
+      "mean_supervised_assistant_characters": 4321.786281588447,
+      "output_token_samples": 1383,
+      "structurally_eligible": 23,
+      "success": 477,
+      "success_rate_known_outcomes": 0.3444043321299639,
+      "training_eligible": 0,
+      "trajectories": 1385,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "deepagent-harbor": {
+      "cost_samples": 380,
+      "duration_samples": 380,
+      "failure": 98,
+      "input_token_samples": 0,
+      "mean_actions": 29.263157894736842,
+      "mean_cost_source_native_units": 0.0,
+      "mean_duration_seconds": 705.5991674925151,
+      "mean_events": 59.747368421052634,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 4741.5447368421055,
+      "output_token_samples": 0,
+      "structurally_eligible": 337,
+      "success": 282,
+      "success_rate_known_outcomes": 0.7421052631578947,
+      "training_eligible": 0,
+      "trajectories": 380,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gemini-cli": {
+      "cost_samples": 755,
+      "duration_samples": 755,
+      "failure": 604,
+      "input_token_samples": 755,
+      "mean_actions": 17.34834437086093,
+      "mean_cost_source_native_units": 27.581561065845143,
+      "mean_duration_seconds": 509.88918244822923,
+      "mean_events": 34.70728476821192,
+      "mean_input_tokens": 613653.8317880795,
+      "mean_output_tokens": 22646.129801324503,
+      "mean_supervised_assistant_characters": 2940.1867549668873,
+      "output_token_samples": 755,
+      "structurally_eligible": 568,
+      "success": 151,
+      "success_rate_known_outcomes": 0.2,
+      "training_eligible": 0,
+      "trajectories": 755,
+      "unique_tasks": 88,
+      "unknown": 0
+    },
+    "judy": {
+      "cost_samples": 157,
+      "duration_samples": 157,
+      "failure": 37,
+      "input_token_samples": 0,
+      "mean_actions": 32.611464968152866,
+      "mean_cost_source_native_units": 0.0,
+      "mean_duration_seconds": 792.0475469212623,
+      "mean_events": 65.35031847133757,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 4949.356687898089,
+      "output_token_samples": 0,
+      "structurally_eligible": 14,
+      "success": 120,
+      "success_rate_known_outcomes": 0.7643312101910829,
+      "training_eligible": 0,
+      "trajectories": 157,
+      "unique_tasks": 83,
+      "unknown": 0
+    },
+    "mini-swe-agent": {
+      "cost_samples": 6437,
+      "duration_samples": 6437,
+      "failure": 4936,
+      "input_token_samples": 6421,
+      "mean_actions": 21.314276837035887,
+      "mean_cost_source_native_units": 29.442191969641353,
+      "mean_duration_seconds": 491.84549342628515,
+      "mean_events": 43.85194966599347,
+      "mean_input_tokens": 743809.1583865442,
+      "mean_output_tokens": 19476.359601308206,
+      "mean_supervised_assistant_characters": 11188.48003728445,
+      "output_token_samples": 6421,
+      "structurally_eligible": 6339,
+      "success": 1501,
+      "success_rate_known_outcomes": 0.23318315985707627,
+      "training_eligible": 0,
+      "trajectories": 6437,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "minisweagent.run.benchmarks.swebench.ProgressTrackingAgent": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 1152,
+      "input_token_samples": 0,
+      "mean_actions": 47.387984154929576,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 97.00748239436619,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 12330.321082746479,
+      "output_token_samples": 0,
+      "structurally_eligible": 4532,
+      "success": 3392,
+      "success_rate_known_outcomes": 0.7464788732394366,
+      "training_eligible": 0,
+      "trajectories": 4544,
+      "unique_tasks": 500,
+      "unknown": 0
+    },
+    "openhands": {
+      "cost_samples": 6033,
+      "duration_samples": 6033,
+      "failure": 4296,
+      "input_token_samples": 5593,
+      "mean_actions": 82.31791811702304,
+      "mean_cost_source_native_units": 161.66761875649664,
+      "mean_duration_seconds": 885.4345594482321,
+      "mean_events": 144.69766285430134,
+      "mean_input_tokens": 1835762.452887538,
+      "mean_output_tokens": 27097.81995351332,
+      "mean_supervised_assistant_characters": 18699.04243328361,
+      "output_token_samples": 5593,
+      "structurally_eligible": 284,
+      "success": 1737,
+      "success_rate_known_outcomes": 0.28791645947289907,
+      "training_eligible": 0,
+      "trajectories": 6033,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "rl-rag BrowseComp research scaffold": {
+      "cost_samples": 0,
+      "duration_samples": 5063,
+      "failure": 3852,
+      "input_token_samples": 0,
+      "mean_actions": 27.13786292711831,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": 108.07624142800492,
+      "mean_events": 54.28303377444203,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 15385.260912502468,
+      "output_token_samples": 0,
+      "structurally_eligible": 5063,
+      "success": 1211,
+      "success_rate_known_outcomes": 0.23918625320955955,
+      "training_eligible": 0,
+      "trajectories": 5063,
+      "unique_tasks": 1266,
+      "unknown": 0
+    },
+    "tau benchmark agent and user simulator": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 100,
+      "input_token_samples": 0,
+      "mean_actions": 14.5,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 30.0,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 4221.82,
+      "output_token_samples": 0,
+      "structurally_eligible": 0,
+      "success": 100,
+      "success_rate_known_outcomes": 0.5,
+      "training_eligible": 0,
+      "trajectories": 200,
+      "unique_tasks": 50,
+      "unknown": 0
+    },
+    "terminus-2": {
+      "cost_samples": 15921,
+      "duration_samples": 15921,
+      "failure": 10743,
+      "input_token_samples": 14581,
+      "mean_actions": 25.161610451604798,
+      "mean_cost_source_native_units": 42.898824807674124,
+      "mean_duration_seconds": 743.9324663575665,
+      "mean_events": 52.535456315558065,
+      "mean_input_tokens": 612642.0770866196,
+      "mean_output_tokens": 16685.474110143336,
+      "mean_supervised_assistant_characters": 17296.68670309654,
+      "output_token_samples": 14581,
+      "structurally_eligible": 15822,
+      "success": 5178,
+      "success_rate_known_outcomes": 0.32523082720934615,
+      "training_eligible": 0,
+      "trajectories": 15921,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "terminus-3-3": {
+      "cost_samples": 678,
+      "duration_samples": 678,
+      "failure": 172,
+      "input_token_samples": 678,
+      "mean_actions": 29.125368731563423,
+      "mean_cost_source_native_units": 74.12365559350073,
+      "mean_duration_seconds": 664.1559937288628,
+      "mean_events": 59.250737463126846,
+      "mean_input_tokens": 990659.5044247787,
+      "mean_output_tokens": 25339.938053097347,
+      "mean_supervised_assistant_characters": 5243.02802359882,
+      "output_token_samples": 678,
+      "structurally_eligible": 678,
+      "success": 506,
+      "success_rate_known_outcomes": 0.7463126843657817,
+      "training_eligible": 0,
+      "trajectories": 678,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "unknown": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 0,
+      "input_token_samples": 0,
+      "mean_actions": 16.883314625933338,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 33.79059099116685,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 10559.201747108487,
+      "output_token_samples": 0,
+      "structurally_eligible": 19634,
+      "success": 491,
+      "success_rate_known_outcomes": 1.0,
+      "training_eligible": 19634,
+      "trajectories": 20491,
+      "unique_tasks": 18080,
+      "unknown": 20000
+    }
+  },
+  "by_model": {
+    "ChatGPT (repository-described; record has no exact version)": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 4,
+      "input_token_samples": 0,
+      "mean_actions": 3.8181818181818183,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 9.090909090909092,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 834.3636363636364,
+      "output_token_samples": 0,
+      "structurally_eligible": 0,
+      "success": 7,
+      "success_rate_known_outcomes": 0.6363636363636364,
+      "training_eligible": 0,
+      "trajectories": 11,
+      "unique_tasks": 11,
+      "unknown": 0
+    },
+    "Multi-SWE-agent-32B-Rust": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 42,
+      "input_token_samples": 0,
+      "mean_actions": 70.65116279069767,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 143.30232558139534,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 26940.906976744187,
+      "output_token_samples": 0,
+      "structurally_eligible": 41,
+      "success": 1,
+      "success_rate_known_outcomes": 0.023255813953488372,
+      "training_eligible": 0,
+      "trajectories": 43,
+      "unique_tasks": 43,
+      "unknown": 0
+    },
+    "Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8@together_ai": {
+      "cost_samples": 1798,
+      "duration_samples": 1798,
+      "failure": 1348,
+      "input_token_samples": 1794,
+      "mean_actions": 72.29365962180201,
+      "mean_cost_source_native_units": 255.6923547450772,
+      "mean_duration_seconds": 648.0904410365426,
+      "mean_events": 119.13681868743048,
+      "mean_input_tokens": 1270502.5462653288,
+      "mean_output_tokens": 10809.755852842809,
+      "mean_supervised_assistant_characters": 27958.512791991103,
+      "output_token_samples": 1794,
+      "structurally_eligible": 928,
+      "success": 450,
+      "success_rate_known_outcomes": 0.25027808676307006,
+      "training_eligible": 0,
+      "trajectories": 1798,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "Qwen2.5-Coder-32B-Instruct": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 43,
+      "input_token_samples": 0,
+      "mean_actions": 29.74418604651163,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 61.48837209302326,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 57204.72093023256,
+      "output_token_samples": 0,
+      "structurally_eligible": 27,
+      "success": 0,
+      "success_rate_known_outcomes": 0.0,
+      "training_eligible": 0,
+      "trajectories": 43,
+      "unique_tasks": 43,
+      "unknown": 0
+    },
+    "Qwen3.8-27B": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 13666,
+      "input_token_samples": 0,
+      "mean_actions": 62.70772295792804,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 134.19887495605298,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 6554.125161139107,
+      "output_token_samples": 0,
+      "structurally_eligible": 25599,
+      "success": 11933,
+      "success_rate_known_outcomes": 0.4661510215242783,
+      "training_eligible": 25599,
+      "trajectories": 25599,
+      "unique_tasks": 25599,
+      "unknown": 0
+    },
+    "Qwen3.8-Flash-Next": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 13217,
+      "input_token_samples": 0,
+      "mean_actions": 54.06604679755638,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 114.20494102278403,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 2777.217082260729,
+      "output_token_samples": 0,
+      "structurally_eligible": 26027,
+      "success": 12810,
+      "success_rate_known_outcomes": 0.49218119644984054,
+      "training_eligible": 26027,
+      "trajectories": 26027,
+      "unique_tasks": 26027,
+      "unknown": 0
+    },
+    "SWE-agent-LM-32B": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 36,
+      "input_token_samples": 0,
+      "mean_actions": 41.74418604651163,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 85.48837209302326,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 23868.302325581397,
+      "output_token_samples": 0,
+      "structurally_eligible": 42,
+      "success": 7,
+      "success_rate_known_outcomes": 0.16279069767441862,
+      "training_eligible": 0,
+      "trajectories": 43,
+      "unique_tasks": 43,
+      "unknown": 0
+    },
+    "accounts/fireworks/models/glm-4p6@fireworks_ai": {
+      "cost_samples": 904,
+      "duration_samples": 904,
+      "failure": 680,
+      "input_token_samples": 904,
+      "mean_actions": 41.66814159292036,
+      "mean_cost_source_native_units": 11.691257349520395,
+      "mean_duration_seconds": 476.20702301295455,
+      "mean_events": 85.45575221238938,
+      "mean_input_tokens": 1122957.5884955751,
+      "mean_output_tokens": 11417.679203539823,
+      "mean_supervised_assistant_characters": 30965.702433628318,
+      "output_token_samples": 904,
+      "structurally_eligible": 904,
+      "success": 224,
+      "success_rate_known_outcomes": 0.24778761061946902,
+      "training_eligible": 0,
+      "trajectories": 904,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "accounts/fireworks/models/minimax-m2@fireworks_ai": {
+      "cost_samples": 898,
+      "duration_samples": 898,
+      "failure": 622,
+      "input_token_samples": 894,
+      "mean_actions": 35.66369710467706,
+      "mean_cost_source_native_units": 32.10366033558591,
+      "mean_duration_seconds": 931.9153214180656,
+      "mean_events": 75.79955456570156,
+      "mean_input_tokens": 1006714.9149888143,
+      "mean_output_tokens": 17048.77852348993,
+      "mean_supervised_assistant_characters": 24994.038975501113,
+      "output_token_samples": 894,
+      "structurally_eligible": 888,
+      "success": 276,
+      "success_rate_known_outcomes": 0.30734966592427615,
+      "training_eligible": 0,
+      "trajectories": 898,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "anthropic/claude-4-opus-thinking-off": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 134,
+      "input_token_samples": 0,
+      "mean_actions": 15.327338129496402,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 41.37410071942446,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5759.190647482014,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 144,
+      "success_rate_known_outcomes": 0.5179856115107914,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "anthropic/claude-4-opus-thinking-on-10k": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 135,
+      "input_token_samples": 0,
+      "mean_actions": 14.964028776978417,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 40.18345323741007,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5505.651079136691,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 143,
+      "success_rate_known_outcomes": 0.5143884892086331,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "anthropic/claude-4-sonnet-thinking-off": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 150,
+      "input_token_samples": 0,
+      "mean_actions": 14.158273381294965,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 36.330935251798564,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5403.838129496403,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 128,
+      "success_rate_known_outcomes": 0.460431654676259,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "anthropic/claude-4-sonnet-thinking-on-10k": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 133,
+      "input_token_samples": 0,
+      "mean_actions": 13.611510791366907,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 34.989208633093526,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5727.251798561151,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 145,
+      "success_rate_known_outcomes": 0.5215827338129496,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "anthropic/claude-4.5-sonnet-thinking-off": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 141,
+      "input_token_samples": 0,
+      "mean_actions": 15.31294964028777,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 42.37410071942446,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5430.76618705036,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 137,
+      "success_rate_known_outcomes": 0.49280575539568344,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "anthropic/claude-4.5-sonnet-thinking-on-10k": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 126,
+      "input_token_samples": 0,
+      "mean_actions": 13.507194244604317,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 39.55755395683453,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5573.884892086331,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 152,
+      "success_rate_known_outcomes": 0.5467625899280576,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "claude-3-5-sonnet-20241022": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 3567,
+      "input_token_samples": 0,
+      "mean_actions": 21.48913043478261,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 43.97826086956522,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 8249.628260869566,
+      "output_token_samples": 0,
+      "structurally_eligible": 5980,
+      "success": 2413,
+      "success_rate_known_outcomes": 0.4035117056856187,
+      "training_eligible": 5980,
+      "trajectories": 5980,
+      "unique_tasks": 2194,
+      "unknown": 0
+    },
+    "claude-3-7-sonnet-20250219": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 10909,
+      "input_token_samples": 0,
+      "mean_actions": 34.72726765280491,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 70.45453530560982,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 20883.35668434273,
+      "output_token_samples": 0,
+      "structurally_eligible": 17715,
+      "success": 7006,
+      "success_rate_known_outcomes": 0.3910689366452693,
+      "training_eligible": 17715,
+      "trajectories": 17915,
+      "unique_tasks": 9193,
+      "unknown": 0
+    },
+    "claude-4.5-haiku-high": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 167,
+      "input_token_samples": 0,
+      "mean_actions": 66.118,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 133.272,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 21335.944,
+      "output_token_samples": 0,
+      "structurally_eligible": 500,
+      "success": 333,
+      "success_rate_known_outcomes": 0.666,
+      "training_eligible": 0,
+      "trajectories": 500,
+      "unique_tasks": 500,
+      "unknown": 0
+    },
+    "claude-4.5-opus-high": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 116,
+      "input_token_samples": 0,
+      "mean_actions": 32.892,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 72.41,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 11651.378,
+      "output_token_samples": 0,
+      "structurally_eligible": 500,
+      "success": 384,
+      "success_rate_known_outcomes": 0.768,
+      "training_eligible": 0,
+      "trajectories": 500,
+      "unique_tasks": 500,
+      "unknown": 0
+    },
+    "claude-haiku-4-5-20251001@anthropic": {
+      "cost_samples": 1656,
+      "duration_samples": 1656,
+      "failure": 1230,
+      "input_token_samples": 1655,
+      "mean_actions": 80.83393719806763,
+      "mean_cost_source_native_units": 207.72488678684053,
+      "mean_duration_seconds": 743.0986861180568,
+      "mean_events": 202.96497584541063,
+      "mean_input_tokens": 3116007.3438066463,
+      "mean_output_tokens": 18089.96193353474,
+      "mean_supervised_assistant_characters": 18744.347222222223,
+      "output_token_samples": 1655,
+      "structurally_eligible": 965,
+      "success": 426,
+      "success_rate_known_outcomes": 0.2572463768115942,
+      "training_eligible": 0,
+      "trajectories": 1656,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "claude-opus-4-1-20250805@anthropic": {
+      "cost_samples": 1712,
+      "duration_samples": 1712,
+      "failure": 1072,
+      "input_token_samples": 1710,
+      "mean_actions": 47.4696261682243,
+      "mean_cost_source_native_units": 255.39316846193555,
+      "mean_duration_seconds": 620.2949460087535,
+      "mean_events": 80.42815420560747,
+      "mean_input_tokens": 1029506.9678362573,
+      "mean_output_tokens": 9701.605847953217,
+      "mean_supervised_assistant_characters": 12596.474299065421,
+      "output_token_samples": 1710,
+      "structurally_eligible": 872,
+      "success": 640,
+      "success_rate_known_outcomes": 0.37383177570093457,
+      "training_eligible": 0,
+      "trajectories": 1712,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "claude-opus-4-5-20251101@anthropic": {
+      "cost_samples": 1307,
+      "duration_samples": 1307,
+      "failure": 594,
+      "input_token_samples": 445,
+      "mean_actions": 41.711553175210405,
+      "mean_cost_source_native_units": 19.595637352118167,
+      "mean_duration_seconds": 664.1525678860841,
+      "mean_events": 78.16296863045142,
+      "mean_input_tokens": 943597.584269663,
+      "mean_output_tokens": 14243.811235955056,
+      "mean_supervised_assistant_characters": 13112.324407039021,
+      "output_token_samples": 445,
+      "structurally_eligible": 601,
+      "success": 713,
+      "success_rate_known_outcomes": 0.5455241009946442,
+      "training_eligible": 0,
+      "trajectories": 1307,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "claude-opus-4-6@anthropic": {
+      "cost_samples": 1127,
+      "duration_samples": 1127,
+      "failure": 403,
+      "input_token_samples": 692,
+      "mean_actions": 29.628216503992903,
+      "mean_cost_source_native_units": 35.3747195851623,
+      "mean_duration_seconds": 667.7565543810244,
+      "mean_events": 54.184560780834076,
+      "mean_input_tokens": 543390.0708092486,
+      "mean_output_tokens": 16201.849710982659,
+      "mean_supervised_assistant_characters": 8093.924578527063,
+      "output_token_samples": 692,
+      "structurally_eligible": 689,
+      "success": 724,
+      "success_rate_known_outcomes": 0.642413487133984,
+      "training_eligible": 0,
+      "trajectories": 1127,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "claude-opus-4.6": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 114,
+      "input_token_samples": 0,
+      "mean_actions": 28.932,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 60.844,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 6521.846,
+      "output_token_samples": 0,
