chore: refresh CI-compatible clean-lock evidence
This commit is contained in:
@@ -3,7 +3,7 @@
|
||||
"ok": true,
|
||||
"generated_at": "2026-06-14",
|
||||
"skill_dir": ".",
|
||||
"commit": "ef0089170c1f57ab337e452a8948ceb3ae625ca1",
|
||||
"commit": "a364f5766fd0dcb338ffea827d0d8f1058ad2ebc",
|
||||
"git_status": {
|
||||
"available": true,
|
||||
"dirty": false,
|
||||
@@ -17,9 +17,9 @@
|
||||
"methodology_complete": true,
|
||||
"required_artifact_count": 24,
|
||||
"missing_artifact_count": 0,
|
||||
"evidence_bundle_sha256": "f81d4dd5985eff30c70e3b4720c44a9d5b4984318463637b98645b526bfcbd6b",
|
||||
"evidence_bundle_sha256": "d2af8e406c6fb43fd48bac179752e4050f5b214eef3baed9cc7920822c8e2e2e",
|
||||
"source_contract_sha256": "d3341950a7fb52a18a3a2a7453f7c697cc92e443c6604facdeeb83848d4f867c",
|
||||
"archive_sha256": "c2028f0a12c88f2cae3462c80af972be934333829252faf7c92816bcc54dc541",
|
||||
"archive_sha256": "587e2efa7dc1012628af57bf3e518f963f53d12d56acc74e7d772021cad8845a",
|
||||
"output_case_count": 5,
|
||||
"failure_disclosure_count": 3,
|
||||
"command_count": 21,
|
||||
@@ -50,7 +50,7 @@
|
||||
},
|
||||
"release_lock": {
|
||||
"ready": true,
|
||||
"commit": "ef0089170c1f57ab337e452a8948ceb3ae625ca1",
|
||||
"commit": "a364f5766fd0dcb338ffea827d0d8f1058ad2ebc",
|
||||
"status_scope": "generation-time status before this report is written",
|
||||
"reason": "clean generation-time HEAD"
|
||||
},
|
||||
@@ -60,7 +60,7 @@
|
||||
"existing_count": 24,
|
||||
"missing_count": 0,
|
||||
"missing_paths": [],
|
||||
"sha256": "f81d4dd5985eff30c70e3b4720c44a9d5b4984318463637b98645b526bfcbd6b"
|
||||
"sha256": "d2af8e406c6fb43fd48bac179752e4050f5b214eef3baed9cc7920822c8e2e2e"
|
||||
},
|
||||
"methodology": {
|
||||
"path": "reports/benchmark_methodology.md",
|
||||
@@ -133,8 +133,8 @@
|
||||
"label": "output_execution",
|
||||
"path": "reports/output_execution_runs.json",
|
||||
"exists": true,
|
||||
"bytes": 7967,
|
||||
"sha256": "46a65f5db66718f2f3443467c06445bee6a6269586555c307d124b03784c2d1c"
|
||||
"bytes": 7966,
|
||||
"sha256": "1e32e9c33626c2f5d8cfc4de0c47acbca23285e0d19f54d49a3a0027b4808953"
|
||||
},
|
||||
{
|
||||
"label": "blind_review",
|
||||
@@ -183,14 +183,14 @@
|
||||
"path": "reports/registry_audit.json",
|
||||
"exists": true,
|
||||
"bytes": 3183,
|
||||
"sha256": "eba3fd128cb58de877f8e165a743192a65b5cc0c10a9086cf02fbf324a4e0fb7"
|
||||
"sha256": "85bc616e7171da92f7c8060a43701810320417c23bc8ef07f8f50a281d2c1540"
|
||||
},
|
||||
{
|
||||
"label": "package_verification",
|
||||
"path": "reports/package_verification.json",
|
||||
"exists": true,
|
||||
"bytes": 19325,
|
||||
"sha256": "f7382b0152d969ed99fec9e0063dde02cf598f4122c4d58ec1e88198a125410b"
|
||||
"sha256": "e7a5905285c83aab965df96190f850d53a7173a09e8a277394c771232cf4c0ac"
|
||||
},
|
||||
{
|
||||
"label": "install_simulation",
|
||||
|
||||
@@ -1,9 +1,9 @@
|
||||
# Benchmark Reproducibility
|
||||
|
||||
Generated at: `2026-06-14`
|
||||
Commit: `ef0089170c1f57ab337e452a8948ceb3ae625ca1`
|
||||
Commit: `a364f5766fd0dcb338ffea827d0d8f1058ad2ebc`
