Refresh final evidence consistency

This commit is contained in:
yaojingang
2026-06-16 04:49:58 +08:00
parent 5457ad2d23
commit 44cfb472fc
+20 -20
View File
@@ -202,8 +202,8 @@
"key": "benchmark-release-lock-self-consistency",
"label": "Benchmark release lock matches git dirty state",
"status": "pass",
"expected": false,
"actual": false,
"expected": true,
"actual": true,
"paths": [
"reports/benchmark_reproducibility.json"
],
@@ -243,8 +243,8 @@
"key": "overview-benchmark-commit",
"label": "overview embeds the benchmark commit",
"status": "pass",
"expected": "3677dd4664b0777391fb82ab9987a7b6ea975af5",
"actual": "3677dd4664b0777391fb82ab9987a7b6ea975af5",
"expected": "904969b91f6ce1407ea5d973f359b73d5f80ca7a",
"actual": "904969b91f6ce1407ea5d973f359b73d5f80ca7a",
"paths": [
"reports/benchmark_reproducibility.json",
"reports/skill-overview.json"
@@ -256,7 +256,7 @@
"label": "overview embeds benchmark summary fields",
"status": "pass",
"expected": {
"release_lock_ready": false,
"release_lock_ready": true,
"required_artifact_count": 25,
"missing_artifact_count": 0,
"source_contract_sha256": "7de584946cac8f2ed241aa331ba4c0c2d69fa1c4eae2c06b92882b8af6ad07d7",
@@ -266,10 +266,10 @@
"world_class_source_pass_count": 6,
"world_class_source_blocked_count": 7,
"public_claim_ready": false,
"public_claim_blocker_count": 5
"public_claim_blocker_count": 4
},
"actual": {
"release_lock_ready": false,
"release_lock_ready": true,
"required_artifact_count": 25,
"missing_artifact_count": 0,
"source_contract_sha256": "7de584946cac8f2ed241aa331ba4c0c2d69fa1c4eae2c06b92882b8af6ad07d7",
@@ -279,7 +279,7 @@
"world_class_source_pass_count": 6,
"world_class_source_blocked_count": 7,
"public_claim_ready": false,
"public_claim_blocker_count": 5
"public_claim_blocker_count": 4
},
"paths": [
"reports/benchmark_reproducibility.json",
@@ -389,8 +389,8 @@
"key": "interpretation-benchmark-commit",
"label": "interpretation embeds the benchmark commit",
"status": "pass",
"expected": "3677dd4664b0777391fb82ab9987a7b6ea975af5",
"actual": "3677dd4664b0777391fb82ab9987a7b6ea975af5",
"expected": "904969b91f6ce1407ea5d973f359b73d5f80ca7a",
"actual": "904969b91f6ce1407ea5d973f359b73d5f80ca7a",
"paths": [
"reports/benchmark_reproducibility.json",
"reports/skill-interpretation.json"
@@ -402,7 +402,7 @@
"label": "interpretation embeds benchmark summary fields",
"status": "pass",
"expected": {
"release_lock_ready": false,
"release_lock_ready": true,
"required_artifact_count": 25,
"missing_artifact_count": 0,
"source_contract_sha256": "7de584946cac8f2ed241aa331ba4c0c2d69fa1c4eae2c06b92882b8af6ad07d7",
@@ -412,10 +412,10 @@
"world_class_source_pass_count": 6,
"world_class_source_blocked_count": 7,
"public_claim_ready": false,
"public_claim_blocker_count": 5
"public_claim_blocker_count": 4
},
"actual": {
"release_lock_ready": false,
"release_lock_ready": true,
"required_artifact_count": 25,
"missing_artifact_count": 0,
"source_contract_sha256": "7de584946cac8f2ed241aa331ba4c0c2d69fa1c4eae2c06b92882b8af6ad07d7",
@@ -425,7 +425,7 @@
"world_class_source_pass_count": 6,
"world_class_source_blocked_count": 7,
"public_claim_ready": false,
"public_claim_blocker_count": 5
"public_claim_blocker_count": 4
},
"paths": [
"reports/benchmark_reproducibility.json",
@@ -1519,10 +1519,10 @@
"path": "reports",
"label": "Generated evidence and overview artifacts",
"kind": "folder",
"file_count": 229
"file_count": 227
}
],
"file_count": 413,
"file_count": 411,
"folder_count": 4,
"distribution": [
{
@@ -1555,7 +1555,7 @@
},
{
"label": "reports",
"value": 229
"value": 227
}
]
},
@@ -1607,10 +1607,10 @@
"path": "reports",
"label": "Generated evidence and overview artifacts",
"kind": "folder",
"file_count": 229
"file_count": 227
}
],
"file_count": 413,
"file_count": 411,
"folder_count": 4,
"distribution": [
{
@@ -1643,7 +1643,7 @@
},
{
"label": "reports",
"value": 229
"value": 227
}
]
},