+      "structurally_eligible": 500,
+      "success": 386,
+      "success_rate_known_outcomes": 0.772,
+      "training_eligible": 0,
+      "trajectories": 500,
+      "unique_tasks": 500,
+      "unknown": 0
+    },
+    "claude-opus-4.6@anthropic": {
+      "cost_samples": 157,
+      "duration_samples": 157,
+      "failure": 37,
+      "input_token_samples": 0,
+      "mean_actions": 32.611464968152866,
+      "mean_cost_source_native_units": 0.0,
+      "mean_duration_seconds": 792.0475469212623,
+      "mean_events": 65.35031847133757,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 4949.356687898089,
+      "output_token_samples": 0,
+      "structurally_eligible": 14,
+      "success": 120,
+      "success_rate_known_outcomes": 0.7643312101910829,
+      "training_eligible": 0,
+      "trajectories": 157,
+      "unique_tasks": 83,
+      "unknown": 0
+    },
+    "claude-sonnet-4-5-20250929@anthropic": {
+      "cost_samples": 1696,
+      "duration_samples": 1696,
+      "failure": 958,
+      "input_token_samples": 1692,
+      "mean_actions": 56.375,
+      "mean_cost_source_native_units": 66.14636548995128,
+      "mean_duration_seconds": 662.6581952335699,
+      "mean_events": 96.6745283018868,
+      "mean_input_tokens": 1397487.5827423169,
+      "mean_output_tokens": 11408.179669030733,
+      "mean_supervised_assistant_characters": 16556.10554245283,
+      "output_token_samples": 1692,
+      "structurally_eligible": 868,
+      "success": 738,
+      "success_rate_known_outcomes": 0.43514150943396224,
+      "training_eligible": 0,
+      "trajectories": 1696,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "deepseek-ai/DeepSeek-R1-0528": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 157,
+      "input_token_samples": 0,
+      "mean_actions": 13.305755395683454,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 36.50359712230216,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5477.802158273381,
+      "output_token_samples": 0,
+      "structurally_eligible": 277,
+      "success": 121,
+      "success_rate_known_outcomes": 0.4352517985611511,
+      "training_eligible": 277,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "deepseek-ai/DeepSeek-V3-0324": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 168,
+      "input_token_samples": 0,
+      "mean_actions": 14.212230215827338,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 40.07194244604317,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 8366.03237410072,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 110,
+      "success_rate_known_outcomes": 0.39568345323741005,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "deepseek-ai/DeepSeek-V3.2-Exp-thinking-off": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 151,
+      "input_token_samples": 0,
+      "mean_actions": 15.356115107913668,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 40.989208633093526,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5493.4604316546765,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 127,
+      "success_rate_known_outcomes": 0.4568345323741007,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "deepseek-ai/DeepSeek-V3.2-Exp-thinking-on": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 154,
+      "input_token_samples": 0,
+      "mean_actions": 14.676258992805755,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 39.856115107913666,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 4574.9604316546765,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 124,
+      "success_rate_known_outcomes": 0.4460431654676259,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "deepseek-v3.2@deepseek": {
+      "cost_samples": 373,
+      "duration_samples": 373,
+      "failure": 220,
+      "input_token_samples": 373,
+      "mean_actions": 22.630026809651476,
+      "mean_cost_source_native_units": 1.1921455508981889,
+      "mean_duration_seconds": 724.4749572385732,
+      "mean_events": 46.265415549597854,
+      "mean_input_tokens": 517430.51742627344,
+      "mean_output_tokens": 15099.662198391421,
+      "mean_supervised_assistant_characters": 14214.538873994637,
+      "output_token_samples": 373,
+      "structurally_eligible": 373,
+      "success": 153,
+      "success_rate_known_outcomes": 0.4101876675603217,
+      "training_eligible": 0,
+      "trajectories": 373,
+      "unique_tasks": 87,
+      "unknown": 0
+    },
+    "gemini-2.5-flash@gemini": {
+      "cost_samples": 1678,
+      "duration_samples": 1678,
+      "failure": 1386,
+      "input_token_samples": 1677,
+      "mean_actions": 42.11442193087009,
+      "mean_cost_source_native_units": 20.479215825333995,
+      "mean_duration_seconds": 630.2167853126936,
+      "mean_events": 73.4523241954708,
+      "mean_input_tokens": 2427590.1562313656,
+      "mean_output_tokens": 37394.98270721526,
+      "mean_supervised_assistant_characters": 12389.981525625744,
+      "output_token_samples": 1677,
+      "structurally_eligible": 1151,
+      "success": 292,
+      "success_rate_known_outcomes": 0.17401668653158522,
+      "training_eligible": 0,
+      "trajectories": 1678,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gemini-2.5-pro@gemini": {
+      "cost_samples": 1687,
+      "duration_samples": 1687,
+      "failure": 1270,
+      "input_token_samples": 1685,
+      "mean_actions": 33.0,
+      "mean_cost_source_native_units": 53.944440666818025,
+      "mean_duration_seconds": 612.0814591494268,
+      "mean_events": 57.75874333135744,
+      "mean_input_tokens": 1062349.9721068249,
+      "mean_output_tokens": 28035.420771513353,
+      "mean_supervised_assistant_characters": 9635.45880260818,
+      "output_token_samples": 1685,
+      "structurally_eligible": 1149,
+      "success": 417,
+      "success_rate_known_outcomes": 0.24718435091879076,
+      "training_eligible": 0,
+      "trajectories": 1687,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gemini-3-flash-high": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 121,
+      "input_token_samples": 0,
+      "mean_actions": 56.108,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 113.236,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 68234.066,
+      "output_token_samples": 0,
+      "structurally_eligible": 498,
+      "success": 379,
+      "success_rate_known_outcomes": 0.758,
+      "training_eligible": 0,
+      "trajectories": 500,
+      "unique_tasks": 500,
+      "unknown": 0
+    },
+    "gemini-3-flash-preview@gemini": {
+      "cost_samples": 444,
+      "duration_samples": 444,
+      "failure": 214,
+      "input_token_samples": 0,
+      "mean_actions": 19.763513513513512,
+      "mean_cost_source_native_units": 0.0,
+      "mean_duration_seconds": 545.6824201635412,
+      "mean_events": 40.556306306306304,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 11352.623873873874,
+      "output_token_samples": 0,
+      "structurally_eligible": 444,
+      "success": 230,
+      "success_rate_known_outcomes": 0.5180180180180181,
+      "training_eligible": 0,
+      "trajectories": 444,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gemini-3-pro": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 0,
+      "input_token_samples": 0,
+      "mean_actions": 51.256,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 103.722,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 203.868,
+      "output_token_samples": 0,
+      "structurally_eligible": 490,
+      "success": 500,
+      "success_rate_known_outcomes": 1.0,
+      "training_eligible": 0,
+      "trajectories": 500,
+      "unique_tasks": 500,
+      "unknown": 0
+    },
+    "gemini-3-pro-preview@gemini": {
+      "cost_samples": 445,
+      "duration_samples": 445,
+      "failure": 192,
+      "input_token_samples": 445,
+      "mean_actions": 13.262921348314606,
+      "mean_cost_source_native_units": 45.99529477558779,
+      "mean_duration_seconds": 548.5905832183495,
+      "mean_events": 27.557303370786517,
+      "mean_input_tokens": 320152.2426966292,
+      "mean_output_tokens": 24330.34831460674,
+      "mean_supervised_assistant_characters": 7961.357303370786,
+      "output_token_samples": 445,
+      "structurally_eligible": 445,
+      "success": 253,
+      "success_rate_known_outcomes": 0.5685393258426966,
+      "training_eligible": 0,
+      "trajectories": 445,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gemini-3.1-pro-preview@Google": {
+      "cost_samples": 431,
+      "duration_samples": 431,
+      "failure": 111,
+      "input_token_samples": 431,
+      "mean_actions": 26.454756380510442,
+      "mean_cost_source_native_units": 75.75997570551189,
+      "mean_duration_seconds": 632.5132716946702,
+      "mean_events": 53.909512761020885,
+      "mean_input_tokens": 992285.0626450116,
+      "mean_output_tokens": 29053.654292343388,
+      "mean_supervised_assistant_characters": 4961.839907192575,
+      "output_token_samples": 431,
+      "structurally_eligible": 431,
+      "success": 320,
+      "success_rate_known_outcomes": 0.7424593967517401,
+      "training_eligible": 0,
+      "trajectories": 431,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "glm-5-high": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 136,
+      "input_token_samples": 0,
+      "mean_actions": 76.026,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 154.152,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 350.692,
+      "output_token_samples": 0,
+      "structurally_eligible": 500,
+      "success": 364,
+      "success_rate_known_outcomes": 0.728,
+      "training_eligible": 0,
+      "trajectories": 500,
+      "unique_tasks": 500,
+      "unknown": 0
+    },
+    "glm-5@z-ai": {
+      "cost_samples": 376,
+      "duration_samples": 376,
+      "failure": 165,
+      "input_token_samples": 376,
+      "mean_actions": 28.388297872340427,
+      "mean_cost_source_native_units": 0.0,
+      "mean_duration_seconds": 809.2636813914522,
+      "mean_events": 57.81382978723404,
+      "mean_input_tokens": 0.0,
+      "mean_output_tokens": 0.0,
+      "mean_supervised_assistant_characters": 14967.672872340425,
+      "output_token_samples": 376,
+      "structurally_eligible": 376,
+      "success": 211,
+      "success_rate_known_outcomes": 0.5611702127659575,
+      "training_eligible": 0,
+      "trajectories": 376,
+      "unique_tasks": 88,
+      "unknown": 0
+    },
+    "google/gemini-2.5-flash-thinking-off": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 168,
+      "input_token_samples": 0,
+      "mean_actions": 11.456834532374101,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 25.154676258992804,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 1610.4496402877699,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 110,
+      "success_rate_known_outcomes": 0.39568345323741005,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "google/gemini-2.5-flash-thinking-on": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 176,
+      "input_token_samples": 0,
+      "mean_actions": 15.902877697841726,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 39.02517985611511,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 2495.3920863309354,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 102,
+      "success_rate_known_outcomes": 0.3669064748201439,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "google/gemini-2.5-pro-thinking-on": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 165,
+      "input_token_samples": 0,
+      "mean_actions": 15.02158273381295,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 37.53237410071942,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 2018.86690647482,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 113,
+      "success_rate_known_outcomes": 0.4064748201438849,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "gpt-4o-2024-08-06": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 5308,
+      "input_token_samples": 0,
+      "mean_actions": 21.264601320467243,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 44.01574403250381,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 7167.808362959201,
+      "output_token_samples": 0,
+      "structurally_eligible": 5907,
+      "success": 599,
+      "success_rate_known_outcomes": 0.10140511257829693,
+      "training_eligible": 5907,
+      "trajectories": 5907,
+      "unique_tasks": 2649,
+      "unknown": 0
+    },
+    "gpt-5-codex@openai": {
+      "cost_samples": 1027,
+      "duration_samples": 1027,
+      "failure": 579,
+      "input_token_samples": 1025,
+      "mean_actions": 23.137293086660176,
+      "mean_cost_source_native_units": 34.65822178652798,
+      "mean_duration_seconds": 894.1482539265158,
+      "mean_events": 47.249269717624145,
+      "mean_input_tokens": 404544.2819512195,
+      "mean_output_tokens": 24892.82731707317,
+      "mean_supervised_assistant_characters": 9759.669912366115,
+      "output_token_samples": 1025,
+      "structurally_eligible": 882,
+      "success": 448,
+      "success_rate_known_outcomes": 0.436222005842259,
+      "training_eligible": 0,
+      "trajectories": 1027,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gpt-5-mini": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 219,
+      "input_token_samples": 0,
+      "mean_actions": 20.108,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 41.45,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 1133.53,
+      "output_token_samples": 0,
+      "structurally_eligible": 500,
+      "success": 281,
+      "success_rate_known_outcomes": 0.562,
+      "training_eligible": 0,
+      "trajectories": 500,
+      "unique_tasks": 500,
+      "unknown": 0
+    },
+    "gpt-5-mini@openai": {
+      "cost_samples": 1705,
+      "duration_samples": 1705,
+      "failure": 1231,
+      "input_token_samples": 1703,
+      "mean_actions": 20.30909090909091,
+      "mean_cost_source_native_units": 6.237601142212792,
+      "mean_duration_seconds": 738.1692138985105,
+      "mean_events": 36.47214076246334,
+      "mean_input_tokens": 362939.1297709924,
+      "mean_output_tokens": 20953.19553728714,
+      "mean_supervised_assistant_characters": 7645.26568914956,
+      "output_token_samples": 1703,
+      "structurally_eligible": 857,
+      "success": 474,
+      "success_rate_known_outcomes": 0.2780058651026393,
+      "training_eligible": 0,
+      "trajectories": 1705,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gpt-5-nano@openai": {
+      "cost_samples": 1707,
+      "duration_samples": 1707,
+      "failure": 1546,
+      "input_token_samples": 1703,
+      "mean_actions": 27.11833626244874,
+      "mean_cost_source_native_units": 2.833243009294413,
+      "mean_duration_seconds": 795.4362061986046,
+      "mean_events": 49.25834797891037,
+      "mean_input_tokens": 561219.6964180857,
+      "mean_output_tokens": 53272.514386376984,
+      "mean_supervised_assistant_characters": 11135.304042179261,
+      "output_token_samples": 1703,
+      "structurally_eligible": 861,
+      "success": 161,
+      "success_rate_known_outcomes": 0.09431751611013474,
+      "training_eligible": 0,
+      "trajectories": 1707,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gpt-5.1-codex@openai": {
+      "cost_samples": 441,
+      "duration_samples": 441,
+      "failure": 277,
+      "input_token_samples": 440,
+      "mean_actions": 22.573696145124718,
+      "mean_cost_source_native_units": 31.287152531770083,
+      "mean_duration_seconds": 1209.0573483940695,
+      "mean_events": 46.14965986394558,
+      "mean_input_tokens": 481445.4818181818,
+      "mean_output_tokens": 20365.454545454544,
+      "mean_supervised_assistant_characters": 11924.947845804989,
+      "output_token_samples": 440,
+      "structurally_eligible": 429,
+      "success": 164,
+      "success_rate_known_outcomes": 0.37188208616780044,
+      "training_eligible": 0,
+      "trajectories": 441,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gpt-5.1@openai": {
+      "cost_samples": 445,
+      "duration_samples": 445,
+      "failure": 233,
+      "input_token_samples": 444,
+      "mean_actions": 8.55056179775281,
+      "mean_cost_source_native_units": 41.71665708241838,
+      "mean_duration_seconds": 845.7475696006518,
+      "mean_events": 18.10112359550562,
+      "mean_input_tokens": 134251.63288288287,
+      "mean_output_tokens": 37152.01351351351,
+      "mean_supervised_assistant_characters": 6334.330337078652,
+      "output_token_samples": 444,
+      "structurally_eligible": 438,
+      "success": 212,
+      "success_rate_known_outcomes": 0.4764044943820225,
+      "training_eligible": 0,
+      "trajectories": 445,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gpt-5.2-codex": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 86,
+      "input_token_samples": 0,
+      "mean_actions": 35.49090909090909,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 73.75636363636363,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 949.08,
+      "output_token_samples": 0,
+      "structurally_eligible": 275,
+      "success": 189,
+      "success_rate_known_outcomes": 0.6872727272727273,
+      "training_eligible": 0,
+      "trajectories": 275,
+      "unique_tasks": 275,
+      "unknown": 0
+    },
+    "gpt-5.2-codex@openai": {
+      "cost_samples": 380,
+      "duration_samples": 380,
+      "failure": 98,
+      "input_token_samples": 0,
+      "mean_actions": 29.263157894736842,
+      "mean_cost_source_native_units": 0.0,
+      "mean_duration_seconds": 705.5991674925151,
+      "mean_events": 59.747368421052634,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 4741.5447368421055,
+      "output_token_samples": 0,
+      "structurally_eligible": 337,
+      "success": 282,
+      "success_rate_known_outcomes": 0.7421052631578947,
+      "training_eligible": 0,
+      "trajectories": 380,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gpt-5.2-high": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 72,
+      "input_token_samples": 0,
+      "mean_actions": 35.806691449814124,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 74.45724907063197,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 943.1858736059479,
+      "output_token_samples": 0,
+      "structurally_eligible": 269,
+      "success": 197,
+      "success_rate_known_outcomes": 0.7323420074349443,
+      "training_eligible": 0,
+      "trajectories": 269,
+      "unique_tasks": 269,
+      "unknown": 0
+    },
+    "gpt-5.2@openai": {
+      "cost_samples": 442,
+      "duration_samples": 442,
+      "failure": 203,
+      "input_token_samples": 0,
+      "mean_actions": 8.339366515837105,
+      "mean_cost_source_native_units": 0.0,
+      "mean_duration_seconds": 812.1008663954239,
+      "mean_events": 17.67873303167421,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 9987.174208144796,
+      "output_token_samples": 0,
+      "structurally_eligible": 435,
+      "success": 239,
+      "success_rate_known_outcomes": 0.5407239819004525,
+      "training_eligible": 0,
+      "trajectories": 442,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gpt-5@openai": {
+      "cost_samples": 1707,
+      "duration_samples": 1707,
+      "failure": 989,
+      "input_token_samples": 1705,
+      "mean_actions": 18.446983011130637,
+      "mean_cost_source_native_units": 32.87692377919262,
+      "mean_duration_seconds": 789.6129793145884,
+      "mean_events": 34.26479203280609,
+      "mean_input_tokens": 288354.0152492669,
+      "mean_output_tokens": 26094.951906158356,
+      "mean_supervised_assistant_characters": 6976.230228471002,
+      "output_token_samples": 1705,
+      "structurally_eligible": 854,
+      "success": 718,
+      "success_rate_known_outcomes": 0.4206209724663152,
+      "training_eligible": 0,
+      "trajectories": 1707,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "gpt-oss-120b": {
+      "cost_samples": 0,
+      "duration_samples": 5063,
+      "failure": 3852,
+      "input_token_samples": 0,
+      "mean_actions": 27.13786292711831,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": 108.07624142800492,
+      "mean_events": 54.28303377444203,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 15385.260912502468,
+      "output_token_samples": 0,
+      "structurally_eligible": 5063,
+      "success": 1211,
+      "success_rate_known_outcomes": 0.23918625320955955,
+      "training_eligible": 0,
+      "trajectories": 5063,
+      "unique_tasks": 1266,
+      "unknown": 0
+    },
+    "grok-4-0709@xai": {
+      "cost_samples": 1264,
+      "duration_samples": 1264,
+      "failure": 927,
+      "input_token_samples": 1254,
+      "mean_actions": 24.593354430379748,
+      "mean_cost_source_native_units": -2.574454700680379,
+      "mean_duration_seconds": 1095.5605030421968,
+      "mean_events": 43.932753164556964,
+      "mean_input_tokens": 292064.3724082935,
+      "mean_output_tokens": 3859.8508771929824,
+      "mean_supervised_assistant_characters": 7819.378955696203,
+      "output_token_samples": 1254,
+      "structurally_eligible": 821,
+      "success": 337,
+      "success_rate_known_outcomes": 0.2666139240506329,
+      "training_eligible": 0,
+      "trajectories": 1264,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "grok-code-fast-1@xai": {
+      "cost_samples": 880,
+      "duration_samples": 880,
+      "failure": 702,
+      "input_token_samples": 878,
+      "mean_actions": 15.525,
+      "mean_cost_source_native_units": 1.126558439504482,
+      "mean_duration_seconds": 348.8899872476404,
+      "mean_events": 32.86022727272727,
+      "mean_input_tokens": 258086.56719817768,
+      "mean_output_tokens": 2870.0159453302963,
+      "mean_supervised_assistant_characters": 7485.394318181819,
+      "output_token_samples": 878,
+      "structurally_eligible": 876,
+      "success": 178,
+      "success_rate_known_outcomes": 0.20227272727272727,
+      "training_eligible": 0,
+      "trajectories": 880,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "minimax-m2.1@minimax": {
+      "cost_samples": 440,
+      "duration_samples": 440,
+      "failure": 277,
