mirror of
https://github.com/codestable/CodeStable.git
synced 2026-09-19 09:03:09 +08:00
dafang
106 lines
7.8 KiB
JSON
106 lines
7.8 KiB
JSON
{
|
|
"schema_version": 1,
|
|
"state": "prepared-awaiting-commit",
|
|
"prepared_on": "2026-08-02",
|
|
"hash_algorithm": "sha256",
|
|
"source_commit": "pending",
|
|
"real_llm_runs_started": false,
|
|
"model_target_probe": {
|
|
"status": "pending",
|
|
"source_commit": "pending",
|
|
"targets": []
|
|
},
|
|
"primary_metric": {
|
|
"name": "overall_paired_hidden_pass_delta",
|
|
"threshold": 0.25,
|
|
"aggregation": "overall across both model families and four positive fixtures",
|
|
"family_guard": "each family delta > 0"
|
|
},
|
|
"budget_usd": 50.0,
|
|
"required_scale": {
|
|
"model_families": 2,
|
|
"positive_fixtures": 4,
|
|
"guard_fixtures": 2,
|
|
"k": 5,
|
|
"pairs": 60,
|
|
"agent_invocations": 240,
|
|
"hook_runs": 20
|
|
},
|
|
"offline_preflight": {
|
|
"status": "passed",
|
|
"golden_hidden_rate": 1.0,
|
|
"naive_hidden_rate": 0.0,
|
|
"golden_regression_rate": 1.0,
|
|
"naive_regression_rate": 1.0,
|
|
"a_seed_passed": 0,
|
|
"a_seed_total": 4,
|
|
"a_golden_passed": 4,
|
|
"a_golden_total": 4,
|
|
"estimated_cost_usd": 3.41,
|
|
"estimated_cost_tag": "soft"
|
|
},
|
|
"external_inputs": {
|
|
".claude/skills/eval-cs-skill/scripts/_model.py": "5847997e90aba27d998dbd811e9f1507d0979bd89bfb76775da716337d47f24f",
|
|
".claude/skills/eval-cs-skill/scripts/buildprompt.py": "5d1bc95157b7c04fbc9cf8f74d235ce8782c85cf391bf15d301f3f68f7f19fe7",
|
|
".claude/skills/eval-cs-skill/scripts/config.py": "094923d9b0fb8676cc2910758fd8660a9a5d3fd9637ed2189ddaf83442f3d113",
|
|
".claude/skills/eval-cs-skill/scripts/e2e_env.py": "1d7c9b704b79d64b29a5b4d1124a9285433fe44bd7fd3fcbcb11cf023b8a670a",
|
|
".claude/skills/eval-cs-skill/scripts/fixtures.py": "cfb5f41817eec3ec28be4322a5959d31afd8f9b90e10fa6a842b270083dd7305",
|
|
".claude/skills/eval-cs-skill/scripts/harness/__init__.py": "c4acec6d69f515784bdab53afbb99ac5c3f7f5e527a62fea290f534f7f021e41",
|
|
".claude/skills/eval-cs-skill/scripts/harness/adapter_api.py": "b8fafd8fe14d8856274880a45c9f470de17624adc6e0a88f99664acaa6f48dfb",
|
|
".claude/skills/eval-cs-skill/scripts/harness/adapter_claude.py": "905dcaee41a8a9b5629489a8734c0a5b968d8e3cc73d6c20173e0a0dbda1c2a8",
|
|
".claude/skills/eval-cs-skill/scripts/harness/adapter_codex.py": "7e6b6330b92eb6ee9a1814e22b895cbb5e970b1f16fd55ba35439c6ecf074426",
|
|
".claude/skills/eval-cs-skill/scripts/harness/adapter_paseo.py": "15ec2e456b8af9493da2b405d6379136a7872a7aaea0e7c6d8486073468c1a17",
|
|
".claude/skills/eval-cs-skill/scripts/harness/base.py": "2e440a7a2bef03f3f6db1116a7ec59b1378281d2638b9618310504c35c9be061",
|
|
".claude/skills/eval-cs-skill/scripts/harness/mock.py": "e9de0dad8dc219a27756e5827bdf4ad0dfada9bb17ca461249ff31627de93fae",
|
|
".claude/skills/eval-cs-skill/scripts/harness/mock_weak.py": "b50f2fadf541158791f918cf46dd55f30d13c1eba948bdc2f71026a3750a6234",
|
|
".claude/skills/eval-cs-skill/scripts/metrics.py": "59372930ebf1b55baeea3492055288ce4d9decf3599e919c360a2be731d505df",
|
|
".claude/skills/eval-cs-skill/scripts/probe_targets.py": "8a63559a30645c82b260ef9093ed9ce935d07703eac03ae5377b2f1f70109188",
|
|
".claude/skills/eval-cs-skill/scripts/runner.py": "a80b55d9ef1ae07fc579a9d56cd832cd08dbb3d33fb5eaea055335e6a10d5a61",