|
||||
Working tree dirty at generation: `false`
|
||||
Evidence bundle SHA256: `f81d4dd5985eff30c70e3b4720c44a9d5b4984318463637b98645b526bfcbd6b`
|
||||
Evidence bundle SHA256: `d2af8e406c6fb43fd48bac179752e4050f5b214eef3baed9cc7920822c8e2e2e`
|
||||
|
||||
## Summary
|
||||
|
||||
@@ -13,7 +13,7 @@ Evidence bundle SHA256: `f81d4dd5985eff30c70e3b4720c44a9d5b4984318463637b98645b5
|
||||
- required artifacts: `24`
|
||||
- missing artifacts: `0`
|
||||
- source contract sha256: `d3341950a7fb`
|
||||
- archive sha256: `c2028f0a12c8`
|
||||
- archive sha256: `587e2efa7dc1`
|
||||
- output cases: `5`
|
||||
- disclosed failure cases: `3`
|
||||
- reproduction commands: `21`
|
||||
@@ -48,7 +48,7 @@ This report proves local benchmark reproducibility only. It keeps external provi
|
||||
|
||||
- algorithm: `sha256(path,label,exists,artifact_sha256)`
|
||||
- artifacts: `24` / `24`
|
||||
- sha256: `f81d4dd5985eff30c70e3b4720c44a9d5b4984318463637b98645b526bfcbd6b`
|
||||
- sha256: `d2af8e406c6fb43fd48bac179752e4050f5b214eef3baed9cc7920822c8e2e2e`
|
||||
|
||||
## Methodology Sections
|
||||
|
||||
@@ -70,15 +70,15 @@ This report proves local benchmark reproducibility only. It keeps external provi
|
||||
| output_cases | `evals/output/cases.jsonl` | present | `a6ae96857116` |
|
||||
| output_schema | `evals/output/schema.json` | present | `8ee340c95064` |
|
||||
| output_scorecard | `reports/output_quality_scorecard.json` | present | `0806258a8e08` |
|
||||
| output_execution | `reports/output_execution_runs.json` | present | `46a65f5db667` |
|
||||
| output_execution | `reports/output_execution_runs.json` | present | `1e32e9c33626` |
|
||||
| blind_review | `reports/output_blind_review_pack.json` | present | `bbe2db8ec277` |
|
||||
| review_adjudication | `reports/output_review_adjudication.json` | present | `240485a721af` |
|
||||
| trigger_scorecard | `reports/route_scorecard.json` | present | `c164e83e36d0` |
|
||||
| runtime_conformance | `reports/conformance_matrix.json` | present | `8251329e663d` |
|
||||
| trust_report | `reports/security_trust_report.json` | present | `d827b410e989` |
|
||||
| python_compatibility | `reports/python_compatibility.json` | present | `ae16e17266e4` |
|
||||
| registry_audit | `reports/registry_audit.json` | present | `eba3fd128cb5` |
|
||||
| package_verification | `reports/package_verification.json` | present | `f7382b0152d9` |
|
||||
| registry_audit | `reports/registry_audit.json` | present | `85bc616e7171` |
|
||||
| package_verification | `reports/package_verification.json` | present | `e7a5905285c8` |
|
||||
| install_simulation | `reports/install_simulation.json` | present | `8f987e805c92` |
|
||||
| skill_os2_audit | `reports/skill_os2_audit.json` | present | `6bb2dcb0e1e5` |
|
||||
| world_class_evidence_plan | `reports/world_class_evidence_plan.json` | present | `164803bb1cea` |
|
||||
|
||||
+12
-12
@@ -1336,9 +1336,9 @@
|
||||
"methodology_complete": true,
|
||||
"required_artifact_count": 24,
|
||||
"missing_artifact_count": 0,
|
||||