+      "input_token_samples": 0,
+      "mean_actions": 47.49318181818182,
+      "mean_cost_source_native_units": 0.0,
+      "mean_duration_seconds": 779.6249267968265,
+      "mean_events": 85.37727272727273,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 8243.768181818183,
+      "output_token_samples": 0,
+      "structurally_eligible": 1,
+      "success": 163,
+      "success_rate_known_outcomes": 0.3704545454545455,
+      "training_eligible": 0,
+      "trajectories": 440,
+      "unique_tasks": 88,
+      "unknown": 0
+    },
+    "minimax-m2.1@openai": {
+      "cost_samples": 438,
+      "duration_samples": 438,
+      "failure": 308,
+      "input_token_samples": 0,
+      "mean_actions": 30.929223744292237,
+      "mean_cost_source_native_units": 0.0,
+      "mean_duration_seconds": 916.7981030538202,
+      "mean_events": 62.858447488584474,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 19007.906392694065,
+      "output_token_samples": 0,
+      "structurally_eligible": 438,
+      "success": 130,
+      "success_rate_known_outcomes": 0.2968036529680365,
+      "training_eligible": 0,
+      "trajectories": 438,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "minimax-m2.5-high": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 121,
+      "input_token_samples": 0,
+      "mean_actions": 60.438,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 121.894,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 1597.206,
+      "output_token_samples": 0,
+      "structurally_eligible": 500,
+      "success": 379,
+      "success_rate_known_outcomes": 0.758,
+      "training_eligible": 0,
+      "trajectories": 500,
+      "unique_tasks": 500,
+      "unknown": 0
+    },
+    "minimax-m2.5@Minimax": {
+      "cost_samples": 411,
+      "duration_samples": 411,
+      "failure": 230,
+      "input_token_samples": 411,
+      "mean_actions": 29.394160583941606,
+      "mean_cost_source_native_units": 0.0,
+      "mean_duration_seconds": 1034.1070254434924,
+      "mean_events": 60.02676399026764,
+      "mean_input_tokens": 181155.46228710463,
+      "mean_output_tokens": 12802.311435523114,
+      "mean_supervised_assistant_characters": 12416.776155717762,
+      "output_token_samples": 411,
+      "structurally_eligible": 410,
+      "success": 181,
+      "success_rate_known_outcomes": 0.44038929440389296,
+      "training_eligible": 0,
+      "trajectories": 411,
+      "unique_tasks": 88,
+      "unknown": 0
+    },
+    "moonshotai/Kimi-K2-Instruct-0905@together_ai": {
+      "cost_samples": 1783,
+      "duration_samples": 1783,
+      "failure": 1287,
+      "input_token_samples": 1783,
+      "mean_actions": 69.88951205832866,
+      "mean_cost_source_native_units": 117.50347072651596,
+      "mean_duration_seconds": 679.7632403584992,
+      "mean_events": 114.4823331463825,
+      "mean_input_tokens": 1144773.9444756028,
+      "mean_output_tokens": 10086.920919798093,
+      "mean_supervised_assistant_characters": 29183.99046550757,
+      "output_token_samples": 1783,
+      "structurally_eligible": 923,
+      "success": 496,
+      "success_rate_known_outcomes": 0.2781828379136287,
+      "training_eligible": 0,
+      "trajectories": 1783,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "moonshotai/Kimi-K2-Thinking@together_ai": {
+      "cost_samples": 892,
+      "duration_samples": 892,
+      "failure": 572,
+      "input_token_samples": 890,
+      "mean_actions": 34.10538116591928,
+      "mean_cost_source_native_units": 119.29089524606952,
+      "mean_duration_seconds": 971.4054908068191,
+      "mean_events": 70.29820627802691,
+      "mean_input_tokens": 949991.6898876404,
+      "mean_output_tokens": 18533.206741573034,
+      "mean_supervised_assistant_characters": 29966.461883408072,
+      "output_token_samples": 890,
+      "structurally_eligible": 892,
+      "success": 320,
+      "success_rate_known_outcomes": 0.35874439461883406,
+      "training_eligible": 0,
+      "trajectories": 892,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "openai/gpt-4.1": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 153,
+      "input_token_samples": 0,
+      "mean_actions": 14.446043165467627,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 40.118705035971225,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 3550.039568345324,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 125,
+      "success_rate_known_outcomes": 0.44964028776978415,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "openai/gpt-4.1-mini": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 158,
+      "input_token_samples": 0,
+      "mean_actions": 19.776978417266186,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 50.32014388489208,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 3496.9460431654675,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 120,
+      "success_rate_known_outcomes": 0.4316546762589928,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "openai/gpt-4.1-nano": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 238,
+      "input_token_samples": 0,
+      "mean_actions": 33.71223021582734,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 77.18705035971223,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 4114.230215827338,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 40,
+      "success_rate_known_outcomes": 0.14388489208633093,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "openai/gpt-4o-20240806": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 170,
+      "input_token_samples": 0,
+      "mean_actions": 17.758992805755394,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 48.302158273381295,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 3614.1582733812947,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 108,
+      "success_rate_known_outcomes": 0.38848920863309355,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "openai/gpt-4o-mini": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 205,
+      "input_token_samples": 0,
+      "mean_actions": 45.21223021582734,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 93.01079136690647,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 1972.478417266187,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 73,
+      "success_rate_known_outcomes": 0.26258992805755393,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "openai/gpt-5": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 138,
+      "input_token_samples": 0,
+      "mean_actions": 12.305755395683454,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 40.26618705035971,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 4689.446043165468,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 140,
+      "success_rate_known_outcomes": 0.5035971223021583,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "openai/gpt-5-nano": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 183,
+      "input_token_samples": 0,
+      "mean_actions": 24.07913669064748,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 61.83453237410072,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 8785.273381294965,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 95,
+      "success_rate_known_outcomes": 0.34172661870503596,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "openai/gpt-oss-120b@together_ai": {
+      "cost_samples": 1805,
+      "duration_samples": 1805,
+      "failure": 1502,
+      "input_token_samples": 1801,
+      "mean_actions": 10.353462603878116,
+      "mean_cost_source_native_units": 2.2321994562228302,
+      "mean_duration_seconds": 383.20367774012345,
+      "mean_events": 22.829362880886425,
+      "mean_input_tokens": 118470.27984453081,
+      "mean_output_tokens": 7668.382565241533,
+      "mean_supervised_assistant_characters": 10867.516897506925,
+      "output_token_samples": 1801,
+      "structurally_eligible": 1789,
+      "success": 303,
+      "success_rate_known_outcomes": 0.16786703601108033,
+      "training_eligible": 0,
+      "trajectories": 1805,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "openai/gpt-oss-20b@together_ai": {
+      "cost_samples": 1807,
+      "duration_samples": 1807,
+      "failure": 1749,
+      "input_token_samples": 1805,
+      "mean_actions": 22.045379081350305,
+      "mean_cost_source_native_units": 2.7351576161561053,
+      "mean_duration_seconds": 225.4812311870193,
+      "mean_events": 47.460431654676256,
+      "mean_input_tokens": 503665.8925207756,
+      "mean_output_tokens": 10992.939612188366,
+      "mean_supervised_assistant_characters": 16832.06917542889,
+      "output_token_samples": 1805,
+      "structurally_eligible": 1745,
+      "success": 58,
+      "success_rate_known_outcomes": 0.032097399003873824,
+      "training_eligible": 0,
+      "trajectories": 1807,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "openai/o3-high": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 136,
+      "input_token_samples": 0,
+      "mean_actions": 15.010791366906474,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 42.780575539568346,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 3309.9172661870502,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 142,
+      "success_rate_known_outcomes": 0.5107913669064749,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "openai/o4-mini-high": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 162,
+      "input_token_samples": 0,
+      "mean_actions": 14.733812949640289,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 39.410071942446045,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 2741.8057553956833,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 116,
+      "success_rate_known_outcomes": 0.4172661870503597,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "togetherai/Qwen/Qwen3-235B-A22B-FP8": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 181,
+      "input_token_samples": 0,
+      "mean_actions": 14.18705035971223,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 35.55755395683453,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 3833.568345323741,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 97,
+      "success_rate_known_outcomes": 0.3489208633093525,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "togetherai/Qwen/Qwen3-235B-A22B-Instruct-2507-FP8": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 154,
+      "input_token_samples": 0,
+      "mean_actions": 16.37410071942446,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 41.82374100719424,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 3558.928057553957,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 124,
+      "success_rate_known_outcomes": 0.4460431654676259,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "togetherai/Qwen/Qwen3-235B-A22B-Thinking-2507-FP8": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 162,
+      "input_token_samples": 0,
+      "mean_actions": 15.913669064748202,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 40.17985611510792,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 7153.723021582734,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 116,
+      "success_rate_known_outcomes": 0.4172661870503597,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "togetherai/Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 218,
+      "input_token_samples": 0,
+      "mean_actions": 14.129496402877697,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 38.697841726618705,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 3825.9172661870502,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 60,
+      "success_rate_known_outcomes": 0.2158273381294964,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "togetherai/moonshotai/Kimi-K2-Instruct": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 141,
+      "input_token_samples": 0,
+      "mean_actions": 14.881294964028777,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 40.35251798561151,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5261.179856115108,
+      "output_token_samples": 0,
+      "structurally_eligible": 276,
+      "success": 137,
+      "success_rate_known_outcomes": 0.49280575539568344,
+      "training_eligible": 276,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "togetherai/moonshotai/Kimi-K2-Instruct-0905": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 139,
+      "input_token_samples": 0,
+      "mean_actions": 16.007194244604317,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 42.42446043165467,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5449.546762589928,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 139,
+      "success_rate_known_outcomes": 0.5,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "togetherai/openai/gpt-oss-120b": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 221,
+      "input_token_samples": 0,
+      "mean_actions": 13.420863309352518,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 35.97122302158273,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 14511.31654676259,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 57,
+      "success_rate_known_outcomes": 0.20503597122302158,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "togetherai/openai/gpt-oss-20b": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 248,
+      "input_token_samples": 0,
+      "mean_actions": 45.29136690647482,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 100.81294964028777,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 29205.543165467625,
+      "output_token_samples": 0,
+      "structurally_eligible": 278,
+      "success": 30,
+      "success_rate_known_outcomes": 0.1079136690647482,
+      "training_eligible": 278,
+      "trajectories": 278,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "unknown": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 0,
+      "input_token_samples": 0,
+      "mean_actions": 16.45260465116279,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 32.974976744186044,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 10075.690186046511,
+      "output_token_samples": 0,
+      "structurally_eligible": 20643,
+      "success": 491,
+      "success_rate_known_outcomes": 1.0,
+      "training_eligible": 20643,
+      "trajectories": 21500,
+      "unique_tasks": 19089,
+      "unknown": 21009
+    }
+  },
+  "by_outcome": {
+    "failure": {
+      "cost_samples": 23212,
+      "duration_samples": 27064,
+      "failure": 79973,
+      "input_token_samples": 21454,
+      "mean_actions": 45.62608630412765,
+      "mean_cost_source_native_units": 75.64188306777397,
+      "mean_duration_seconds": 701.5190378383018,
+      "mean_events": 94.278981656309,
+      "mean_input_tokens": 1156962.542835835,
+      "mean_output_tokens": 21745.191386221682,
+      "mean_supervised_assistant_characters": 12323.303715003814,
+      "output_token_samples": 21454,
+      "structurally_eligible": 73054,
+      "success": 0,
+      "success_rate_known_outcomes": 0.0,
+      "training_eligible": 51532,
+      "trajectories": 79973,
+      "unique_tasks": 28927,
+      "unknown": 0
+    },
+    "success": {
+      "cost_samples": 11051,
+      "duration_samples": 12262,
+      "failure": 0,
+      "input_token_samples": 9161,
+      "mean_actions": 36.86015544423162,
+      "mean_cost_source_native_units": 31.06763293005613,
+      "mean_duration_seconds": 410.78903144437595,
+      "mean_events": 76.32702224841609,
+      "mean_input_tokens": 382399.21984499507,
+      "mean_output_tokens": 11283.977950005457,
+      "mean_supervised_assistant_characters": 7551.789303079417,
+      "output_token_samples": 9161,
+      "structurally_eligible": 50793,
+      "success": 54296,
+      "success_rate_known_outcomes": 1.0,
+      "training_eligible": 38524,
+      "trajectories": 54296,
+      "unique_tasks": 23353,
+      "unknown": 0
+    },
+    "unknown": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 0,
+      "input_token_samples": 0,
+      "mean_actions": 16.3787424437146,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 32.805511923461374,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 10051.800371269455,
+      "output_token_samples": 0,
+      "structurally_eligible": 20152,
+      "success": 0,
+      "success_rate_known_outcomes": null,
+      "training_eligible": 20152,
+      "trajectories": 21009,
+      "unique_tasks": 18796,
+      "unknown": 21009
+    }
+  },
+  "by_quality_tier": {
+    "3": {
+      "cost_samples": 7668,
+      "duration_samples": 8879,
+      "failure": 0,
+      "input_token_samples": 6699,
+      "mean_actions": 35.69987645264661,
+      "mean_cost_source_native_units": 23.312137899828038,
+      "mean_duration_seconds": 387.87317761634836,
+      "mean_events": 75.29464113354696,
+      "mean_input_tokens": 242325.22913867742,
+      "mean_output_tokens": 11504.635617256306,
+      "mean_supervised_assistant_characters": 7296.183294081309,
+      "output_token_samples": 6699,
+      "structurally_eligible": 51802,
+      "success": 50793,
+      "success_rate_known_outcomes": 1.0,
+      "training_eligible": 39533,
+      "trajectories": 51802,
+      "unique_tasks": 24316,
+      "unknown": 1009
+    },
+    "4": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 0,
+      "input_token_samples": 0,
+      "mean_actions": 17.524369221125216,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 35.04873844225043,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 10991.896829128142,
+      "output_token_samples": 0,
+      "structurally_eligible": 19143,
+      "success": 0,
+      "success_rate_known_outcomes": null,
+      "training_eligible": 19143,
+      "trajectories": 19143,
+      "unique_tasks": 16936,
+      "unknown": 19143
+    },
+    "5": {
+      "cost_samples": 26595,
+      "duration_samples": 30447,
+      "failure": 79973,
+      "input_token_samples": 23916,
+      "mean_actions": 45.172364317645524,
+      "mean_cost_source_native_units": 72.20794615015373,
+      "mean_duration_seconds": 675.8984596042691,
+      "mean_events": 92.5128360191147,
+      "mean_input_tokens": 1116461.6548335843,
+      "mean_output_tokens": 20606.468640240844,
+      "mean_supervised_assistant_characters": 12075.546061446883,
+      "output_token_samples": 23916,
+      "structurally_eligible": 73054,
+      "success": 3503,
+      "success_rate_known_outcomes": 0.041964157362595234,
+      "training_eligible": 51532,
+      "trajectories": 84333,
+      "unique_tasks": 29808,
+      "unknown": 857
+    }
+  },
+  "by_source": {
+    "agentsuite_tau2": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 4965,
+      "input_token_samples": 0,
+      "mean_actions": 17.801798561151077,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 45.433093525179856,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5750.385491606715,
+      "output_token_samples": 0,
+      "structurally_eligible": 8337,
+      "success": 3375,
+      "success_rate_known_outcomes": 0.40467625899280574,
+      "training_eligible": 8337,
+      "trajectories": 8340,
+      "unique_tasks": 228,
+      "unknown": 0
+    },
+    "browsecomp_rollouts": {
+      "cost_samples": 0,
+      "duration_samples": 5063,
+      "failure": 3852,
+      "input_token_samples": 0,
+      "mean_actions": 27.13786292711831,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": 108.07624142800492,
+      "mean_events": 54.28303377444203,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 15385.260912502468,
+      "output_token_samples": 0,
+      "structurally_eligible": 5063,
+      "success": 1211,
+      "success_rate_known_outcomes": 0.23918625320955955,
+      "training_eligible": 0,
+      "trajectories": 5063,
+      "unique_tasks": 1266,
+      "unknown": 0
+    },
+    "mind2web": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 0,
+      "input_token_samples": 0,
+      "mean_actions": 7.705649157581764,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 16.411298315163528,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 256.4281466798811,
+      "output_token_samples": 0,
+      "structurally_eligible": 1009,
+      "success": 0,
+      "success_rate_known_outcomes": null,
+      "training_eligible": 1009,
+      "trajectories": 1009,
+      "unique_tasks": 1009,
+      "unknown": 1009
+    },
+    "nvidia_open_swe_traces": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 26883,
+      "input_token_samples": 0,
+      "mean_actions": 58.35106341765777,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 124.11902917134778,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 4650.015089296091,
+      "output_token_samples": 0,
+      "structurally_eligible": 51626,
+      "success": 24743,
+      "success_rate_known_outcomes": 0.4792740092201604,
+      "training_eligible": 51626,
+      "trajectories": 51626,
+      "unique_tasks": 33444,
+      "unknown": 0
+    },
+    "official_tau3": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 100,
+      "input_token_samples": 0,
+      "mean_actions": 14.5,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 30.0,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 4221.82,
+      "output_token_samples": 0,
+      "structurally_eligible": 0,
+      "success": 100,
+      "success_rate_known_outcomes": 0.5,
+      "training_eligible": 0,
+      "trajectories": 200,
+      "unique_tasks": 50,
+      "unknown": 0
+    },
+    "swebench_multilingual_trajectories": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 121,
+      "input_token_samples": 0,
+      "mean_actions": 47.37984496124031,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 96.75968992248062,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 36004.64341085271,
+      "output_token_samples": 0,
+      "structurally_eligible": 110,
+      "success": 8,
+      "success_rate_known_outcomes": 0.06201550387596899,
+      "training_eligible": 0,
+      "trajectories": 129,
+      "unique_tasks": 43,
+      "unknown": 0
+    },
+    "swebench_verified_trajectories": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 1152,
+      "input_token_samples": 0,
