Download & verify the evidence
Nothing on this site asks you to trust it. Every verdict is backed by a raw evidence file with a published SHA-256, and the ratio is recomputed from the raw per-trial samples. Verification runs entirely in your browser or terminal: no server call, no model, no cost.
- Download an evidence file below (or open the repository's evidence/ folder).
- Click "Verify SHA-256 in my browser": the page hashes the exact bytes it serves and compares with the value recorded in the bundle.
- Recompute offline: PowerShell Get-FileHash <file> -Algorithm SHA256 · macOS/Linux sha256sum <file>.
- Incident 2 files: re-run the evaluator on the raw samples with python scripts/verify_incident_evidence.py <file> (read-only; no network, no model). Incident 1 files use an earlier format that this script does not read; verify them by SHA-256 (step 3).
- Frozen evidence: python scripts/verify_frozen_evidence.py (compares every frozen file with evidence/FROZEN_EVIDENCE_HASHES.json).
Incident 1 · N+1 query fanout
These two files are different artifacts (different SHA-256, different timings), not duplicates: the same four hypotheses were raced twice, once with a warm in-process timer and once with a fresh process per trial. Both are reported; the difference is explained on the Courtroom page.
Artifact 1 of 2 · warm in-process regime
incident1-v2-warm-container-001 · MEASURED LOCAL · nemotron_in_this_run=false · 15 trials per arm
warm in-process (same regime as the original 210x benchmark)
Proof bundle
evidence/incident1-v2-warm-container-001.race-evidence.json
sha256 9ef78a767b860d8c73a37de8081e9b2070197a95d91b2e10543dde0a4a399fbf (26,326 bytes)
The frozen original contains local file paths, so it is not published. A path-redacted public copy is: same numbers, raw samples and verdicts; 4 path strings replaced by a placeholder.
public copy sha256 b87e1e429fb3f31448ac1178eb90b05b65a948f1bc45381c4e489afe3777aa64 (24,044 bytes)
frozen original sha256 9ef78a767b860d8c73a37de8081e9b2070197a95d91b2e10543dde0a4a399fbf
The original's hash is checked offline in the repository: python scripts/verify_frozen_evidence.py and sha256sum evidence/incident1-v2-warm-container-001.race-evidence.json.
Offline re-check: sha256sum <file> must print the SHA-256 shown above; the stored raw samples reproduce every ratio on this page
Artifact 2 of 2 · cold-process regime
incident1-v2-container-001 · MEASURED LOCAL · nemotron_in_this_run=false · 15 trials per arm
cold fresh process per trial (each trial includes cold start)
Proof bundle
evidence/incident1-v2-container-001.race-evidence.json
sha256 e1c5d2d58b8df3e915e4c3aa962175051e4860e879324b9a01ab49ddee689bf2 (26,258 bytes)
The frozen original contains local file paths, so it is not published. A path-redacted public copy is: same numbers, raw samples and verdicts; 4 path strings replaced by a placeholder.
public copy sha256 414c58586ce140f471ca54444c7f58cb5967f810713154828706c92bbd7ba1ac (23,976 bytes)
frozen original sha256 e1c5d2d58b8df3e915e4c3aa962175051e4860e879324b9a01ab49ddee689bf2
The original's hash is checked offline in the repository: python scripts/verify_frozen_evidence.py and sha256sum evidence/incident1-v2-container-001.race-evidence.json.
Offline re-check: sha256sum <file> must print the SHA-256 shown above; the stored raw samples reproduce every ratio on this page
Incident 2 · missing index
Run 1 of 3
incident2-container-001 · MEASURED LOCAL · nemotron_in_this_run=false · 15 trials per arm
warm in-process, interleaved arms, threads=1
Proof bundle
evidence/incident2/incident2-container-001.race-evidence.json
sha256 5a59359f31a5de777f7abbbffce257a6daba82fcce912d9b1b5e3edbc7782f0f (5,763 bytes)
Offline re-check: python scripts/verify_incident_evidence.py <file>
Run 2 of 3
incident2-container-002 · MEASURED LOCAL · nemotron_in_this_run=false · 15 trials per arm
warm in-process, interleaved arms, threads=1
Proof bundle
evidence/incident2/incident2-container-002.race-evidence.json
sha256 b382025251fd9c5e690cb29159ea69fa66f7df6b5732f840c112ad59c2a1ce1c (5,771 bytes)
Offline re-check: python scripts/verify_incident_evidence.py <file>
Run 3 of 3
incident2-agentic-offline-001 · MEASURED LOCAL · nemotron_in_this_run=false · 15 trials per arm
warm in-process, interleaved arms, threads=1 · offline model-workflow demo
Proof bundle
evidence/incident2/incident2-agentic-offline-001.race-evidence.json
sha256 f2494c177e48e930ae99accdd0f12b4a6193d06a4e4a11f4e3ec53f5fffcedd2 (7,729 bytes)
Offline re-check: python scripts/verify_incident_evidence.py <file>