"evidence_bundle_sha256": "f81d4dd5985eff30c70e3b4720c44a9d5b4984318463637b98645b526bfcbd6b",
|
||||
"evidence_bundle_sha256": "d2af8e406c6fb43fd48bac179752e4050f5b214eef3baed9cc7920822c8e2e2e",
|
||||
"source_contract_sha256": "d3341950a7fb52a18a3a2a7453f7c697cc92e443c6604facdeeb83848d4f867c",
|
||||
"archive_sha256": "c2028f0a12c88f2cae3462c80af972be934333829252faf7c92816bcc54dc541",
|
||||
"archive_sha256": "587e2efa7dc1012628af57bf3e518f963f53d12d56acc74e7d772021cad8845a",
|
||||
"output_case_count": 5,
|
||||
"failure_disclosure_count": 3,
|
||||
"command_count": 21,
|
||||
@@ -1357,7 +1357,7 @@
|
||||
"working_tree_dirty": false,
|
||||
"changed_file_count": 0
|
||||
},
|
||||
"commit": "ef0089170c1f57ab337e452a8948ceb3ae625ca1",
|
||||
"commit": "a364f5766fd0dcb338ffea827d0d8f1058ad2ebc",
|
||||
"missing_artifacts": [],
|
||||
"limitations": [
|
||||
"The git commit and dirty flag are generation-time context; the evidence bundle hash is the durable artifact anchor inside a committed report.",
|
||||
@@ -4764,7 +4764,7 @@
|
||||
"ok": true,
|
||||
"generated_at": "2026-06-14",
|
||||
"skill_dir": ".",
|
||||
"commit": "ef0089170c1f57ab337e452a8948ceb3ae625ca1",
|
||||
"commit": "a364f5766fd0dcb338ffea827d0d8f1058ad2ebc",
|
||||
"git_status": {
|
||||
"available": true,
|
||||
"dirty": false,
|
||||
@@ -4778,9 +4778,9 @@
|
||||
"methodology_complete": true,
|
||||
"required_artifact_count": 24,
|
||||
"missing_artifact_count": 0,
|
||||
"evidence_bundle_sha256": "f81d4dd5985eff30c70e3b4720c44a9d5b4984318463637b98645b526bfcbd6b",
|
||||
"evidence_bundle_sha256": "d2af8e406c6fb43fd48bac179752e4050f5b214eef3baed9cc7920822c8e2e2e",
|
||||
"source_contract_sha256": "d3341950a7fb52a18a3a2a7453f7c697cc92e443c6604facdeeb83848d4f867c",
|
||||
"archive_sha256": "c2028f0a12c88f2cae3462c80af972be934333829252faf7c92816bcc54dc541",
|
||||
"archive_sha256": "587e2efa7dc1012628af57bf3e518f963f53d12d56acc74e7d772021cad8845a",
|
||||
"output_case_count": 5,
|
||||
"failure_disclosure_count": 3,
|
||||
"command_count": 21,
|
||||
@@ -4811,7 +4811,7 @@
|
||||
},
|
||||
"release_lock": {
|
||||
"ready": true,
|
||||
"commit": "ef0089170c1f57ab337e452a8948ceb3ae625ca1",
|
||||
"commit": "a364f5766fd0dcb338ffea827d0d8f1058ad2ebc",
|
||||
"status_scope": "generation-time status before this report is written",
|
||||
"reason": "clean generation-time HEAD"
|
||||
},
|
||||
@@ -4821,7 +4821,7 @@
|
||||
"existing_count": 24,
|
||||
"missing_count": 0,
|
||||
"missing_paths": [],
|
||||
"sha256": "f81d4dd5985eff30c70e3b4720c44a9d5b4984318463637b98645b526bfcbd6b"
|
||||
"sha256": "d2af8e406c6fb43fd48bac179752e4050f5b214eef3baed9cc7920822c8e2e2e"
|
||||
},
|
||||
"methodology": {
|
||||
"path": "reports/benchmark_methodology.md",
|
||||
@@ -4894,8 +4894,8 @@
|
||||
"label": "output_execution",
|
||||
"path": "reports/output_execution_runs.json",
|
||||
"exists": true,
|
||||
"bytes": 7967,
|
||||
"sha256": "46a65f5db66718f2f3443467c06445bee6a6269586555c307d124b03784c2d1c"
|
||||
"bytes": 7966,
|
||||
"sha256": "1e32e9c33626c2f5d8cfc4de0c47acbca23285e0d19f54d49a3a0027b4808953"
|
||||
},
|
||||
{