+      "mean_actions": 47.387984154929576,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 97.00748239436619,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 12330.321082746479,
+      "output_token_samples": 0,
+      "structurally_eligible": 4532,
+      "success": 3392,
+      "success_rate_known_outcomes": 0.7464788732394366,
+      "training_eligible": 0,
+      "trajectories": 4544,
+      "unique_tasks": 500,
+      "unknown": 0
+    },
+    "swegym_openhands": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 0,
+      "input_token_samples": 0,
+      "mean_actions": 19.613034623217924,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 40.22606924643585,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 11097.892057026476,
+      "output_token_samples": 0,
+      "structurally_eligible": 491,
+      "success": 491,
+      "success_rate_known_outcomes": 1.0,
+      "training_eligible": 491,
+      "trajectories": 491,
+      "unique_tasks": 293,
+      "unknown": 0
+    },
+    "swegym_sampled": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 5011,
+      "input_token_samples": 0,
+      "mean_actions": 20.382406397673574,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 42.287168302435475,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 7133.47528171574,
+      "output_token_samples": 0,
+      "structurally_eligible": 5502,
+      "success": 491,
+      "success_rate_known_outcomes": 0.08924027626317703,
+      "training_eligible": 5502,
+      "trajectories": 5502,
+      "unique_tasks": 2438,
+      "unknown": 0
+    },
+    "swesmith_trajectories": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 14673,
+      "input_token_samples": 0,
+      "mean_actions": 31.585477178423236,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 64.17095435684648,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 17664.133236514524,
+      "output_token_samples": 0,
+      "structurally_eligible": 24100,
+      "success": 9427,
+      "success_rate_known_outcomes": 0.39116182572614105,
+      "training_eligible": 24100,
+      "trajectories": 24100,
+      "unique_tasks": 10637,
+      "unknown": 0
+    },
+    "terminalbench_trajectories": {
+      "cost_samples": 34263,
+      "duration_samples": 34263,
+      "failure": 23212,
+      "input_token_samples": 30615,
+      "mean_actions": 36.32335172051484,
+      "mean_cost_source_native_units": 61.26514903187753,
+      "mean_duration_seconds": 685.1652316865642,
+      "mean_events": 69.00125499810291,
+      "mean_input_tokens": 925188.0988731014,
+      "mean_output_tokens": 18614.85735750449,
+      "mean_supervised_assistant_characters": 14526.395032542392,
+      "output_token_samples": 30615,
+      "structurally_eligible": 24086,
+      "success": 11051,
+      "success_rate_known_outcomes": 0.32253451244783005,
+      "training_eligible": 0,
+      "trajectories": 34263,
+      "unique_tasks": 89,
+      "unknown": 0
+    },
+    "terminaltraj": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 0,
+      "input_token_samples": 0,
+      "mean_actions": 16.8163,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 33.6326,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 10545.9769,
+      "output_token_samples": 0,
+      "structurally_eligible": 19143,
+      "success": 0,
+      "success_rate_known_outcomes": null,
+      "training_eligible": 19143,
+      "trajectories": 20000,
+      "unique_tasks": 17787,
+      "unknown": 20000
+    },
+    "toolbench": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 4,
+      "input_token_samples": 0,
+      "mean_actions": 3.8181818181818183,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 9.090909090909092,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 834.3636363636364,
+      "output_token_samples": 0,
+      "structurally_eligible": 0,
+      "success": 7,
+      "success_rate_known_outcomes": 0.6363636363636364,
+      "training_eligible": 0,
+      "trajectories": 11,
+      "unique_tasks": 11,
+      "unknown": 0
+    }
+  },
+  "by_task_family": {
+    "software-engineering": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 47840,
+      "input_token_samples": 0,
+      "mean_actions": 47.65324335586629,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 100.24062413186407,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 8926.038452634502,
+      "output_token_samples": 0,
+      "structurally_eligible": 86361,
+      "success": 38552,
+      "success_rate_known_outcomes": 0.4462450226872859,
+      "training_eligible": 81719,
+      "trajectories": 86392,
+      "unique_tasks": 47170,
+      "unknown": 0
+    },
+    "terminal-agent": {
+      "cost_samples": 34263,
+      "duration_samples": 34263,
+      "failure": 23212,
+      "input_token_samples": 30615,
+      "mean_actions": 29.13353482114885,
+      "mean_cost_source_native_units": 61.26514903187753,
+      "mean_duration_seconds": 685.1652316865642,
+      "mean_events": 55.965243351823524,
+      "mean_input_tokens": 925188.0988731014,
+      "mean_output_tokens": 18614.85735750449,
+      "mean_supervised_assistant_characters": 13059.311335532499,
+      "output_token_samples": 30615,
+      "structurally_eligible": 43229,
+      "success": 11051,
+      "success_rate_known_outcomes": 0.32253451244783005,
+      "training_eligible": 19143,
+      "trajectories": 54263,
+      "unique_tasks": 17876,
+      "unknown": 20000
+    },
+    "tool-agent-user-simulation": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 5065,
+      "input_token_samples": 0,
+      "mean_actions": 17.724473067915692,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 45.071662763466044,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 5714.587704918033,
+      "output_token_samples": 0,
+      "structurally_eligible": 8337,
+      "success": 3475,
+      "success_rate_known_outcomes": 0.4069086651053864,
+      "training_eligible": 8337,
+      "trajectories": 8540,
+      "unique_tasks": 278,
+      "unknown": 0
+    },
+    "tool-use": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 4,
+      "input_token_samples": 0,
+      "mean_actions": 3.8181818181818183,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 9.090909090909092,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 834.3636363636364,
+      "output_token_samples": 0,
+      "structurally_eligible": 0,
+      "success": 7,
+      "success_rate_known_outcomes": 0.6363636363636364,
+      "training_eligible": 0,
+      "trajectories": 11,
+      "unique_tasks": 11,
+      "unknown": 0
+    },
+    "web-navigation": {
+      "cost_samples": 0,
+      "duration_samples": 0,
+      "failure": 0,
+      "input_token_samples": 0,
+      "mean_actions": 7.705649157581764,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": null,
+      "mean_events": 16.411298315163528,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 256.4281466798811,
+      "output_token_samples": 0,
+      "structurally_eligible": 1009,
+      "success": 0,
+      "success_rate_known_outcomes": null,
+      "training_eligible": 1009,
+      "trajectories": 1009,
+      "unique_tasks": 1009,
+      "unknown": 1009
+    },
+    "web-research": {
+      "cost_samples": 0,
+      "duration_samples": 5063,
+      "failure": 3852,
+      "input_token_samples": 0,
+      "mean_actions": 27.13786292711831,
+      "mean_cost_source_native_units": null,
+      "mean_duration_seconds": 108.07624142800492,
+      "mean_events": 54.28303377444203,
+      "mean_input_tokens": null,
+      "mean_output_tokens": null,
+      "mean_supervised_assistant_characters": 15385.260912502468,
+      "output_token_samples": 0,
+      "structurally_eligible": 5063,
+      "success": 1211,
+      "success_rate_known_outcomes": 0.23918625320955955,
+      "training_eligible": 0,
+      "trajectories": 5063,
+      "unique_tasks": 1266,
+      "unknown": 0
+    }
+  },
+  "compatible_task_environments": 67660,
+  "evaluated_trajectories": 155278,
+  "evaluation_scope": "all imported canonical trajectories, including every alternative attempt",
+  "identical_duplicate_records": 1208,
+  "independently_reproduced": 0,
+  "label_policy": {
+    "cost_units_are_source_native_and_not_globally_comparable": true,
+    "historical_outcomes_are_target_model_scores": false,
+    "independent_reproduction_requires_explicit_evidence": true,
+    "unknown_is_failure": false
+  },
+  "outcomes": {
+    "failure": 79973,
+    "known_denominator": 134269,
+    "success": 54296,
+    "success_rate_all_trajectories": 0.34966962480196806,
+    "success_rate_known_outcomes": 0.40438224757762403,
+    "unknown": 21009
+  },
+  "quality_tiers": {
+    "3": 51802,
+    "4": 19143,
+    "5": 84333
+  },
+  "repeated_attempts": {
+    "attempt_count_distribution": {
+      "1": 40296,
+      "10": 135,
+      "11": 58,
+      "117": 1,
+      "12": 112,
+      "13": 20,
+      "139": 1,
+      "14": 31,
+      "15": 6,
+      "16": 11,
+      "17": 2,
+      "18": 5,
+      "2": 22077,
+      "20": 6,
+      "21": 3,
+      "22": 2,
+      "23": 1,
+      "27": 2,
+      "28": 1,
+      "29": 2,
+      "3": 825,
+      "30": 278,
+      "31": 1,
+      "321": 1,
+      "345": 2,
+      "346": 1,
+      "358": 2,
+      "361": 1,
+      "362": 2,
+      "366": 2,
+      "369": 1,
+      "371": 1,
+      "372": 1,
+      "373": 2,
+      "375": 2,
+      "377": 4,
+      "378": 1,
+      "379": 2,
+      "380": 1,
+      "383": 2,
+      "384": 1,
+      "385": 1,
+      "386": 4,
+      "387": 8,
+      "388": 6,
+      "389": 2,
+      "39": 1,
+      "390": 1,
+      "391": 5,
+      "392": 8,
+      "393": 4,
+      "394": 5,
+      "395": 5,
+      "396": 2,
+      "397": 1,
+      "398": 3,
+      "399": 1,
+      "4": 1867,
+      "401": 1,
+      "403": 1,
+      "404": 1,
+      "405": 1,
+      "412": 1,
+      "46": 1,
+      "5": 593,
+      "6": 271,
+      "67": 1,
+      "7": 161,
+      "8": 217,
+      "9": 582
+    },
+    "empirical_success_at_k": [
+      {
+        "eligible_tasks": 48864,
+        "empirical_success_at_k": 0.4352079240340537,
+        "k": 1,
+        "ordering": "source record locator then trajectory ID; empirical, not estimated pass@k",
+        "solved_tasks": 21266
+      },
+      {
+        "eligible_tasks": 26897,
+        "empirical_success_at_k": 0.5153362828568242,
+        "k": 2,
+        "ordering": "source record locator then trajectory ID; empirical, not estimated pass@k",
+        "solved_tasks": 13861
+      },
+      {
+        "eligible_tasks": 5057,
+        "empirical_success_at_k": 0.4601542416452442,
+        "k": 3,
+        "ordering": "source record locator then trajectory ID; empirical, not estimated pass@k",
+        "solved_tasks": 2327
+      },
+      {
+        "eligible_tasks": 4313,
+        "empirical_success_at_k": 0.4817992116856017,
+        "k": 4,
+        "ordering": "source record locator then trajectory ID; empirical, not estimated pass@k",
+        "solved_tasks": 2078
+      },
+      {
+        "eligible_tasks": 2487,
+        "empirical_success_at_k": 0.5158825894652191,
+        "k": 5,
+        "ordering": "source record locator then trajectory ID; empirical, not estimated pass@k",
+        "solved_tasks": 1283
+      },
+      {
+        "eligible_tasks": 1919,
+        "empirical_success_at_k": 0.6242834809796769,
+        "k": 6,
+        "ordering": "source record locator then trajectory ID; empirical, not estimated pass@k",
+        "solved_tasks": 1198
+      },
+      {
+        "eligible_tasks": 1663,
+        "empirical_success_at_k": 0.6915213469633193,
+        "k": 7,
+        "ordering": "source record locator then trajectory ID; empirical, not estimated pass@k",
+        "solved_tasks": 1150
+      },
+      {
+        "eligible_tasks": 1511,
+        "empirical_success_at_k": 0.7346128391793514,
+        "k": 8,
+        "ordering": "source record locator then trajectory ID; empirical, not estimated pass@k",
+        "solved_tasks": 1110
+      },
+      {
+        "eligible_tasks": 1303,
+        "empirical_success_at_k": 0.7705295471987721,
+        "k": 9,
+        "ordering": "source record locator then trajectory ID; empirical, not estimated pass@k",
+        "solved_tasks": 1004
+      },
+      {
+        "eligible_tasks": 724,
+        "empirical_success_at_k": 0.649171270718232,
+        "k": 10,
+        "ordering": "source record locator then trajectory ID; empirical, not estimated pass@k",
+        "solved_tasks": 470
+      }
+    ],
+    "fraction_solved_at_least_once": 0.47888015717092336,
+    "tasks_solved_at_least_once": 23400,
+    "tasks_with_known_outcomes": 48864,
+    "tasks_with_multiple_trajectories": 27364
+  },
+  "schema_version": 1,
+  "structurally_eligible": 143999,
+  "training_eligible": 110208,
+  "unique_tasks": 67610
+}
diff --git a/examples.html b/examples.html
new file mode 100644
index 0000000000000000000000000000000000000000..d194efe6a0b8dcb8b20784a87a70dacdeae3e7b5
--- /dev/null
+++ b/examples.html
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b543762180caad13dcc65fced9398d7afbe791dee86c0d27d9772d0a6b7fda19
+size 14119755
diff --git a/examples.selection.json b/examples.selection.json
new file mode 100644
index 0000000000000000000000000000000000000000..3d5b87c5310ba4f52418cbf0ffd82f6f47b161ba
--- /dev/null
+++ b/examples.selection.json
@@ -0,0 +1,142 @@
+{
+  "comparison": {
+    "attempts": 412,
+    "environment_key": "Terminal-Bench@2.0",
+    "task_key": "terminalbench:count-dataset-tokens"
+  },
+  "records": [
+    {
+      "reason": "browsecomp_rollouts/success/tier-3",
+      "trajectory_id": "trajectory:f6afa2d095e1a4ebbe1de80b"
+    },
+    {
+      "reason": "browsecomp_rollouts/failure/tier-5",
+      "trajectory_id": "trajectory:aad1409d32b2ad5bfa3717fb"
+    },
+    {
+      "reason": "mind2web/unknown/tier-3",
+      "trajectory_id": "trajectory:37ff002c486f5bbc671c7f05"
+    },
+    {
+      "reason": "swebench_multilingual_trajectories/success/tier-3",
+      "trajectory_id": "trajectory:ceb8c403bec585635db5c885"
+    },
+    {
+      "reason": "swebench_multilingual_trajectories/failure/tier-5",
+      "trajectory_id": "trajectory:230dd0fadb9aa207896b2745"
+    },
+    {
+      "reason": "swegym_sampled/failure/tier-5",
+      "trajectory_id": "trajectory:17b01c2476ae7525876e3ca3"
+    },
+    {
+      "reason": "nvidia_open_swe_traces/success/tier-3",
+      "trajectory_id": "trajectory:dc2a7493e851fb4205dabf60"
+    },
+    {
+      "reason": "swebench_verified_trajectories/failure/tier-5",
+      "trajectory_id": "trajectory:1939c1be6fbe482a9fd74b4b"
+    },
+    {
+      "reason": "swegym_sampled/success/tier-3",
+      "trajectory_id": "trajectory:465fdac1a1b3856392aa77c7"
+    },
+    {
+      "reason": "nvidia_open_swe_traces/failure/tier-5",
+      "trajectory_id": "trajectory:803ce21a16aed837c0db037b"
+    },
+    {
+      "reason": "swegym_openhands/success/tier-3",
+      "trajectory_id": "trajectory:679220a4ab9e2685d09d8b8f"
+    },
+    {
+      "reason": "swebench_verified_trajectories/success/tier-5",
+      "trajectory_id": "trajectory:7cd76b774e3d1f674cd36ca9"
+    },
+    {
+      "reason": "swebench_verified_trajectories/success/tier-3",
+      "trajectory_id": "trajectory:c39124375e6ebd36c9f65512"
+    },
+    {
+      "reason": "swesmith_trajectories/failure/tier-5",
+      "trajectory_id": "trajectory:b506991f4199d7819fd6fe87"
+    },
+    {
+      "reason": "swesmith_trajectories/success/tier-3",
+      "trajectory_id": "trajectory:08cddd9336ac3a944e153a35"
+    },
+    {
+      "reason": "agentsuite_tau2/success/tier-3",
+      "trajectory_id": "trajectory:0cbb52ef305b68ca6a0e28b8"
+    },
+    {
+      "reason": "agentsuite_tau2/success/tier-5",
+      "trajectory_id": "trajectory:8f3680cf4f445e8511d91a17"
+    },
+    {
+      "reason": "agentsuite_tau2/failure/tier-5",
+      "trajectory_id": "trajectory:ab0d84b3dc737fca1610ac4d"
+    },
+    {
+      "reason": "official_tau3/failure/tier-5",
+      "trajectory_id": "trajectory:a2dc8e895192d065455cf275"
+    },
+    {
+      "reason": "official_tau3/success/tier-5",
+      "trajectory_id": "trajectory:7c7d376b6e7e99e06a878784"
+    },
+    {
+      "reason": "terminaltraj/unknown/tier-4",
+      "trajectory_id": "trajectory:049aa1b8a9751320c05978f9"
+    },
+    {
+      "reason": "terminaltraj/unknown/tier-5",
+      "trajectory_id": "trajectory:29a619875a0a331fc3fe3515"
+    },
+    {
+      "reason": "terminalbench_trajectories/success/tier-5",
+      "trajectory_id": "trajectory:42e80eb22663e48fd9b2b497"
+    },
+    {
+      "reason": "terminalbench_trajectories/failure/tier-5",
+      "trajectory_id": "trajectory:afd3c68daf9d09b9cd03d3fd"
+    },
+    {
+      "reason": "repeated-attempt-comparison",
+      "trajectory_id": "trajectory:01bf48ea0f6c3beb1162bf22"
+    },
+    {
+      "reason": "repeated-attempt-comparison",
+      "trajectory_id": "trajectory:0698279b8ea54e0e439e4e74"
+    },
+    {
+      "reason": "repeated-attempt-comparison",
+      "trajectory_id": "trajectory:0ea8a532668fdf63eadc25c7"
+    },
+    {
+      "reason": "repeated-attempt-comparison",
+      "trajectory_id": "trajectory:8969e9a90720972236678351"
+    },
+    {
+      "reason": "repeated-attempt-comparison",
+      "trajectory_id": "trajectory:d0ce4ac1d315c8c143c8a915"
+    },
+    {
+      "reason": "repeated-attempt-comparison",
+      "trajectory_id": "trajectory:e4618556d3f972e3974f7b5a"
+    },
+    {
+      "reason": "terminalbench_trajectories/success/tier-3",
+      "trajectory_id": "trajectory:abce4a0d6188f4510f4d0bc0"
+    },
+    {
+      "reason": "toolbench/success/tier-5",
+      "trajectory_id": "trajectory:5653af79ecc03fead3f540ff"
+    },
+    {
+      "reason": "toolbench/failure/tier-5",
+      "trajectory_id": "trajectory:e3784d724ae3e7dd55befd52"
+    }
+  ],
+  "selection_policy": "lexicographically first trajectory per source/outcome/tier plus the largest non-BFCL repeated-attempt group"
+}
diff --git a/examples/nemotron_sft_demo.json b/examples/nemotron_sft_demo.json
new file mode 100644
index 0000000000000000000000000000000000000000..9ee221a453b4292c7c34cc4ca99e38b62260d54b
--- /dev/null
+++ b/examples/nemotron_sft_demo.json
@@ -0,0 +1,191 @@
+{
+  "examples": [
+    {
+      "assistant_loss_tokens": 696,
+      "assistant_only_mask": true,
+      "ignored_context_tokens": 2478,
+      "input_ids_preview": [
+        10,
+        25708,
+        1010,
+        1035,
+        7466,
+        2839,
+        38114,
+        28093,
+        1595,
+        13404,
+        5010,
+        28036,
+        2794,
+        1595,
+        11,
+        1010
+      ],
+      "labels_preview": [
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100
+      ],
+      "sample_id": "full:736e63752d3cff92f0d883d1",
+      "token_length": 3174,
+      "view": "full_trajectory"
+    },
+    {
+      "assistant_loss_tokens": 60,
+      "assistant_only_mask": true,
+      "causal_context_events": 5,
+      "causal_context_tokens": 1817,
+      "ignored_context_tokens": 1817,
+      "input_ids_preview": [
+        10,
+        25708,
+        1010,
+        1035,
+        7466,
+        2839,
+        38114,
+        28093,
+        1561,
+        1885,
+        5165,
+        1561,
+        15,
+        1010,
+        11,
+        1010
+      ],
+      "labels_preview": [
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        -100,
+        1561,
+        1885,
+        5165,
+        1561,
+        15,
+        1010,
+        11,
+        1010
+      ],
+      "prior_assistant_tokens_masked_out": 80,
+      "sample_id": "turn:61eec58539653d88c1971b68",
+      "target_event_id": "event:f8253b7938b844a48ae3ab91",
+      "target_event_index": 5,
+      "token_length": 1877,
+      "view": "assistant_turn"
+    }
+  ],
+  "model_ready_batch": {
+    "all_padding_labels_ignored": true,
+    "fields": [
+      "input_ids",
+      "attention_mask",
+      "labels"
+    ],
+    "framework_neutral_python_batch": true,
+    "label_ignore_index": -100,
+    "non_padding_tokens": [
+      3174,
+      1877
+    ],
+    "right_padded": true,
+    "sha256": "d680ae7e4ea605b3db96370de6760d86ed227411dac901d74a511f72c3346ae8",
+    "shape": [
+      2,
+      3174
+    ],
+    "supervised_label_tokens": [
+      696,
+      60
+    ]
+  },
+  "real_trajectory": {
+    "assistant_target_count": 6,
+    "event_count": 13,
+    "outcome_success": true,
+    "preview": [
+      {
+        "content": "# Airline Agent Policy\n\nThe current time is 2024-05-15 15:00:00 EST.\n\nAs an airline agent, you can help users **book**, **modify**, or **cancel** flight reservations. You also handle **refunds and compensation**.\n\nBefore taking any actions ",
+        "role": "system",
+        "supervised": false,
+        "tool_calls": []
+      },
+      {
+        "content": "Hi! How can I help you today?",
+        "role": "assistant",
+        "supervised": true,
+        "tool_calls": []
+      },
+      {
+        "content": "Hi, I’d like to cancel my reservation with the code EHGLP3. Can you help me with that?",
+        "role": "user",
+        "supervised": false,
+        "tool_calls": []
+      },
+      {
+        "content": "I'd be happy to help you cancel your reservation. Before proceeding, I'll need a few pieces of information.\n\nCould you please provide your user ID? Also, may I ask what is the reason for cancellation (change of plan, airline cancelled fligh",
+        "role": "assistant",
+        "supervised": true,
+        "tool_calls": []
+      },
+      {
+        "content": "Sure, my user ID is emma_kim_9957. The reason is that my plans changed unexpectedly.",
+        "role": "user",
+        "supervised": false,
+        "tool_calls": []
+      },
+      {
+        "content": "Thank you for providing your user ID. Let me check the details of your reservation before proceeding with the cancellation.",
+        "role": "assistant",
+        "supervised": true,
+        "tool_calls": [
+          "get_reservation_details"
+        ]
+      },
+      {
+        "content": "{\"reservation_id\": \"EHGLP3\", \"user_id\": \"emma_kim_9957\", \"origin\": \"PHX\", \"destination\": \"JFK\", \"flight_type\": \"one_way\", \"cabin\": \"basic_economy\", \"flights\": [{\"flight_number\": \"HAT156\", \"origin\": \"PHX\", \"destination\": \"SEA\", \"date\": \"2024",
+        "role": "tool",
+        "supervised": false,
+        "tool_calls": []
+      }
+    ],
+    "record_locator": "data/cache/official_tau3/source.json#simulations[1]",
+    "source_id": "official_tau3",
+    "source_revision": "dd8757d4c8791dad7b3cdaf02a105a3d5f9d86bd",
+    "task_id": "0",
+    "task_key": "tau:airline:0",
+    "trajectory_id": "trajectory:f9cf8588a314fa9804d57402",
+    "verification_method": "tau environment reward"
+  },
+  "schema_version": 1,
+  "tokenizer": {
+    "assistant_mask_method": "pinned native chat template with Transformers generation spans around native assistant rendering",
+    "class": "TokenizersBackend",
+    "eos_token_id": 11,
+    "model": "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
+    "model_max_length": 262144,
+    "pad_token_id": 11,
+    "revision": "a9904d24bcc1d289a1950fa9d2b978c47cf903b9",
+    "truncate_history_thinking": false,
+    "vocabulary_size": 131072
+  }
+}
diff --git a/index.sqlite b/index.sqlite
new file mode 100644
index 0000000000000000000000000000000000000000..284c3a2c89c5903ba3cf2f12fbed7cbbb5aa7346
--- /dev/null
+++ b/index.sqlite
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25b97793c7b002995cdbea6daddb4e0068b81b40aaeb11179a2915f9810857fd
+size 178581504
diff --git a/manifest.json b/manifest.json
new file mode 100644
index 0000000000000000000000000000000000000000..08e33150b0b6a2d579d4235b3d57dc5a0f149107
--- /dev/null
+++ b/manifest.json
@@ -0,0 +1,1277 @@
+{
+  "content_complete": false,
+  "duplicate_provenance_records": 1208,
+  "duplicate_trajectory_ids_skipped": 0,
+  "evaluated": 155278,
+  "files": {
+    "all": null,
+    "duplicate_provenance": "duplicate_provenance.jsonl.gz",
+    "partition_indices": {
+      "structurally_eligible": "structurally_eligible.index.jsonl.gz",
+      "successful": "successful.index.jsonl.gz",
+      "training_eligible": "training_eligible.index.jsonl.gz",
+      "unknown": "unknown.index.jsonl.gz",
+      "unsuccessful": "unsuccessful.index.jsonl.gz"
+    }
+  },
+  "hosted_source_counts": {
+    "mind2web": 1009,
+    "nvidia_open_swe_traces": 51626,
+    "official_tau3": 200,
+    "swebench_multilingual_trajectories": 129,
+    "swebench_verified_trajectories": 4544,
+    "swegym_openhands": 491,
+    "swesmith_trajectories": 24100,
+    "terminalbench_trajectories": 34263,
+    "terminaltraj": 20000,
+    "toolbench": 11
+  },
+  "hosted_trajectories": 136373,
+  "identical_records_collapsed": 1208,
+  "inputs": [
+    {
+      "path": "canonical/mind2web/default--train--0000.jsonl.gz",
+      "records": 100,
+      "sha256": "654357077e6249cd030e516a74bc3ac7551e26553809581bbebb56a4f87e4c37",
+      "size_bytes": 60960835,
+      "source_counts": {
+        "mind2web": 100
+      }
+    },
+    {
+      "path": "canonical/mind2web/default--train--0001.jsonl.gz",
+      "records": 100,
+      "sha256": "ea62d59ed99fa6632c59bb62cf339054fe72e717c38d36109526414bb9ce2f60",
+      "size_bytes": 61759916,
+      "source_counts": {
+        "mind2web": 100
+      }
+    },
+    {
+      "path": "canonical/mind2web/default--train--0002.jsonl.gz",
+      "records": 109,
+      "sha256": "77d3d9b0ef99cf5ff8300c7e0681847780752ce60e2c2e0887e247b491b48e22",
+      "size_bytes": 61958505,
+      "source_counts": {
+        "mind2web": 109
+      }
+    },
+    {
+      "path": "canonical/mind2web/default--train--0003.jsonl.gz",
+      "records": 100,
+      "sha256": "f3807d46087383372f43ca0e3ed95d1b14e9ceca917940c652d136eca5157d7c",
+      "size_bytes": 64515683,
+      "source_counts": {
+        "mind2web": 100
+      }
+    },
+    {
+      "path": "canonical/mind2web/default--train--0004.jsonl.gz",
+      "records": 100,
+      "sha256": "f4e03e9b52f92aecac56c551d2a83e1af8fe26933e8366a26cfb90249c9b1f5f",
+      "size_bytes": 59759759,
+      "source_counts": {
+        "mind2web": 100
+      }
+    },
+    {
+      "path": "canonical/mind2web/default--train--0005.jsonl.gz",
+      "records": 100,
+      "sha256": "cd075e514c9006f2781cec84a0bbb816e1bc6ccdf22217885925e2ab5d02df8b",
+      "size_bytes": 59608293,
+      "source_counts": {
+        "mind2web": 100
+      }
+    },
+    {
+      "path": "canonical/mind2web/default--train--0006.jsonl.gz",
+      "records": 200,
+      "sha256": "fa7ab83febb58ab7a51ecdee07a03369ccc55b8b8ae811dd5679693f98a70c52",
+      "size_bytes": 115862382,
+      "source_counts": {
+        "mind2web": 200
+      }
+    },
+    {
+      "path": "canonical/mind2web/default--train--0007.jsonl.gz",
+      "records": 200,
+      "sha256": "aa20a9fa7022a87f523148839fbbccae8576eecd011eb8a8b7a12be2472c8e5e",
+      "size_bytes": 116176682,
+      "source_counts": {
+        "mind2web": 200
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0000.jsonl.gz",
+      "records": 2500,
+      "sha256": "22d874f857dd91bef6f539e5c5809b5f654bb72c310caad8d42cd1d552189c45",
+      "size_bytes": 303785042,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0001.jsonl.gz",
+      "records": 2500,
+      "sha256": "be0c9efbdfb9947734634f75102ad4e2be154b4ab63770a748ebcfb0a5677927",
+      "size_bytes": 295831492,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0002.jsonl.gz",
+      "records": 2500,
+      "sha256": "3067df96c93975319e290aa0b5f7b4be0ceecb6275f761cea33db9b6b3c54ffd",
+      "size_bytes": 305310880,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0003.jsonl.gz",
+      "records": 2500,
+      "sha256": "2b3324ce204a000df83ad9e207b4318af57ed19d36c11d7cbbae7bf5db90b453",
+      "size_bytes": 290056585,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0004.jsonl.gz",
+      "records": 2500,
+      "sha256": "5373e01f7f6527a0611ae933bb73b92d6e2ae312ea81d0d3aef85b1b241e55a7",
+      "size_bytes": 287437480,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0005.jsonl.gz",
+      "records": 41,
+      "sha256": "92b92e2aba80e79cbb07e9ba071db946fe4aa885deef4aefa9d934ae3bff193e",
+      "size_bytes": 5161644,
+      "source_counts": {
+        "nvidia_open_swe_traces": 41
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0006.jsonl.gz",
+      "records": 2500,
+      "sha256": "065ba7c18fbf854df437e352fe79cf54237f22d9290ef0b55283ff7f5c9e7e77",
+      "size_bytes": 368841972,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0007.jsonl.gz",
+      "records": 2500,
+      "sha256": "af1f7ed30b9037df06ebf7e7f0d146913ad813511e065e8f735a520b646b74bf",
+      "size_bytes": 371128455,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0008.jsonl.gz",
+      "records": 2500,
+      "sha256": "1b8247219e917730ef629eb0e66f2ec19d91ea404ea546c1ed10961eba7e0d30",
+      "size_bytes": 367495181,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0009.jsonl.gz",
+      "records": 2500,
+      "sha256": "0aec09bafc26fcd3254794c23b358cb597e94a492d8e0fd153650e859f5997be",
+      "size_bytes": 370761485,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0010.jsonl.gz",
+      "records": 2500,
+      "sha256": "2eff407239d58e2d9d525aa70402dcf01501ef3d3007d6407a69765c9c4936b5",
+      "size_bytes": 370224921,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0011.jsonl.gz",
+      "records": 558,
+      "sha256": "2a68e3249212dbf0661db6e7d7a5b9afc6a2a4bf0172d0f32e9ad0cbf272abbc",
+      "size_bytes": 83589148,
+      "source_counts": {
+        "nvidia_open_swe_traces": 558
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0012.jsonl.gz",
+      "records": 2500,
+      "sha256": "d54f3aebffe6bfdd1889488db25bf851210f000d6f6ca2fa2bebf83155fea40a",
+      "size_bytes": 243965767,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0013.jsonl.gz",
+      "records": 2500,
+      "sha256": "454932733cdc2ce5d6570d064a239ebbe497b0b1915c51f1ea1c3837edc5fae3",
+      "size_bytes": 241910439,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0014.jsonl.gz",
+      "records": 2500,
+      "sha256": "e964d62766afd638a8526c5e44fca9581e2cea0d3ec0ffd0a14beeefc1e16c6c",
+      "size_bytes": 249402952,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0015.jsonl.gz",
+      "records": 2500,
+      "sha256": "97fee226a8dc416dadb5673cace864585c36a933d42238abd640526814a5b2c5",
+      "size_bytes": 240747809,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0016.jsonl.gz",
+      "records": 2500,
+      "sha256": "95b32f5071fe69c1949e66623413b9e02970973ea501040336955b56d6de0b16",
+      "size_bytes": 242546761,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0017.jsonl.gz",
+      "records": 771,
+      "sha256": "2c17ad85f0ef74b315314fc6929e3efe7875f5d88f319208bc07fbbbc621c296",
+      "size_bytes": 74960935,
+      "source_counts": {
+        "nvidia_open_swe_traces": 771
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0018.jsonl.gz",
+      "records": 2500,
+      "sha256": "331b6e4d1611cc71c643b7a1af41744218d25aa5d6be33675f3ca2bcc20ab183",
+      "size_bytes": 373422381,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0019.jsonl.gz",
+      "records": 2500,
+      "sha256": "a42a718e33c3b46b9bc01867f57b2166a3ef6de0063e41d6dfb419bb55df971f",
+      "size_bytes": 371590204,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0020.jsonl.gz",
+      "records": 2500,
+      "sha256": "e412984c9e629a4621083e68f60fa1713ca1678f7eab46c8d0ee25813f967f90",
+      "size_bytes": 376298397,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0021.jsonl.gz",
+      "records": 2500,
+      "sha256": "22ae50d9032b0850fcf93080fcc81143d6af8f6ecad74c732c6d0428a6b4000c",
+      "size_bytes": 376587752,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0022.jsonl.gz",
+      "records": 2500,
+      "sha256": "379caa74f2bedf81325a98874fc804c40fd8741f7bf6571b66d72a4b905b6911",
+      "size_bytes": 380085424,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      }
+    },
+    {
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0023.jsonl.gz",
+      "records": 256,
+      "sha256": "863309aa10f41341676f0fd71e7e2d68db2d50b3d8451b5b8adab83ff44c515c",
+      "size_bytes": 38763239,
+      "source_counts": {
+        "nvidia_open_swe_traces": 256
+      }
+    },
+    {
+      "path": "canonical/official_tau3+toolbench/official_tau3_toolbench.jsonl.gz",
+      "records": 211,
+      "sha256": "0e08775a1a2e47543fb6c3ddd1789660ece1d9fe3bbf6189e457fc06f7bcc594",
+      "size_bytes": 2412172,
+      "source_counts": {
+        "official_tau3": 200,
+        "toolbench": 11
+      }
+    },
+    {
+      "path": "canonical/swebench_multilingual_trajectories/default--Multi_SWE_agent_32B_Rust--0000.jsonl.gz",
+      "records": 43,
+      "sha256": "0034093e5c0b717c4957b481d85c9381f9cba025ad16d1b0fcac9152c9698656",
+      "size_bytes": 17147283,
+      "source_counts": {
+        "swebench_multilingual_trajectories": 43
+      }
+    },
+    {
+      "path": "canonical/swebench_multilingual_trajectories/default--Qwen2.5_Coder_32B_Instruct--0001.jsonl.gz",
+      "records": 43,
+      "sha256": "146462000c0168746f0ca6fccec7f0cf859db13353ccd7504d8e264964eb4441",
+      "size_bytes": 36236356,
+      "source_counts": {
+        "swebench_multilingual_trajectories": 43
+      }
+    },
+    {
+      "path": "canonical/swebench_multilingual_trajectories/default--SWE_agent_LM_32B--0002.jsonl.gz",
+      "records": 43,
+      "sha256": "bf4d1265c6f2928fc1b8c188299fc5e46892bc3624ca377249c9855bf3e766fa",
+      "size_bytes": 18456716,
+      "source_counts": {
+        "swebench_multilingual_trajectories": 43
+      }
+    },
+    {
+      "path": "canonical/swebench_verified_trajectories/all.jsonl.gz",
+      "records": 5000,
+      "sha256": "7da41ea495e4f899daacf55989fcd5be6ef103c870f1f22e2aa11c0a14ce8982",
+      "size_bytes": 336638978,
+      "source_counts": {
+        "swebench_verified_trajectories": 5000
+      }
+    },
+    {
+      "path": "canonical/swegym_openhands/default--train.success.oss--0000.jsonl.gz",
+      "records": 491,
+      "sha256": "8d5a0fb7c17af0eefdf7ed0748ded69aef38fab9e9c6d661bf384915847777f9",
+      "size_bytes": 13990587,
+      "source_counts": {
+        "swegym_openhands": 491
+      }
+    },
+    {
+      "path": "canonical/swesmith_trajectories/default--tool--0000.jsonl.gz",
+      "records": 3013,
+      "sha256": "909cd6654ba6b253bf83874168fcfef57eb7ae700d94d39d4e10da15e2d4df22",
+      "size_bytes": 208736286,
+      "source_counts": {
+        "swesmith_trajectories": 3013
+      }
+    },
+    {
+      "path": "canonical/swesmith_trajectories/default--tool--0001.jsonl.gz",
+      "records": 3013,
+      "sha256": "8efdc46253958b9bfdba85da004f1de435709881773472d627a21372f197ba78",
+      "size_bytes": 128599828,
+      "source_counts": {
+        "swesmith_trajectories": 3013
+      }
+    },
+    {
+      "path": "canonical/swesmith_trajectories/default--tool--0002.jsonl.gz",
+      "records": 3013,
+      "sha256": "80d7a0105bee1b9cde168524b9cd702005786014464a3d5adcb40dcb7a0603ff",
+      "size_bytes": 115440687,
+      "source_counts": {
+        "swesmith_trajectories": 3013
+      }
+    },
+    {
+      "path": "canonical/swesmith_trajectories/default--tool--0003.jsonl.gz",
+      "records": 3013,
+      "sha256": "48df7f6939fd5619d1376c3e6909cfb71d88db56c11ce0511989b51620e44a9c",
+      "size_bytes": 147332694,
+      "source_counts": {
+        "swesmith_trajectories": 3013
+      }
+    },
+    {
+      "path": "canonical/swesmith_trajectories/default--tool--0004.jsonl.gz",
+      "records": 3012,
+      "sha256": "da2007c58f68ec2de3736cf7dbb6b1356e1d1657bfe8ef5f03ecb77b03a7e227",
+      "size_bytes": 93517927,
+      "source_counts": {
+        "swesmith_trajectories": 3012
+      }
+    },
+    {
+      "path": "canonical/swesmith_trajectories/default--tool--0005.jsonl.gz",
+      "records": 3012,
+      "sha256": "97e66d5f2c7e74a53d2d56528d9972ad3ad455ba7472900f8ba474e7bfecba9a",
+      "size_bytes": 147229136,
+      "source_counts": {
+        "swesmith_trajectories": 3012
+      }
+    },
+    {
+      "path": "canonical/swesmith_trajectories/default--tool--0006.jsonl.gz",
+      "records": 3012,
+      "sha256": "fcf23d8b7b0f907ce65e99fd25983b479e099e69506eaeef4c6794f6a663b1cd",
+      "size_bytes": 162574019,
+      "source_counts": {
+        "swesmith_trajectories": 3012
+      }
+    },
+    {
+      "path": "canonical/swesmith_trajectories/default--tool--0007.jsonl.gz",
+      "records": 3012,
+      "sha256": "ff5c4100eb4fa6610b91012d93d071989a7cb33510f9b5e273e19e41a05d7b39",
+      "size_bytes": 141721185,
+      "source_counts": {
+        "swesmith_trajectories": 3012
+      }
+    },
+    {
+      "path": "canonical/terminalbench_trajectories/default--train--0000.jsonl.gz",
+      "records": 17432,
+      "sha256": "ab68edea96136704624b82ec2b8de28bf07135b8cb93da708e6f199ee873e06c",
+      "size_bytes": 167147418,
+      "source_counts": {
+        "terminalbench_trajectories": 17432
+      }
+    },
+    {
+      "path": "canonical/terminalbench_trajectories/default--train--0001.jsonl.gz",
+      "records": 17030,
+      "sha256": "8cacfd4217b7cdfd2e53e481d8a8845a73925f4f44759da29e2c6a550f4f9e63",
+      "size_bytes": 154916466,
+      "source_counts": {
+        "terminalbench_trajectories": 17030
+      }
+    },
+    {
+      "path": "canonical/terminaltraj/default--train--0000.jsonl.gz",
+      "records": 10000,
+      "sha256": "6e17b247614ddfd84792e72de91ad2a5217f2f86086cb05afbcf01a9cfd9e5d3",
+      "size_bytes": 89960134,
+      "source_counts": {
+        "terminaltraj": 10000
+      }
+    },
+    {
+      "path": "canonical/terminaltraj/default--train--0001.jsonl.gz",
+      "records": 10000,
+      "sha256": "f01da77f7ac09de890a7d8b0cb208c3ddc23ef5f2fe125b0d5237bbf1f4b120f",
+      "size_bytes": 91844522,
+      "source_counts": {
+        "terminaltraj": 10000
+      }
+    }
+  ],
+  "inventory_inputs": [
+    {
+      "canonical_sha256": "c17428d18c66788e5bbb8c705467f7df49bfa3807c18479c788d3b784ae8cc23",
+      "content_availability": "metadata_only",
+      "path": null,
+      "records": 8340,
+      "size_bytes": 116362991,
+      "source_counts": {
+        "agentsuite_tau2": 8340
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/AgentSuite/tau2-bench-trajectories/tree/07264f1d253a618fc20966670eddc44246212916"
+      ]
+    },
+    {
+      "canonical_sha256": "4329332610ac06539794aa2781b452c95ffef9a5b633d82c97e083b1ef1961d6",
+      "content_availability": "metadata_only",
+      "path": null,
+      "records": 2531,
+      "size_bytes": 217366029,
+      "source_counts": {
+        "browsecomp_rollouts": 2531
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/rl-rag/browsecomp-gpt-oss-120b-260222/tree/ac4854be0eb527cdf90f6c7d09446eeb7a3d1dbd"
+      ]
+    },
+    {
+      "canonical_sha256": "a4ba3ccfa48f233d4be820e50a3c1d0c5e8c0213c5af4c6a4dd7f5eea30b2753",
+      "content_availability": "metadata_only",
+      "path": null,
+      "records": 2532,
+      "size_bytes": 223325709,
+      "source_counts": {
+        "browsecomp_rollouts": 2532
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/rl-rag/browsecomp-gpt-oss-120b-260222/tree/ac4854be0eb527cdf90f6c7d09446eeb7a3d1dbd"
+      ]
+    },
+    {
+      "canonical_sha256": "654357077e6249cd030e516a74bc3ac7551e26553809581bbebb56a4f87e4c37",
+      "content_availability": "hosted",
+      "path": "canonical/mind2web/default--train--0000.jsonl.gz",
+      "records": 100,
+      "size_bytes": 60960835,
+      "source_counts": {
+        "mind2web": 100
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/osunlp/Mind2Web/tree/17ece8eb89862368edc0cc806acee6fca5163474"
+      ]
+    },
+    {
+      "canonical_sha256": "ea62d59ed99fa6632c59bb62cf339054fe72e717c38d36109526414bb9ce2f60",
+      "content_availability": "hosted",
+      "path": "canonical/mind2web/default--train--0001.jsonl.gz",
+      "records": 100,
+      "size_bytes": 61759916,
+      "source_counts": {
+        "mind2web": 100
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/osunlp/Mind2Web/tree/17ece8eb89862368edc0cc806acee6fca5163474"
+      ]
+    },
+    {
+      "canonical_sha256": "77d3d9b0ef99cf5ff8300c7e0681847780752ce60e2c2e0887e247b491b48e22",
+      "content_availability": "hosted",
+      "path": "canonical/mind2web/default--train--0002.jsonl.gz",
+      "records": 109,
+      "size_bytes": 61958505,
+      "source_counts": {
+        "mind2web": 109
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/osunlp/Mind2Web/tree/17ece8eb89862368edc0cc806acee6fca5163474"
+      ]
+    },
+    {
+      "canonical_sha256": "f3807d46087383372f43ca0e3ed95d1b14e9ceca917940c652d136eca5157d7c",
+      "content_availability": "hosted",
+      "path": "canonical/mind2web/default--train--0003.jsonl.gz",
+      "records": 100,
+      "size_bytes": 64515683,
+      "source_counts": {
+        "mind2web": 100
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/osunlp/Mind2Web/tree/17ece8eb89862368edc0cc806acee6fca5163474"
+      ]
+    },
+    {
+      "canonical_sha256": "f4e03e9b52f92aecac56c551d2a83e1af8fe26933e8366a26cfb90249c9b1f5f",
+      "content_availability": "hosted",
+      "path": "canonical/mind2web/default--train--0004.jsonl.gz",
+      "records": 100,
+      "size_bytes": 59759759,
+      "source_counts": {
+        "mind2web": 100
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/osunlp/Mind2Web/tree/17ece8eb89862368edc0cc806acee6fca5163474"
+      ]
+    },
+    {
+      "canonical_sha256": "cd075e514c9006f2781cec84a0bbb816e1bc6ccdf22217885925e2ab5d02df8b",
+      "content_availability": "hosted",
+      "path": "canonical/mind2web/default--train--0005.jsonl.gz",
+      "records": 100,
+      "size_bytes": 59608293,
+      "source_counts": {
+        "mind2web": 100
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/osunlp/Mind2Web/tree/17ece8eb89862368edc0cc806acee6fca5163474"
+      ]
+    },
+    {
+      "canonical_sha256": "fa7ab83febb58ab7a51ecdee07a03369ccc55b8b8ae811dd5679693f98a70c52",
+      "content_availability": "hosted",
+      "path": "canonical/mind2web/default--train--0006.jsonl.gz",
+      "records": 200,
+      "size_bytes": 115862382,
+      "source_counts": {
+        "mind2web": 200
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/osunlp/Mind2Web/tree/17ece8eb89862368edc0cc806acee6fca5163474"
+      ]
+    },
+    {
+      "canonical_sha256": "aa20a9fa7022a87f523148839fbbccae8576eecd011eb8a8b7a12be2472c8e5e",
+      "content_availability": "hosted",
+      "path": "canonical/mind2web/default--train--0007.jsonl.gz",
+      "records": 200,
+      "size_bytes": 116176682,
+      "source_counts": {
+        "mind2web": 200
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/osunlp/Mind2Web/tree/17ece8eb89862368edc0cc806acee6fca5163474"
+      ]
+    },
+    {
+      "canonical_sha256": "22d874f857dd91bef6f539e5c5809b5f654bb72c310caad8d42cd1d552189c45",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0000.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 303785042,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "be0c9efbdfb9947734634f75102ad4e2be154b4ab63770a748ebcfb0a5677927",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0001.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 295831492,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "3067df96c93975319e290aa0b5f7b4be0ceecb6275f761cea33db9b6b3c54ffd",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0002.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 305310880,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "2b3324ce204a000df83ad9e207b4318af57ed19d36c11d7cbbae7bf5db90b453",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0003.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 290056585,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "5373e01f7f6527a0611ae933bb73b92d6e2ae312ea81d0d3aef85b1b241e55a7",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0004.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 287437480,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "92b92e2aba80e79cbb07e9ba071db946fe4aa885deef4aefa9d934ae3bff193e",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0005.jsonl.gz",
+      "records": 41,
+      "size_bytes": 5161644,