|
||||
"label": "blind_review",
|
||||
@@ -4944,14 +4944,14 @@
|
||||
"path": "reports/registry_audit.json",
|
||||
"exists": true,
|
||||
"bytes": 3183,
|
||||
"sha256": "eba3fd128cb58de877f8e165a743192a65b5cc0c10a9086cf02fbf324a4e0fb7"
|
||||
"sha256": "85bc616e7171da92f7c8060a43701810320417c23bc8ef07f8f50a281d2c1540"
|
||||
},
|
||||
{
|
||||
"label": "package_verification",
|
||||
"path": "reports/package_verification.json",
|
||||
"exists": true,
|
||||
"bytes": 19325,
|
||||
"sha256": "f7382b0152d969ed99fec9e0063dde02cf598f4122c4d58ec1e88198a125410b"
|
||||
"sha256": "e7a5905285c83aab965df96190f850d53a7173a09e8a277394c771232cf4c0ac"
|
||||
},
|
||||
{
|
||||
"label": "install_simulation",
|
||||
|
||||
@@ -923,9 +923,9 @@
|
||||
"methodology_complete": true,
|
||||
"required_artifact_count": 24,
|
||||
"missing_artifact_count": 0,
|
||||
"evidence_bundle_sha256": "f81d4dd5985eff30c70e3b4720c44a9d5b4984318463637b98645b526bfcbd6b",
|
||||
"evidence_bundle_sha256": "d2af8e406c6fb43fd48bac179752e4050f5b214eef3baed9cc7920822c8e2e2e",
|
||||
"source_contract_sha256": "d3341950a7fb52a18a3a2a7453f7c697cc92e443c6604facdeeb83848d4f867c",
|
||||
"archive_sha256": "c2028f0a12c88f2cae3462c80af972be934333829252faf7c92816bcc54dc541",
|
||||
"archive_sha256": "587e2efa7dc1012628af57bf3e518f963f53d12d56acc74e7d772021cad8845a",
|
||||
"output_case_count": 5,
|
||||
"failure_disclosure_count": 3,
|
||||
"command_count": 21,
|
||||
@@ -944,7 +944,7 @@
|
||||
"working_tree_dirty": false,
|
||||
"changed_file_count": 0
|
||||
},
|
||||
"commit": "ef0089170c1f57ab337e452a8948ceb3ae625ca1",
|
||||
"commit": "a364f5766fd0dcb338ffea827d0d8f1058ad2ebc",
|
||||
"missing_artifacts": [],
|
||||
"limitations": [
|
||||
"The git commit and dirty flag are generation-time context; the evidence bundle hash is the durable artifact anchor inside a committed report.",
|
||||
|
||||
@@ -922,9 +922,9 @@
|
||||
"methodology_complete": true,
|
||||
"required_artifact_count": 24,
|
||||
"missing_artifact_count": 0,
|
||||
"evidence_bundle_sha256": "f81d4dd5985eff30c70e3b4720c44a9d5b4984318463637b98645b526bfcbd6b",
|
||||
"evidence_bundle_sha256": "d2af8e406c6fb43fd48bac179752e4050f5b214eef3baed9cc7920822c8e2e2e",
|
||||
"source_contract_sha256": "d3341950a7fb52a18a3a2a7453f7c697cc92e443c6604facdeeb83848d4f867c",
|
||||
"archive_sha256": "c2028f0a12c88f2cae3462c80af972be934333829252faf7c92816bcc54dc541",
|
||||
"archive_sha256": "587e2efa7dc1012628af57bf3e518f963f53d12d56acc74e7d772021cad8845a",
|
||||
"output_case_count": 5,
|
||||
"failure_disclosure_count": 3,
|
||||
"command_count": 21,
|
||||
@@ -943,7 +943,7 @@
|
||||
"working_tree_dirty": false,
|
||||
"changed_file_count": 0
|
||||
},
|
||||
"commit": "ef0089170c1f57ab337e452a8948ceb3ae625ca1",
|
||||
"commit": "a364f5766fd0dcb338ffea827d0d8f1058ad2ebc",
|
||||
"missing_artifacts": [],
|
||||
"limitations": [
|
||||
"The git commit and dirty flag are generation-time context; the evidence bundle hash is the durable artifact anchor inside a committed report.",
|
||||
|
||||
Reference in New Issue
Block a user