+      "source_counts": {
+        "nvidia_open_swe_traces": 41
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "065ba7c18fbf854df437e352fe79cf54237f22d9290ef0b55283ff7f5c9e7e77",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0006.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 368841972,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "af1f7ed30b9037df06ebf7e7f0d146913ad813511e065e8f735a520b646b74bf",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0007.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 371128455,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "1b8247219e917730ef629eb0e66f2ec19d91ea404ea546c1ed10961eba7e0d30",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0008.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 367495181,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "0aec09bafc26fcd3254794c23b358cb597e94a492d8e0fd153650e859f5997be",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0009.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 370761485,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "2eff407239d58e2d9d525aa70402dcf01501ef3d3007d6407a69765c9c4936b5",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0010.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 370224921,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "2a68e3249212dbf0661db6e7d7a5b9afc6a2a4bf0172d0f32e9ad0cbf272abbc",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0011.jsonl.gz",
+      "records": 558,
+      "size_bytes": 83589148,
+      "source_counts": {
+        "nvidia_open_swe_traces": 558
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "d54f3aebffe6bfdd1889488db25bf851210f000d6f6ca2fa2bebf83155fea40a",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0012.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 243965767,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "454932733cdc2ce5d6570d064a239ebbe497b0b1915c51f1ea1c3837edc5fae3",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0013.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 241910439,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "e964d62766afd638a8526c5e44fca9581e2cea0d3ec0ffd0a14beeefc1e16c6c",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0014.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 249402952,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "97fee226a8dc416dadb5673cace864585c36a933d42238abd640526814a5b2c5",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0015.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 240747809,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "95b32f5071fe69c1949e66623413b9e02970973ea501040336955b56d6de0b16",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0016.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 242546761,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "2c17ad85f0ef74b315314fc6929e3efe7875f5d88f319208bc07fbbbc621c296",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0017.jsonl.gz",
+      "records": 771,
+      "size_bytes": 74960935,
+      "source_counts": {
+        "nvidia_open_swe_traces": 771
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "331b6e4d1611cc71c643b7a1af41744218d25aa5d6be33675f3ca2bcc20ab183",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0018.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 373422381,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "a42a718e33c3b46b9bc01867f57b2166a3ef6de0063e41d6dfb419bb55df971f",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0019.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 371590204,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "e412984c9e629a4621083e68f60fa1713ca1678f7eab46c8d0ee25813f967f90",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0020.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 376298397,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "22ae50d9032b0850fcf93080fcc81143d6af8f6ecad74c732c6d0428a6b4000c",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0021.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 376587752,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "379caa74f2bedf81325a98874fc804c40fd8741f7bf6571b66d72a4b905b6911",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0022.jsonl.gz",
+      "records": 2500,
+      "size_bytes": 380085424,
+      "source_counts": {
+        "nvidia_open_swe_traces": 2500
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "863309aa10f41341676f0fd71e7e2d68db2d50b3d8451b5b8adab83ff44c515c",
+      "content_availability": "hosted",
+      "path": "canonical/nvidia_open_swe_traces/v1.2--openhands--0023.jsonl.gz",
+      "records": 256,
+      "size_bytes": 38763239,
+      "source_counts": {
+        "nvidia_open_swe_traces": 256
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586"
+      ]
+    },
+    {
+      "canonical_sha256": "0e08775a1a2e47543fb6c3ddd1789660ece1d9fe3bbf6189e457fc06f7bcc594",
+      "content_availability": "hosted",
+      "path": "canonical/official_tau3+toolbench/official_tau3_toolbench.jsonl.gz",
+      "records": 211,
+      "size_bytes": 2412172,
+      "source_counts": {
+        "official_tau3": 200,
+        "toolbench": 11
+      },
+      "upstream": [
+        "https://github.com/sierra-research/tau2-bench/tree/dd8757d4c8791dad7b3cdaf02a105a3d5f9d86bd",
+        "https://github.com/OpenBMB/ToolBench/tree/d56fdd89faf8c91fa135090b212bb9057ee5cfc2"
+      ]
+    },
+    {
+      "canonical_sha256": "0034093e5c0b717c4957b481d85c9381f9cba025ad16d1b0fcac9152c9698656",
+      "content_availability": "hosted",
+      "path": "canonical/swebench_multilingual_trajectories/default--Multi_SWE_agent_32B_Rust--0000.jsonl.gz",
+      "records": 43,
+      "size_bytes": 17147283,
+      "source_counts": {
+        "swebench_multilingual_trajectories": 43
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/AlienKevin/SWE-bench-Multilingual-trajectories/tree/7221c09c6cc0ed16d51c9cb9886a1ab484465670"
+      ]
+    },
+    {
+      "canonical_sha256": "146462000c0168746f0ca6fccec7f0cf859db13353ccd7504d8e264964eb4441",
+      "content_availability": "hosted",
+      "path": "canonical/swebench_multilingual_trajectories/default--Qwen2.5_Coder_32B_Instruct--0001.jsonl.gz",
+      "records": 43,
+      "size_bytes": 36236356,
+      "source_counts": {
+        "swebench_multilingual_trajectories": 43
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/AlienKevin/SWE-bench-Multilingual-trajectories/tree/7221c09c6cc0ed16d51c9cb9886a1ab484465670"
+      ]
+    },
+    {
+      "canonical_sha256": "bf4d1265c6f2928fc1b8c188299fc5e46892bc3624ca377249c9855bf3e766fa",
+      "content_availability": "hosted",
+      "path": "canonical/swebench_multilingual_trajectories/default--SWE_agent_LM_32B--0002.jsonl.gz",
+      "records": 43,
+      "size_bytes": 18456716,
+      "source_counts": {
+        "swebench_multilingual_trajectories": 43
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/AlienKevin/SWE-bench-Multilingual-trajectories/tree/7221c09c6cc0ed16d51c9cb9886a1ab484465670"
+      ]
+    },
+    {
+      "canonical_sha256": "7da41ea495e4f899daacf55989fcd5be6ef103c870f1f22e2aa11c0a14ce8982",
+      "content_availability": "hosted",
+      "path": "canonical/swebench_verified_trajectories/all.jsonl.gz",
+      "records": 5000,
+      "size_bytes": 336638978,
+      "source_counts": {
+        "swebench_verified_trajectories": 5000
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/tarsur385/swebench-verified-trajectories/tree/773748a7c1222e8a642a7059821498e14293562a"
+      ]
+    },
+    {
+      "canonical_sha256": "8d5a0fb7c17af0eefdf7ed0748ded69aef38fab9e9c6d661bf384915847777f9",
+      "content_availability": "hosted",
+      "path": "canonical/swegym_openhands/default--train.success.oss--0000.jsonl.gz",
+      "records": 491,
+      "size_bytes": 13990587,
+      "source_counts": {
+        "swegym_openhands": 491
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/SWE-Gym/OpenHands-SFT-Trajectories/tree/4aaa5a4a4b5861f4799d2336908760c190ac3b17"
+      ]
+    },
+    {
+      "canonical_sha256": "7a13d09314190f2caefd8b6ccfac49a74c81cb3ff444a62d5b6a38cd89300772",
+      "content_availability": "metadata_only",
+      "path": null,
+      "records": 2019,
+      "size_bytes": 65321488,
+      "source_counts": {
+        "swegym_sampled": 2019
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/SWE-Gym/OpenHands-Sampled-Trajectories/tree/baf3a4e4bff514d48ddc08a93a2ade5c126212c7"
+      ]
+    },
+    {
+      "canonical_sha256": "a2addae2568f38226261c33bd83106899f1e27e5e87a0cea0b832bad29ec659e",
+      "content_availability": "metadata_only",
+      "path": null,
+      "records": 2018,
+      "size_bytes": 67493699,
+      "source_counts": {
+        "swegym_sampled": 2018
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/SWE-Gym/OpenHands-Sampled-Trajectories/tree/baf3a4e4bff514d48ddc08a93a2ade5c126212c7"
+      ]
+    },
+    {
+      "canonical_sha256": "a472b24ce0f1d557c42c07f20a136c04ec160d77fbf342a0125daf4ff5da25ca",
+      "content_availability": "metadata_only",
+      "path": null,
+      "records": 2018,
+      "size_bytes": 117810358,
+      "source_counts": {
+        "swegym_sampled": 2018
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/SWE-Gym/OpenHands-Sampled-Trajectories/tree/baf3a4e4bff514d48ddc08a93a2ade5c126212c7"
+      ]
+    },
+    {
+      "canonical_sha256": "909cd6654ba6b253bf83874168fcfef57eb7ae700d94d39d4e10da15e2d4df22",
+      "content_availability": "hosted",
+      "path": "canonical/swesmith_trajectories/default--tool--0000.jsonl.gz",
+      "records": 3013,
+      "size_bytes": 208736286,
+      "source_counts": {
+        "swesmith_trajectories": 3013
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories/tree/08e109b4a59eaeebf80e4675cd125d42e7ac99a4"
+      ]
+    },
+    {
+      "canonical_sha256": "8efdc46253958b9bfdba85da004f1de435709881773472d627a21372f197ba78",
+      "content_availability": "hosted",
+      "path": "canonical/swesmith_trajectories/default--tool--0001.jsonl.gz",
+      "records": 3013,
+      "size_bytes": 128599828,
+      "source_counts": {
+        "swesmith_trajectories": 3013
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories/tree/08e109b4a59eaeebf80e4675cd125d42e7ac99a4"
+      ]
+    },
+    {
+      "canonical_sha256": "80d7a0105bee1b9cde168524b9cd702005786014464a3d5adcb40dcb7a0603ff",
+      "content_availability": "hosted",
+      "path": "canonical/swesmith_trajectories/default--tool--0002.jsonl.gz",
+      "records": 3013,
+      "size_bytes": 115440687,
+      "source_counts": {
+        "swesmith_trajectories": 3013
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories/tree/08e109b4a59eaeebf80e4675cd125d42e7ac99a4"
+      ]
+    },
+    {
+      "canonical_sha256": "48df7f6939fd5619d1376c3e6909cfb71d88db56c11ce0511989b51620e44a9c",
+      "content_availability": "hosted",
+      "path": "canonical/swesmith_trajectories/default--tool--0003.jsonl.gz",
+      "records": 3013,
+      "size_bytes": 147332694,
+      "source_counts": {
+        "swesmith_trajectories": 3013
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories/tree/08e109b4a59eaeebf80e4675cd125d42e7ac99a4"
+      ]
+    },
+    {
+      "canonical_sha256": "da2007c58f68ec2de3736cf7dbb6b1356e1d1657bfe8ef5f03ecb77b03a7e227",
+      "content_availability": "hosted",
+      "path": "canonical/swesmith_trajectories/default--tool--0004.jsonl.gz",
+      "records": 3012,
+      "size_bytes": 93517927,
+      "source_counts": {
+        "swesmith_trajectories": 3012
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories/tree/08e109b4a59eaeebf80e4675cd125d42e7ac99a4"
+      ]
+    },
+    {
+      "canonical_sha256": "97e66d5f2c7e74a53d2d56528d9972ad3ad455ba7472900f8ba474e7bfecba9a",
+      "content_availability": "hosted",
+      "path": "canonical/swesmith_trajectories/default--tool--0005.jsonl.gz",
+      "records": 3012,
+      "size_bytes": 147229136,
+      "source_counts": {
+        "swesmith_trajectories": 3012
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories/tree/08e109b4a59eaeebf80e4675cd125d42e7ac99a4"
+      ]
+    },
+    {
+      "canonical_sha256": "fcf23d8b7b0f907ce65e99fd25983b479e099e69506eaeef4c6794f6a663b1cd",
+      "content_availability": "hosted",
+      "path": "canonical/swesmith_trajectories/default--tool--0006.jsonl.gz",
+      "records": 3012,
+      "size_bytes": 162574019,
+      "source_counts": {
+        "swesmith_trajectories": 3012
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories/tree/08e109b4a59eaeebf80e4675cd125d42e7ac99a4"
+      ]
+    },
+    {
+      "canonical_sha256": "ff5c4100eb4fa6610b91012d93d071989a7cb33510f9b5e273e19e41a05d7b39",
+      "content_availability": "hosted",
+      "path": "canonical/swesmith_trajectories/default--tool--0007.jsonl.gz",
+      "records": 3012,
+      "size_bytes": 141721185,
+      "source_counts": {
+        "swesmith_trajectories": 3012
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories/tree/08e109b4a59eaeebf80e4675cd125d42e7ac99a4"
+      ]
+    },
+    {
+      "canonical_sha256": "ab68edea96136704624b82ec2b8de28bf07135b8cb93da708e6f199ee873e06c",
+      "content_availability": "hosted",
+      "path": "canonical/terminalbench_trajectories/default--train--0000.jsonl.gz",
+      "records": 17432,
+      "size_bytes": 167147418,
+      "source_counts": {
+        "terminalbench_trajectories": 17432
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/yoonholee/terminalbench-trajectories/tree/04e8940f5b6736a7ce8d22224fe2f2af74163ed2"
+      ]
+    },
+    {
+      "canonical_sha256": "8cacfd4217b7cdfd2e53e481d8a8845a73925f4f44759da29e2c6a550f4f9e63",
+      "content_availability": "hosted",
+      "path": "canonical/terminalbench_trajectories/default--train--0001.jsonl.gz",
+      "records": 17030,
+      "size_bytes": 154916466,
+      "source_counts": {
+        "terminalbench_trajectories": 17030
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/yoonholee/terminalbench-trajectories/tree/04e8940f5b6736a7ce8d22224fe2f2af74163ed2"
+      ]
+    },
+    {
+      "canonical_sha256": "6e17b247614ddfd84792e72de91ad2a5217f2f86086cb05afbcf01a9cfd9e5d3",
+      "content_availability": "hosted",
+      "path": "canonical/terminaltraj/default--train--0000.jsonl.gz",
+      "records": 10000,
+      "size_bytes": 89960134,
+      "source_counts": {
+        "terminaltraj": 10000
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/m-a-p/TerminalTraj/tree/5c1823f4a8b9ca0cf02c27d2db52c5b35b53a308"
+      ]
+    },
+    {
+      "canonical_sha256": "f01da77f7ac09de890a7d8b0cb208c3ddc23ef5f2fe125b0d5237bbf1f4b120f",
+      "content_availability": "hosted",
+      "path": "canonical/terminaltraj/default--train--0001.jsonl.gz",
+      "records": 10000,
+      "size_bytes": 91844522,
+      "source_counts": {
+        "terminaltraj": 10000
+      },
+      "upstream": [
+        "https://huggingface.co/datasets/m-a-p/TerminalTraj/tree/5c1823f4a8b9ca0cf02c27d2db52c5b35b53a308"
+      ]
+    }
+  ],
+  "metadata_only_source_counts": {
+    "agentsuite_tau2": 8340,
+    "browsecomp_rollouts": 5063,
+    "swegym_sampled": 5502
+  },
+  "metadata_only_trajectories": 18905,
+  "outcome_counts": {
+    "failure": 79973,
+    "success": 54296,
+    "unknown": 21009
+  },
+  "partition_counts": {
+    "structurally_eligible": 143999,
+    "successful": 54296,
+    "training_eligible": 110208,
+    "unknown": 21009,
+    "unsuccessful": 79973
+  },
+  "permissions": "registry/permissions.json",
+  "ranked": 155278,
+  "schema_version": 1,
+  "source_counts": {
+    "agentsuite_tau2": 8340,
+    "browsecomp_rollouts": 5063,
+    "mind2web": 1009,
+    "nvidia_open_swe_traces": 51626,
+    "official_tau3": 200,
+    "swebench_multilingual_trajectories": 129,
+    "swebench_verified_trajectories": 4544,
+    "swegym_openhands": 491,
+    "swegym_sampled": 5502,
+    "swesmith_trajectories": 24100,
+    "terminalbench_trajectories": 34263,
+    "terminaltraj": 20000,
+    "toolbench": 11
+  },
+  "source_record_counts": {
+    "agentsuite_tau2": 8340,
+    "browsecomp_rollouts": 5063,
+    "mind2web": 1009,
+    "nvidia_open_swe_traces": 51626,
+    "official_tau3": 200,
+    "swebench_multilingual_trajectories": 129,
+    "swebench_verified_trajectories": 5000,
+    "swegym_openhands": 491,
+    "swegym_sampled": 6055,
+    "swesmith_trajectories": 24100,
+    "terminalbench_trajectories": 34462,
+    "terminaltraj": 20000,
+    "toolbench": 11
+  },
+  "source_records": 156486,
+  "structurally_eligible": 143999,
+  "task_split_counts": {
+    "protected_evaluation": 1948,
+    "test": 6442,
+    "train": 52592,
+    "validation": 6628
+  },
+  "training_eligible": 110208,
+  "trajectories": 155278
+}
diff --git a/rankings.jsonl.gz b/rankings.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..094efb07ddafe599817c91fbeb3a88c4ec144240
--- /dev/null
+++ b/rankings.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a56a9ae57c860130af600fb534afaaa337124c15646573ec9f8dc86b950fa41
+size 10959026
diff --git a/registry/permissions.json b/registry/permissions.json
new file mode 100644
index 0000000000000000000000000000000000000000..8d54f421559d57c27880dd1f766af40f1396d5ab
--- /dev/null
+++ b/registry/permissions.json
@@ -0,0 +1,215 @@
+{
+  "schema_version": 1,
+  "audit_date": "2026-10-14",
+  "vocabulary": {
+    "permission": ["permitted", "permitted_with_conditions", "uncertain", "restricted"],
+    "verification": ["verified", "inferred", "uncertain"],
+    "public_payload": ["hosted", "metadata_only"]
+  },
+  "scope": "Source-level engineering audit for the 13 sources represented in the 155,278-record canonical inventory. Permissions apply to the identified upstream revision and can differ for embedded repository code, web content, model outputs, or later releases.",
+  "policy": "Keep the complete inventory and derived metadata. Host canonical payload only when redistribution is supported by source-specific evidence; otherwise publish metadata, hashes, pinned upstream locators, and authorized-overlay instructions.",
+  "disclaimer": "This is an evidence-based engineering audit, not legal advice. Downstream users remain responsible for their use case and third-party rights.",
+  "sources": [
+    {
+      "source_id": "swegym_openhands",
+      "upstream": "https://huggingface.co/datasets/SWE-Gym/OpenHands-SFT-Trajectories/tree/4aaa5a4a4b5861f4799d2336908760c190ac3b17",
+      "declared_license": "MIT",
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted", "verification": "verified"},
+      "model_training": {"permission": "permitted", "verification": "verified", "conditions": ["retain MIT notice"]},
+      "commercial_training": {"permission": "permitted", "verification": "verified", "conditions": ["retain MIT notice"]},
+      "redistribution_original": {"permission": "permitted", "verification": "verified", "conditions": ["retain MIT notice"]},
+      "redistribution_transformed": {"permission": "permitted", "verification": "verified", "conditions": ["retain MIT notice and source attribution"]},
+      "public_payload": "hosted",
+      "evidence": [
+        {"finding": "Pinned dataset card declares MIT.", "url": "https://huggingface.co/datasets/SWE-Gym/OpenHands-SFT-Trajectories/blob/4aaa5a4a4b5861f4799d2336908760c190ac3b17/README.md"},
+        {"finding": "The SWE-Gym project explicitly documents this dataset for model training.", "url": "https://github.com/SWE-Gym/SWE-Gym/blob/b681068ca20628c6987b7416cc4cf03f06b77ba5/docs/OpenHands.md"}
+      ]
+    },
+    {
+      "source_id": "nvidia_open_swe_traces",
+      "upstream": "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/tree/6d0a83373ae0d83abe3d363f1f18cb1822e16586",
+      "declared_license": "CC-BY-4.0 plus per-repository permissive licenses",
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted", "verification": "verified", "conditions": ["provide attribution"]},
+      "model_training": {"permission": "permitted", "verification": "verified", "conditions": ["provide attribution", "honor the per-record repository license"]},
+      "commercial_training": {"permission": "permitted", "verification": "verified", "conditions": ["provide attribution", "honor the per-record repository license"]},
+      "redistribution_original": {"permission": "permitted", "verification": "verified", "conditions": ["CC-BY-4.0 attribution", "retain per-record repository license metadata"]},
+      "redistribution_transformed": {"permission": "permitted", "verification": "verified", "conditions": ["CC-BY-4.0 attribution", "identify canonical transformations", "retain per-record repository license metadata"]},
+      "public_payload": "hosted",
+      "evidence": [
+        {"finding": "Card declares CC-BY-4.0, commercial/non-commercial readiness, permissive task repositories, and SFT/distillation as intended uses.", "url": "https://huggingface.co/datasets/nvidia/Open-SWE-Traces/blob/6d0a83373ae0d83abe3d363f1f18cb1822e16586/README.md"}
+      ]
+    },
+    {
+      "source_id": "terminaltraj",
+      "upstream": "https://huggingface.co/datasets/m-a-p/TerminalTraj/tree/5c1823f4a8b9ca0cf02c27d2db52c5b35b53a308",
+      "declared_license": "Apache-2.0 project license; dataset card omits a license field",
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted", "verification": "verified", "conditions": ["honor embedded repository licenses"]},
+      "model_training": {"permission": "permitted", "verification": "verified", "conditions": ["honor embedded repository licenses"]},
+      "commercial_training": {"permission": "permitted_with_conditions", "verification": "inferred", "conditions": ["confirm embedded repository obligations for the selected records"]},
+      "redistribution_original": {"permission": "permitted_with_conditions", "verification": "inferred", "conditions": ["Apache-2.0 notice", "honor embedded MIT/Apache-2.0/BSD-2-Clause/BSD-3-Clause repository terms"]},
+      "redistribution_transformed": {"permission": "permitted_with_conditions", "verification": "inferred", "conditions": ["Apache-2.0 notice", "source attribution", "honor embedded repository terms"]},
+      "public_payload": "hosted",
+      "evidence": [
+        {"finding": "The publisher's project links this exact dataset as its released training dataset, carries Apache-2.0, and says embedded repositories are limited to permissive MIT/Apache/BSD licenses.", "url": "https://github.com/multimodal-art-projection/TerminalTraj"},
+        {"finding": "Pinned dataset card links the same paper but does not repeat the project license.", "url": "https://huggingface.co/datasets/m-a-p/TerminalTraj/blob/5c1823f4a8b9ca0cf02c27d2db52c5b35b53a308/README.md"}
+      ]
+    },
+    {
+      "source_id": "agentsuite_tau2",
+      "upstream": "https://huggingface.co/datasets/AgentSuite/tau2-bench-trajectories/tree/07264f1d253a618fc20966670eddc44246212916",
+      "declared_license": null,
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted_with_conditions", "verification": "inferred", "conditions": ["public-release research context", "honor tau2-bench MIT terms and model-provider terms"]},
+      "model_training": {"permission": "uncertain", "verification": "uncertain", "conditions": ["obtain or confirm trajectory publisher and model-output rights"]},
+      "commercial_training": {"permission": "uncertain", "verification": "uncertain"},
+      "redistribution_original": {"permission": "uncertain", "verification": "uncertain"},
+      "redistribution_transformed": {"permission": "uncertain", "verification": "uncertain"},
+      "public_payload": "metadata_only",
+      "evidence": [
+        {"finding": "Pinned dataset is explicitly a public release but supplies no license or usage terms.", "url": "https://huggingface.co/datasets/AgentSuite/tau2-bench-trajectories/blob/07264f1d253a618fc20966670eddc44246212916/README.md"},
+        {"finding": "AgentSuite vendors tau2-bench under MIT, but that does not by itself license separately hosted model outputs.", "url": "https://github.com/Agent-Suite/AgentSuite/tree/0f9eac1c1a376a411ad807bd974555055f08e6c5/tau2-bench"}
+      ]
+    },
+    {
+      "source_id": "official_tau3",
+      "upstream": "https://github.com/sierra-research/tau2-bench/tree/dd8757d4c8791dad7b3cdaf02a105a3d5f9d86bd",
+      "declared_license": "MIT",
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted", "verification": "verified"},
+      "model_training": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["training on benchmark tasks or rewards creates evaluation overlap and must be disclosed as a custom submission"]},
+      "commercial_training": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain MIT notice", "disclose benchmark contamination"]},
+      "redistribution_original": {"permission": "permitted", "verification": "verified", "conditions": ["retain MIT notice"]},
+      "redistribution_transformed": {"permission": "permitted", "verification": "verified", "conditions": ["retain MIT notice and source attribution"]},
+      "public_payload": "hosted",
+      "evidence": [
+        {"finding": "The exact result JSON files are bundled in the pinned MIT-licensed repository.", "url": "https://github.com/sierra-research/tau2-bench/tree/dd8757d4c8791dad7b3cdaf02a105a3d5f9d86bd/data/tau2/results/final"},
+        {"finding": "Training is supported, while leaderboard rules classify domain/task overlap as a custom submission requiring disclosure.", "url": "https://github.com/sierra-research/tau2-bench/blob/dd8757d4c8791dad7b3cdaf02a105a3d5f9d86bd/docs/leaderboard-submission.md"}
+      ]
+    },
+    {
+      "source_id": "mind2web",
+      "upstream": "https://huggingface.co/datasets/osunlp/Mind2Web/tree/17ece8eb89862368edc0cc806acee6fca5163474",
+      "declared_license": "CC-BY-4.0 with a research-only intended-use statement",
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted", "verification": "verified", "conditions": ["provide attribution"]},
+      "model_training": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["research use", "provide attribution", "avoid harmful use"]},
+      "commercial_training": {"permission": "restricted", "verification": "verified", "conditions": ["dataset card states it was released solely for research purposes despite CC-BY metadata"]},
+      "redistribution_original": {"permission": "permitted", "verification": "verified", "conditions": ["CC-BY-4.0 attribution", "retain research-use disclaimer"]},
+      "redistribution_transformed": {"permission": "permitted", "verification": "verified", "conditions": ["CC-BY-4.0 attribution", "identify canonical transformations", "retain research-use disclaimer"]},
+      "public_payload": "hosted",
+      "evidence": [
+        {"finding": "Pinned card declares CC-BY-4.0 and separately states research-only release intent.", "url": "https://huggingface.co/datasets/osunlp/Mind2Web/blob/17ece8eb89862368edc0cc806acee6fca5163474/README.md"}
+      ]
+    },
+    {
+      "source_id": "toolbench",
+      "upstream": "https://github.com/OpenBMB/ToolBench/tree/d56fdd89faf8c91fa135090b212bb9057ee5cfc2",
+      "declared_license": "Apache-2.0 with research-and-education intended use",
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted", "verification": "verified", "conditions": ["research or education", "consider third-party API rights"]},
+      "model_training": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["research or education", "consider third-party API rights"]},
+      "commercial_training": {"permission": "restricted", "verification": "verified", "conditions": ["README states ToolBench is intended solely for research and educational purposes"]},
+      "redistribution_original": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["Apache-2.0 notice", "research/education intended use", "third-party API content remains subject to third-party rights"]},
+      "redistribution_transformed": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["Apache-2.0 notice", "source attribution", "research/education intended use", "third-party API content remains subject to third-party rights"]},
+      "public_payload": "hosted",
+      "evidence": [
+        {"finding": "Pinned README says the dataset is distributed under Apache-2.0, is intended solely for research and education, and provides SFT data/training instructions.", "url": "https://github.com/OpenBMB/ToolBench/blob/d56fdd89faf8c91fa135090b212bb9057ee5cfc2/README.md"}
+      ]
+    },
+    {
+      "source_id": "swegym_sampled",
+      "upstream": "https://huggingface.co/datasets/SWE-Gym/OpenHands-Sampled-Trajectories/tree/baf3a4e4bff514d48ddc08a93a2ade5c126212c7",
+      "declared_license": null,
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted_with_conditions", "verification": "inferred", "conditions": ["SWE-Gym project is Apache-2.0 and research/training oriented", "confirm dataset-specific terms"]},
+      "model_training": {"permission": "uncertain", "verification": "uncertain"},
+      "commercial_training": {"permission": "uncertain", "verification": "uncertain"},
+      "redistribution_original": {"permission": "uncertain", "verification": "uncertain"},
+      "redistribution_transformed": {"permission": "uncertain", "verification": "uncertain"},
+      "public_payload": "metadata_only",
+      "evidence": [
+        {"finding": "Pinned dataset card has no license or usage statement.", "url": "https://huggingface.co/datasets/SWE-Gym/OpenHands-Sampled-Trajectories/blob/baf3a4e4bff514d48ddc08a93a2ade5c126212c7/README.md"},
+        {"finding": "The related SWE-Gym project is Apache-2.0, but its training documentation references a different SFT trajectory dataset and does not establish terms for this separate repository.", "url": "https://github.com/SWE-Gym/SWE-Gym/tree/b681068ca20628c6987b7416cc4cf03f06b77ba5"}
+      ]
+    },
+    {
+      "source_id": "swebench_verified_trajectories",
+      "upstream": "https://huggingface.co/datasets/tarsur385/swebench-verified-trajectories/tree/773748a7c1222e8a642a7059821498e14293562a",
+      "declared_license": "MIT",
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted", "verification": "verified", "conditions": ["honor underlying task repository licenses"]},
+      "model_training": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain MIT notice", "honor underlying task repository licenses", "avoid evaluation leakage"]},
+      "commercial_training": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["honor underlying task repository and model-provider terms"]},
+      "redistribution_original": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain MIT notice", "underlying task repositories retain their licenses"]},
+      "redistribution_transformed": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain MIT notice and attribution", "underlying task repositories retain their licenses"]},
+      "public_payload": "hosted",
+      "evidence": [
+        {"finding": "Pinned dataset card declares MIT and identifies the trajectory provenance.", "url": "https://huggingface.co/datasets/tarsur385/swebench-verified-trajectories/blob/773748a7c1222e8a642a7059821498e14293562a/README.md"}
+      ]
+    },
+    {
+      "source_id": "swebench_multilingual_trajectories",
+      "upstream": "https://huggingface.co/datasets/AlienKevin/SWE-bench-Multilingual-trajectories/tree/7221c09c6cc0ed16d51c9cb9886a1ab484465670",
+      "declared_license": "MIT",
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted", "verification": "verified", "conditions": ["honor underlying task repository licenses"]},
+      "model_training": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain MIT notice", "honor underlying task repository licenses", "avoid evaluation leakage"]},
+      "commercial_training": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["honor underlying task repository and model-provider terms"]},
+      "redistribution_original": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain MIT notice", "underlying task repositories retain their licenses"]},
+      "redistribution_transformed": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain MIT notice and attribution", "underlying task repositories retain their licenses"]},
+      "public_payload": "hosted",
+      "evidence": [
+        {"finding": "Pinned dataset card declares MIT and describes the three trajectory splits/models.", "url": "https://huggingface.co/datasets/AlienKevin/SWE-bench-Multilingual-trajectories/blob/7221c09c6cc0ed16d51c9cb9886a1ab484465670/README.md"}
+      ]
+    },
+    {
+      "source_id": "terminalbench_trajectories",
+      "upstream": "https://huggingface.co/datasets/yoonholee/terminalbench-trajectories/tree/04e8940f5b6736a7ce8d22224fe2f2af74163ed2",
+      "declared_license": "Apache-2.0",
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted", "verification": "verified"},
+      "model_training": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain Apache-2.0 notice", "avoid or disclose Terminal-Bench evaluation leakage"]},
+      "commercial_training": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["review model-provider terms represented in the scraped public leaderboard trajectories"]},
+      "redistribution_original": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain Apache-2.0 notice and attribution", "source says trajectories were scraped from public leaderboard data"]},
+      "redistribution_transformed": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain Apache-2.0 notice and attribution", "identify canonical transformations"]},
+      "public_payload": "hosted",
+      "evidence": [
+        {"finding": "Pinned dataset card declares Apache-2.0, describes complete trajectories, and identifies the public leaderboard source.", "url": "https://huggingface.co/datasets/yoonholee/terminalbench-trajectories/blob/04e8940f5b6736a7ce8d22224fe2f2af74163ed2/README.md"}
+      ]
+    },
+    {
+      "source_id": "browsecomp_rollouts",
+      "upstream": "https://huggingface.co/datasets/rl-rag/browsecomp-gpt-oss-120b-260222/tree/ac4854be0eb527cdf90f6c7d09446eeb7a3d1dbd",
+      "declared_license": null,
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted_with_conditions", "verification": "inferred", "conditions": ["underlying BrowseComp data is MIT", "web observations retain third-party rights"]},
+      "model_training": {"permission": "uncertain", "verification": "uncertain", "conditions": ["trajectory publisher terms absent", "web observations may contain third-party content"]},
+      "commercial_training": {"permission": "uncertain", "verification": "uncertain"},
+      "redistribution_original": {"permission": "uncertain", "verification": "uncertain"},
+      "redistribution_transformed": {"permission": "uncertain", "verification": "uncertain"},
+      "public_payload": "metadata_only",
+      "evidence": [
+        {"finding": "Pinned trajectory card has no license and includes full model conversations plus search/open results from third-party sites.", "url": "https://huggingface.co/datasets/rl-rag/browsecomp-gpt-oss-120b-260222/blob/ac4854be0eb527cdf90f6c7d09446eeb7a3d1dbd/README.md"},
+        {"finding": "Underlying BrowseComp questions are MIT in OpenAI simple-evals; that does not automatically license the separate rollouts or retrieved web content.", "url": "https://github.com/openai/simple-evals/tree/652c89d0ca9df547706735883097e9537d40dc47"}
+      ]
+    },
+    {
+      "source_id": "swesmith_trajectories",
+      "upstream": "https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories/tree/08e109b4a59eaeebf80e4675cd125d42e7ac99a4",
+      "declared_license": "MIT",
+      "public_access": {"permission": "permitted", "verification": "verified"},
+      "local_research_use": {"permission": "permitted", "verification": "verified", "conditions": ["honor generated task repository licenses"]},
+      "model_training": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain MIT notice", "honor generated task repository licenses"]},
+      "commercial_training": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["honor generated task repository and model-provider terms"]},
+      "redistribution_original": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain MIT notice", "generated task repositories retain their licenses"]},
+      "redistribution_transformed": {"permission": "permitted_with_conditions", "verification": "verified", "conditions": ["retain MIT notice and attribution", "generated task repositories retain their licenses"]},
+      "public_payload": "hosted",
+      "evidence": [
+        {"finding": "Pinned dataset card declares MIT and states the trajectories were used to fine-tune SWE-agent-LM.", "url": "https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories/blob/08e109b4a59eaeebf80e4675cd125d42e7ac99a4/README.md"}
+      ]
+    }
+  ]
+}
diff --git a/registry/sources.json b/registry/sources.json
new file mode 100644
index 0000000000000000000000000000000000000000..c2a812a8c7713494f41df44bca4842f13c058e2f
--- /dev/null
+++ b/registry/sources.json
@@ -0,0 +1,465 @@
+{
+  "registry_version": 1,
+  "audit_date": "2026-10-14",
+  "sources": [
+    {
+      "id": "swegym_openhands",
+      "name": "SWE-Gym OpenHands SFT Trajectories",
+      "url": "https://huggingface.co/datasets/SWE-Gym/OpenHands-SFT-Trajectories",
+      "pinned_revision": "4aaa5a4a4b5861f4799d2336908760c190ac3b17",
+      "related_code_revision": "b681068ca20628c6987b7416cc4cf03f06b77ba5",
+      "available_splits": ["train.success.oss"],
+      "upstream_record_count": 491,
+      "qualifying_trajectory_count": 491,
+      "verified_record_count": 491,
+      "reported_statistics": "Dataset server reports 491 rows; all 491 were imported and passed the sequential-action/context gate.",
+      "schema_summary": "Messages-only conversations with system/user/assistant turns; assistant tool actions and execution results are encoded in message content.",
+      "full_trajectory": "yes for all 491 audited rows",
+      "provenance": "OpenHands agent rollouts; exact model is not supplied per row",
+      "verification": "Published train.success.oss split label; not independently replayed",
+      "eligibility": "conditionally-usable",
+      "license": "MIT",
+      "task_id": "Stable hash of the initial user task because rows do not expose an instance ID",
+      "duplicate_risks": ["SWE-bench task overlap", "multiple attempts per instance"],
+      "limitations": ["Tool calls/results are content-encoded rather than structured fields", "Success is source-label-derived rather than independently replayed"]
+    },
+    {
+      "id": "nvidia_open_swe_traces",
+      "name": "NVIDIA Open-SWE-Traces",
+      "url": "https://huggingface.co/datasets/nvidia/Open-SWE-Traces",
+      "pinned_revision": "6d0a83373ae0d83abe3d363f1f18cb1822e16586",
+      "available_splits": ["v1.0: openhands/sweagent", "v1.1: openhands/sweagent/minisweagent", "v1.2: openhands/sweagent/minisweagent"],
+      "upstream_record_count": 677815,
+      "qualifying_trajectory_count": 677815,
+      "not_imported_due_resource_limit": 626189,
+      "rejected_before_library": 0,
+      "verified_record_count": 51626,
+      "reported_statistics": "The pinned release enumerates 677,815 rows (v1.0 151,219; v1.1 263,226; v1.2 263,370). All 51,626 v1.2 OpenHands rows were imported; 626,189 rows in the other version/scaffold splits remain unimported.",
+      "schema_summary": "Verified fields include trajectory_id, instance_id, repo, resolved, messages, tools, metadata, language and license; messages/tools/metadata are JSON-encoded strings.",
+      "full_trajectory": "yes",
+      "provenance": "Model agent traces with model/scaffold metadata preserved from each row",
+      "verification": "Published resolved field; not independently replayed here",
+      "eligibility": "conditionally-usable",
+      "license": "CC-BY-4.0",
+      "task_id": "Native instance_id",
+      "duplicate_risks": ["SWE-bench and SWE-Gym overlap", "multiple attempts"],
+      "limitations": ["Only v1.2/OpenHands is persisted locally", "The remaining 626,189 rows comprise 55,056,684,304 bytes of source Parquet", "Source outcomes were not independently replayed"]
+    },
+    {
+      "id": "terminaltraj",
+      "name": "TerminalTraj",
+      "url": "https://huggingface.co/datasets/m-a-p/TerminalTraj",
+      "pinned_revision": "5c1823f4a8b9ca0cf02c27d2db52c5b35b53a308",
+      "available_splits": ["train"],
+      "upstream_record_count": 20000,
+      "qualifying_trajectory_count": 19143,
+      "verified_record_count": 20000,
+      "reported_statistics": "All 20,000 train trajectories imported; 19,143 contain at least two fully contextualized assistant actions.",
+      "schema_summary": "Alternating user terminal-output and assistant JSON action messages containing analysis, plan, commands and task_complete.",
+      "full_trajectory": "yes for audited rows",
+      "provenance": "Terminal agent command/output trajectories; exact model is not supplied per row",
+      "verification": "Paired command/output transcript; no trustworthy task-level success field",
+      "eligibility": "conditionally-usable",
+      "license": "unverified at dataset revision",
+      "task_id": "Native task ID required",
+      "duplicate_risks": ["Terminal-Bench overlap", "repeated attempts"],
+      "limitations": ["No trustworthy binary success indicator", "857 imported records are short or lack full intervening context and remain tier-5 diagnostics"]
+    },
+    {
+      "id": "agentsuite_tau2",
+      "name": "AgentSuite tau2-bench trajectories",
+      "url": "https://huggingface.co/datasets/AgentSuite/tau2-bench-trajectories",
+      "pinned_revision": "07264f1d253a618fc20966670eddc44246212916",
+      "available_splits": ["train"],
+      "upstream_record_count": 8340,
+      "qualifying_trajectory_count": 8337,
+      "verified_record_count": 8340,
+      "reported_statistics": "All 8,340 public trajectories (30 models x 278 tasks) imported from the pinned native JSONL files.",
+      "schema_summary": "System/user/assistant/tool messages plus eval_result.score/db_match and meta.is_correct/reward_details/finish_reason.",
+      "full_trajectory": "yes for all 8,340 imported rows",
+      "provenance": "Model agent plus model user simulator rollouts",
+      "verification": "AgentSuite evaluator fields; meta.is_correct is used for the binary outcome",
+      "eligibility": "conditionally-usable",
+      "license": "unverified at dataset revision",
+      "task_id": "Domain plus native task ID",
+      "duplicate_risks": ["official tau task overlap", "tau2/tau3 version mismatch"],
+      "limitations": ["Tool definitions are absent", "Three records are explicitly incomplete because a source tool call has no returned observation", "Must not be treated as version-matched tau3 demonstrations", "Evaluator results were not independently replayed"]
+    },
+    {
+      "id": "official_tau3",
+      "name": "Official tau3-bench",
+      "url": "https://github.com/sierra-research/tau2-bench",
+      "pinned_revision": "dd8757d4c8791dad7b3cdaf02a105a3d5f9d86bd",
+      "available_splits": ["base and repository-defined evaluation splits"],
+      "upstream_record_count": 200,
+      "qualifying_trajectory_count": 200,
+      "verified_record_count": 200,
+      "verified_file_hashes": [
+        {
+          "path": "data/tau2/results/final/claude-3-7-sonnet-20250219_airline_default_gpt-4.1-2025-04-14_4trials.json",
+          "sha256": "40a2c6a246eab27db5cdefda895fbe7f44be78a23d300817248534aa606d66de"
+        }
+      ],
+      "reported_statistics": null,
+      "schema_summary": "Results object with tasks and simulations; each simulation has task_id, messages, reward_info, trial and seed.",
+      "full_trajectory": "yes for audited text simulations",
+      "provenance": "Model agent plus model user simulator",
+      "verification": "Environment/action/communication reward evidence",
+      "eligibility": "evaluation-only",
+      "license": "MIT code; benchmark evaluation constraints apply",
+      "task_id": "domain plus task_id",
+      "duplicate_risks": ["multiple trials", "AgentSuite tau2 overlap"],
+      "limitations": ["Only one checked-in result file counted", "reference actions are not demonstrations"]
+    },
+    {
+      "id": "mind2web",
+      "name": "Mind2Web",
+      "url": "https://huggingface.co/datasets/osunlp/Mind2Web",
+      "pinned_revision": "17ece8eb89862368edc0cc806acee6fca5163474",
+      "available_splits": ["train"],
+      "upstream_record_count": 1009,
+      "qualifying_trajectory_count": 1009,
+      "verified_record_count": 1009,
+      "reported_statistics": "All 1,009 rows exposed by the dataset-server Parquet conversion imported with every human action and pre-action DOM observation.",
+      "schema_summary": "Human action sequence with confirmed_task and, for every action, operation, action representation, target candidates and pre-action cleaned/raw DOM.",
+      "full_trajectory": "yes for audited action/context sequences",
+      "provenance": "Human web demonstrations",
+      "verification": "Per-action DOM context and human annotation are present; no explicit binary execution-success field",
+      "eligibility": "conditionally-usable",
+      "license": "CC-BY-4.0",
+      "task_id": "Native annotation/task ID required",
+      "duplicate_risks": ["same website/task across splits"],
+      "limitations": ["No explicit binary success indicator", "DOM snapshots make full records large"]
+    },
+    {
+      "id": "toolbench",
+      "name": "ToolBench",
+      "url": "https://github.com/OpenBMB/ToolBench",
+      "pinned_revision": "d56fdd89faf8c91fa135090b212bb9057ee5cfc2",
+      "available_splits": ["G1", "G2", "G3 examples"],
+      "upstream_record_count": 15,
+      "qualifying_trajectory_count": 11,
+      "rejected_before_library": 4,
+      "verified_record_count": 15,
+      "verified_file_hashes": [
+        {
+          "path": "data_example/answer/G1_answer/10_ChatGPT_DFS_woFilter_w2.json",
+          "sha256": "76a77721640045e07142c669008611e6b82b25007c73427aec771adbf9dd9841"
+        }
+      ],
+      "reported_statistics": "Full dataset statistics are upstream-reported and not verified here.",
+      "schema_summary": "answer_generation with function definitions and cumulative train_messages; final list entry is the complete function-call conversation.",
+      "full_trajectory": "yes in checked-in examples",
+      "provenance": "ChatGPT-generated search/tool paths",
+      "verification": "valid_data/win generation status; no independent semantic proof",
+      "eligibility": "conditionally-usable",
+      "license": "Repository states Apache-2.0 and research/educational intent; third-party API content needs review",
+      "task_id": "query_id or stable hash of query",
+      "duplicate_risks": ["same query across answer methods", "cumulative message prefixes"],
+      "limitations": ["13 of 15 examples adapted; two lacked a complete cumulative history", "11 passed the strict sequential context gate", "win/valid_data are not independent semantic success proof"]
+    },
+    {
+      "id": "alfworld",
+      "name": "ALFWorld",
+      "url": "https://github.com/alfworld/alfworld",
+      "pinned_revision": "release-0.2.2-sha256-25171f16e20ad7b048c47275c45b0babf3aa1cbab29cec97387922350a9844bc",
+      "related_code_revision": "aaba6870f86c5be6a08a491f32a50b906227bc3e",
+      "available_splits": ["train (6374)", "valid_seen (251)", "valid_train (200)", "valid_unseen (255)"],
+      "upstream_record_count": 7080,
+      "qualifying_trajectory_count": 0,
+      "rejected_before_library": 7080,
+      "verified_record_count": 7080,
+      "verified_file_hashes": [
+        {
+          "path": "json_2.1.1_json.zip",
+          "sha256": "25171f16e20ad7b048c47275c45b0babf3aa1cbab29cec97387922350a9844bc"
+        },
+        {
+          "path": "json_2.1.1/valid_seen/pick_cool_then_place_in_recep-AppleSliced-None-GarbageCan-25/trial_T20190908_193525_796595/traj_data.json",
+          "sha256": "1c8c8dc3b72bea45bcd3e6714705c257f94f5cc7636094f74a4816ae90a0856a"
+        }
+      ],
+      "reported_statistics": "README reports 3,553 train, 140 seen-validation and 134 unseen-validation games; the verified release archive contains different counts and includes valid_train.",
+      "schema_summary": "Verified traj_data.json with task/scene, three Turk annotations, planner high_pddl, low actions, and image references.",
+      "full_trajectory": "planner action trajectory; not a chat transcript",
+      "provenance": "PDDL planner actions plus human language annotations",
+      "verification": "Official release ZIP integrity passed; one record mapped. Planner actions were not replayed in an environment.",
+      "eligibility": "conditionally-usable",
+      "license": "MIT repository; downloaded data terms still require confirmation",
+      "task_id": "task_id/trial ID",
+      "duplicate_risks": ["language annotations over the same plan"],
+      "limitations": ["Only one of 7,080 records received adapter smoke validation", "tool observations are not present as chat events", "image payload availability was not validated"]
+    },
+    {
+      "id": "swebench_verified",
+      "name": "SWE-bench Verified",
+      "url": "https://huggingface.co/datasets/princeton-nlp/SWE-bench_Verified",
+      "pinned_revision": null,
+      "related_code_revision": "02e7a74ffd0b707aab73d203fe87bdc7c76afc8e",
+      "available_splits": [],
+      "verified_record_count": null,
+      "reported_statistics": null,
+      "schema_summary": "Benchmark instances, not expert trajectories.",
+      "full_trajectory": "no",
+      "provenance": "Curated software issue benchmark",
+      "verification": "Benchmark tests",
+      "eligibility": "evaluation-only",
+      "license": "Per-instance repository licenses require review",
+      "task_id": "instance_id",
+      "duplicate_risks": ["overlap with SWE agent trajectory sources"],
+      "limitations": ["Dataset host blocked", "gold patches must never become model-visible context"]
+    },
+    {
+      "id": "swebench_multilingual",
+      "name": "SWE-bench Multilingual",
+      "url": "https://huggingface.co/datasets/SWE-bench/SWE-bench_Multilingual",
+      "pinned_revision": null,
+      "available_splits": [],
+      "verified_record_count": null,
+      "reported_statistics": null,
+      "schema_summary": "Benchmark instances, not expert trajectories; unverified locally.",
+      "full_trajectory": "no",
+      "provenance": "Software issue benchmark",
+      "verification": "Benchmark tests",
+      "eligibility": "evaluation-only",
+      "license": "Per-instance repository licenses require review",
+      "task_id": "instance_id",
+      "duplicate_risks": ["cross-dataset repository/issue overlap"],
+      "limitations": ["Dataset host blocked", "reference patches must remain hidden"]
+    },
+    {
+      "id": "terminal_bench",
+      "name": "Terminal-Bench",
+      "url": "https://github.com/laude-institute/terminal-bench",
+      "pinned_revision": "d28711d0da2675d0bb1d56de45ae5df6082438a3",
+      "available_splits": ["repository task registry"],
+      "verified_record_count": 241,
+      "reported_statistics": null,
+      "schema_summary": "Task manifests, tests and reference scripts; not complete conversational demonstrations.",
+      "full_trajectory": "no in audited repository",
+      "provenance": "Protected benchmark tasks",
+      "verification": "Container tests",
+      "eligibility": "evaluation-only",
+      "license": "Apache-2.0 repository plus explicit anti-training instructions",
+      "task_id": "task directory/registry ID",
+      "duplicate_risks": ["TerminalTraj task overlap"],
+      "limitations": ["966 files carried anti-training canaries", "reference scripts are not trajectories"]
+    },
+    {
+      "id": "deepswe_datacurve",
+      "name": "DeepSWE (Datacurve)",
+      "url": "https://www.datacurve.ai/",
+      "pinned_revision": null,
+      "available_splits": [],
+      "verified_record_count": null,
+      "reported_statistics": null,
+      "schema_summary": "Exact requested artifact was not resolvable from available evidence.",
+      "full_trajectory": "unverified",
+      "provenance": "unverified",
+      "verification": "unverified",
+      "eligibility": "evaluation-only",
+      "license": "unverified",
+      "task_id": "unverified",
+      "duplicate_risks": ["SWE benchmark overlap"],
+      "limitations": ["Treat as protected evaluation data pending an exact source URL and anti-training review"]
+    },
+    {
+      "id": "browsecomp",
+      "name": "BrowseComp",
+      "url": "https://github.com/openai/simple-evals",
+      "pinned_revision": "652c89d0ca9df547706735883097e9537d40dc47",
+      "available_splits": ["evaluation implementation"],
+      "verified_record_count": null,
+      "reported_statistics": null,
+      "schema_summary": "Question/answer evaluation benchmark, not trajectories.",
+      "full_trajectory": "no",
+      "provenance": "Evaluation benchmark",
+      "verification": "Answer grading",
+      "eligibility": "evaluation-only",
+      "license": "Repository license; dataset terms require separate review",
+      "task_id": "benchmark example ID",
+      "duplicate_risks": ["web question leakage"],
+      "limitations": ["No rollout data inspected"]
+    },
+    {
+      "id": "bfcl",
+      "name": "Berkeley Function Calling Leaderboard",
+      "url": "https://github.com/ShishirPatil/gorilla",
+      "pinned_revision": "6ea57973c7a6097fd7c5915698c54c17c5b1b6c8",
+      "available_splits": ["evaluation implementation"],
+      "verified_record_count": null,
+      "reported_statistics": null,
+      "schema_summary": "Function-calling evaluation cases and expected outputs, not full expert trajectories.",
+      "full_trajectory": "no",
+      "provenance": "Evaluation benchmark",
+      "verification": "Leaderboard validators",
+      "eligibility": "evaluation-only",
+      "license": "Repository/dataset terms require split-level review",
+      "task_id": "benchmark case ID",
+      "duplicate_risks": ["ToolBench/function-call overlap"],
+      "limitations": ["Expected calls must not be exported as training targets"]
+    },
+    {
+      "id": "artificial_analysis_index",
+      "name": "Artificial Analysis Intelligence Index",
+      "url": "https://artificialanalysis.ai/",
+      "pinned_revision": null,
+      "available_splits": [],
+      "verified_record_count": null,
+      "reported_statistics": null,
+      "schema_summary": "Public leaderboard/index, not a downloadable trajectory dataset.",
+      "full_trajectory": "no",
+      "provenance": "Aggregated evaluation index",
+      "verification": "Provider methodology not inspected in this environment",
+      "eligibility": "unavailable",
+      "license": "No training-data license verified",
+      "task_id": "not applicable",
+      "duplicate_risks": [],
+      "limitations": ["No qualifying public trajectory records identified"]
+    },
+    {
+      "id": "swegym_sampled",
+      "name": "SWE-Gym OpenHands Sampled Trajectories",
+      "url": "https://huggingface.co/datasets/SWE-Gym/OpenHands-Sampled-Trajectories",
+      "pinned_revision": "baf3a4e4bff514d48ddc08a93a2ade5c126212c7",
+      "available_splits": ["train.raw"],
+      "upstream_record_count": 6055,
+      "qualifying_trajectory_count": 6055,
+      "verified_record_count": 6055,
+      "reported_statistics": "6,055/6,055 rows imported, including every successful and failed sampled attempt; 553 exact interaction/outcome duplicates collapse into retained canonical provenance groups.",
+      "schema_summary": "Full OpenHands messages, structured tools, run ID, resolved verdict and execution report.",
+      "full_trajectory": "yes",
+      "provenance": "OpenHands model rollouts",
+      "verification": "Source execution test result; not independently replayed here",
+      "eligibility": "conditionally-usable",
+      "license": "Dataset card does not declare a license",
+      "task_id": "instance_id",
+      "duplicate_risks": ["multiple run_id attempts per instance", "SWE-Gym SFT overlap"],
+      "limitations": ["Source evaluator was not independently reproduced"]
+    },
+    {
+      "id": "swebench_verified_trajectories",
+      "name": "SWE-bench Verified mini-swe-agent Trajectories",
+      "url": "https://huggingface.co/datasets/tarsur385/swebench-verified-trajectories",
+      "pinned_revision": "773748a7c1222e8a642a7059821498e14293562a",
+      "available_splits": ["10 models x 500 SWE-bench Verified tasks"],
+      "upstream_record_count": 5000,
+      "qualifying_trajectory_count": 5000,
+      "verified_record_count": 5000,
+      "reported_statistics": "All 5,000 native .traj.json files imported; 456 exact duplicates collapse into retained provenance groups and 4,532 retained interactions pass the strict gate.",
+      "schema_summary": "Native mini-swe-agent messages, tool calls/results, model/run IDs and resolved score.",
+      "full_trajectory": "yes",
+      "provenance": "Ten model families using mini-swe-agent",
+      "verification": "Published SWE-bench execution result; protected evaluation split",
+      "eligibility": "evaluation-only",
+      "license": "MIT dataset card; underlying repositories retain their own licenses",
+      "task_id": "instance_id",
+      "duplicate_risks": ["exactly ten attempts per benchmark task", "cross-source SWE-bench overlap"],
+      "limitations": ["Native tests were not independently rerun in this environment"]
+    },
+    {
+      "id": "swebench_multilingual_trajectories",
+      "name": "SWE-bench Multilingual Trajectories",
+      "url": "https://huggingface.co/datasets/AlienKevin/SWE-bench-Multilingual-trajectories",
+      "pinned_revision": "7221c09c6cc0ed16d51c9cb9886a1ab484465670",
+      "available_splits": ["Multi_SWE_agent_32B_Rust", "Qwen2.5_Coder_32B_Instruct", "SWE_agent_LM_32B"],
+      "upstream_record_count": 129,
+      "qualifying_trajectory_count": 110,
+      "verified_record_count": 129,
+      "reported_statistics": "All 129 trajectories (43 tasks x three models) imported.",
+      "schema_summary": "Action, response/thought, terminal observation and state for every step.",
+      "full_trajectory": "yes",
+      "provenance": "Three SWE-agent-compatible model releases",
+      "verification": "Published resolved/completed/errored fields",
+      "eligibility": "evaluation-only",
+      "license": "MIT",
+      "task_id": "id",
+      "duplicate_risks": ["three model attempts for each of 43 Rust tasks"],
+      "limitations": ["Source verdicts were not independently replayed"]
+    },
+    {
+      "id": "terminalbench_trajectories",
+      "name": "Terminal-Bench 2.0 Recorded Trajectories",
+      "url": "https://huggingface.co/datasets/yoonholee/terminalbench-trajectories",
+      "pinned_revision": "04e8940f5b6736a7ce8d22224fe2f2af74163ed2",
+      "available_splits": ["train"],
+      "upstream_record_count": 52104,
+      "qualifying_trajectory_count": 24171,
+      "rejected_before_library": 17642,
+      "verified_record_count": 34462,
+      "reported_statistics": "52,104 trials enumerated; 34,462 have steps and were imported; 17,642 have null steps. Exact deduplication retains 34,263 interactions, of which 24,086 pass the strict gate.",
+      "schema_summary": "Full step sequence with agent message, tool bundle, returned observation, reward, model, scaffold, duration, tokens and cost.",
+      "full_trajectory": "yes for 34,462 rows; absent for 17,642 rows",
+      "provenance": "26 agent scaffolds and 49 underlying models",
+      "verification": "Source-reported Terminal-Bench reward",
+      "eligibility": "evaluation-only",
+      "license": "Apache-2.0 dataset card",
+      "task_id": "task_name with unique trial_id per attempt",
+      "duplicate_risks": ["typically five trials per task/agent combination"],
+      "limitations": ["Observations are source-truncated to 5,000 characters", "17,642 leaderboard trials contain no trajectory steps"]
+    },
+    {
+      "id": "agentsuite_bfcl",
+      "name": "AgentSuite BFCL Outputs",
+      "url": "https://huggingface.co/datasets/AgentSuite/BFCL-trajectories",
+      "pinned_revision": "a86612c6bb68e6566e5e7d206f7b031af889edc0",
+      "available_splits": ["train"],
+      "upstream_record_count": 115466,
+      "qualifying_trajectory_count": 0,
+      "diagnostic_records_examined": 115466,
+      "rejected_before_library": 115466,
+      "verified_record_count": 115466,
+      "reported_statistics": "All 115,466 records imported for diagnostics; zero pass the full-trajectory action-observation gate.",
+      "schema_summary": "User prompt and one or more model function-call answers, evaluator score and latency; no returned tool observations.",
+      "full_trajectory": "no; final function-call outputs only",
+      "provenance": "26 model outputs through AgentSuite",
+      "verification": "BFCL evaluator score",
+      "eligibility": "evaluation-only",
+      "license": "No license declared on the dataset card",
+      "task_id": "meta.id",
+      "duplicate_risks": ["one attempt per model/task"],
+      "limitations": ["No causal tool results; retained only as tier-5 diagnostics, never as trajectory SFT"]
+    },
+    {
+      "id": "browsecomp_rollouts",
+      "name": "BrowseComp GPT-OSS-120B Rollouts",
+      "url": "https://huggingface.co/datasets/rl-rag/browsecomp-gpt-oss-120b-260222",
+      "pinned_revision": "ac4854be0eb527cdf90f6c7d09446eeb7a3d1dbd",
+      "available_splits": ["train"],
+      "upstream_record_count": 5064,
+      "qualifying_trajectory_count": 5063,
+      "rejected_before_library": 1,
+      "verified_record_count": 5064,
+      "reported_statistics": "5,063 full conversations imported; one malformed row lacks a conversation.",
+      "schema_summary": "Four research-agent attempts per question with tool calls, full tool responses, answer judge verdict and latency.",
+      "full_trajectory": "yes for 5,063/5,064 rows",
+      "provenance": "gpt-oss-120b research-agent rollouts",
+      "verification": "Source-reported GPT-4o answer judge; not independently reproduced",
+      "eligibility": "evaluation-only",
+      "license": "No license declared on the dataset card",
+      "task_id": "qid",
+      "duplicate_risks": ["exactly four attempts per question"],
+      "limitations": ["One malformed record", "Reference answers remain raw provenance and are excluded from model_input"]
+    },
+    {
+      "id": "swesmith_trajectories",
+      "name": "SWE-smith Trajectories",
+      "url": "https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories",
+      "pinned_revision": "08e109b4a59eaeebf80e4675cd125d42e7ac99a4",
+      "available_splits": ["tool", "xml", "ticks"],
+      "upstream_record_count": 24100,
+      "qualifying_trajectory_count": 24100,
+      "verified_record_count": 24100,
+      "reported_statistics": "The tool-format population has 24,100 rows; XML/ticks are alternate serializations, not independent rollouts.",
+      "schema_summary": "Full SWE-agent messages, task/trajectory IDs, resolved outcome and patch.",
+      "full_trajectory": "yes",
+      "provenance": "SWE-agent with Claude-generated trajectories",
+      "verification": "Published resolved field",
+      "eligibility": "conditionally-usable",
+      "license": "MIT",
+      "task_id": "instance_id",
+      "duplicate_risks": ["tool/xml/ticks serialization variants", "SWE-bench task overlap"],
+      "limitations": ["Tool definitions are not published; names and arguments remain in messages"]
+    }
+  ]
+}
diff --git a/staging_report.json b/staging_report.json
new file mode 100644
index 0000000000000000000000000000000000000000..64e8d8b9594f70cfe45d0fe7d122dace0abd6b61
--- /dev/null
+++ b/staging_report.json
@@ -0,0 +1,26 @@
+{
+  "hosted_shards": 50,
+  "hosted_sources": [
+    "mind2web",
+    "nvidia_open_swe_traces",
+    "official_tau3",
+    "swebench_multilingual_trajectories",
+    "swebench_verified_trajectories",
+    "swegym_openhands",
+    "swesmith_trajectories",
+    "terminalbench_trajectories",
+    "terminaltraj",
+    "toolbench"
+  ],
+  "hosted_trajectories": 136373,
+  "inventory_shards": 56,
+  "inventory_trajectories": 155278,
+  "metadata_only_sources": [
+    "agentsuite_tau2",
+    "browsecomp_rollouts",
+    "swegym_sampled"
+  ],
+  "metadata_only_trajectories": 18905,
+  "schema_version": 1,
+  "scores_recomputed": false
+}
diff --git a/structurally_eligible.index.jsonl.gz b/structurally_eligible.index.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..c20e8cc6eeb2572f07f2a489f14e8b82f03c2a96
--- /dev/null
+++ b/structurally_eligible.index.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7abe69bdb9cd3d07b48a282c0bf371f7bcff26776ad3caba307e22b45faf40f8
+size 4326891
diff --git a/successful.index.jsonl.gz b/successful.index.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..67fb6cc0530c5c7f35938fd2a41119540f1aab12
--- /dev/null
+++ b/successful.index.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30b4ec2dd4faacb804e78a2826df649af7737807f542ddc6f667e314ae08c3e1
+size 1561257
diff --git a/task_splits.jsonl.gz b/task_splits.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..69e30782ec9554a71819fe43b59bade60ee01847
--- /dev/null
+++ b/task_splits.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f932f6d111636fa2306d1b2fef94a6a5b98e9fdab426c95f0da78fced367f776
+size 582841
diff --git a/training_eligible.index.jsonl.gz b/training_eligible.index.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..61edcd02e0314b3e877438b1a49947b7cae7bad2
--- /dev/null
+++ b/training_eligible.index.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd0619062d95f44824710e32922c7fc330c44c367f41c2adcd3fb2002308a00f
+size 3426523
diff --git a/unknown.index.jsonl.gz b/unknown.index.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..3468c9c66a31a4a7bd8e3afc1c0771fd8f539f0a
--- /dev/null
+++ b/unknown.index.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a11f7e31b6bf29ee513709a4118ad99823a8634b6f0b3f5ecd8cf605b58cc45
+size 663458
diff --git a/unsuccessful.index.jsonl.gz b/unsuccessful.index.jsonl.gz
new file mode 100644
index 0000000000000000000000000000000000000000..10395e15393e3f0d5e1793141cf5f875773a237d
--- /dev/null
+++ b/unsuccessful.index.jsonl.gz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0e83f5fa6f85137c64534acc9f400663ac458d44fb88758db407007ae259273
+size 2233251