|
|
".claude/skills/eval-cs-skill/scripts/scorers/__init__.py": "917a22c5614078caeb4daf20b5f71687e16838af67a624e30e427c4834a649e0",
|
|
".claude/skills/eval-cs-skill/scripts/scorers/base.py": "dba496f77c139f516d09f7c64a1d2e38db2e5a2b981c76ce165b5d56b4218189",
|
|
".claude/skills/eval-cs-skill/scripts/scorers/learning_transfer.py": "d34ba315c065b880c0043af94a9ca87bd275d7b18e4ee3d8346858d33fdbebce",
|
|
".claude/skills/eval-cs-skill/scripts/sequence.py": "c4059892481527bce85964b80375f23524b4427b3d879fa84e7b99c821e958b5",
|
|
"experiments/seeds/dispatchboard-learning/build-seed.py": "4c7f7090abe69aca0b6be2a3568c85a91345374a4879f94da9f6704f028689b1",
|
|
"plugins/codestable/skills/cs-epic/SKILL.md": "e3f4c89ce04e41db022935bf251a012fe2cf613226d24f339569ef9d65e510dd",
|
|
"plugins/codestable/skills/cs-feat/SKILL.md": "df8d3de6f7bb4925024f24d93290c9a644bf24b2359588eeee51cc57ce9c7047",
|
|
"plugins/codestable/skills/cs-issue/SKILL.md": "fa36e43ff0785e13fcfdc222fd87b15fb22a38bbb5e9872c338a401a1320e88a",
|
|
"plugins/codestable/skills/cs-keep/SKILL.md": "1d8f56aae90ab45b0ab966ae15c42d8d958e1a3a1a29ca2e196871afda90354d",
|
|
"plugins/codestable/skills/cs-refactor/SKILL.md": "a3b5e29ae6919969e659799faed8513dbd8dad7b3713cee09aae1edcf35e7871"
|
|
},
|
|
"inputs": {
|
|
"_asset_mutations.py": "fe952ddc35ddc1fd0976627043383f9b4d4ec05fd8202ac5fe963bdd0b904210",
|
|
"checks/test_epic_sequence_a.py": "8e4da3a0323f5c5fc933e1ac51da835e082edbed4a6b27a123ba8ab94fbdce05",
|
|
"checks/test_grouping_a.py": "1824398cb4bb63393eabd02aed096ebc3d85447d27f7ec7f25d1e311dfa52dcc",
|
|
"checks/test_patch_a.py": "1f3aba4d8d0a78037d9463a472406a5d8efe2249724f85256b80e763f7bdfc43",
|
|
"checks/test_reference_job_a.py": "fafd9aff6aa8740ee82c08e4694d1b910d393daa843100db19bb2c97c33c757f",
|
|
"config.json": "709e23cded9e1a76e4cb4d712445180060050c66e7f3f07922c94ab7fa8bcdce",
|
|
"fixtures/positive/lt-epic-sequence.json": "622aba952c16b125c12e5cd278d2079faf2c07bb09a190de4dda34a14b8e9405",
|
|
"fixtures/positive/lt-feat-patch.json": "41542cbdcaf7e12ba075695cf1a1cca62379352c07c2bd7a433af11f8cb0c8a8",
|
|
"fixtures/positive/lt-issue-reference.json": "3234b15ceee742570e502953e639f14ee5cc69f5b28f51bba749a4b05f35addb",
|
|
"fixtures/positive/lt-refactor-events.json": "efebfab2c46ca01608e4c1a6ef3b7e42af328ea587eae39b5e3c466dc3b9b7bb",
|
|
"fixtures/stale/lt-stale-reference-v2.json": "8f7b21a8510e1576f70531ceab67bf86a9534ce1e534e8d683acd6243a02b13e",
|
|
"fixtures/unrelated/lt-unrelated-display.json": "c4cbc8bd7cc6e96b70853ea611547458ecd743c88058b366d1c445f607ffb40e",
|
|
"hidden/test_display_route_label.py": "ac61a72295579ffc9a0b1da6c562ff3afa172b5c6183663833a5c62524453745",
|
|
"hidden/test_group_routes.py": "2559076bb914e68e2252d452e43e70f6842c6fa7908fc3ac3abdfd2855cb1c67",
|
|
"hidden/test_patch_route.py": "a5edc897899a131e5c4735c12ce066068cb1c69fab07d156e9b133e38f0a7fb7",
|
|
"hidden/test_reference_channel.py": "99d3cf5cf66a96ec45de890c1610692266b56d4c4bc149ae7101073ea53be1ab",
|
|
"hidden/test_reference_route_v2.py": "74d8cc747446cb7d939658a6e5fe3b322de4ec1fbdd9c075ce37a368a43b56f4",
|
|
"hidden/test_route_sequence.py": "f53c2759c5d0eb9b649ba53f01259fa4e3b245a4397d39eb97e26ab34a66d29e",
|
|
"hooks/stale-reference-opaque-v2.py": "3fefea656f9317514d64eee4a0e156d69964456a0800e53bb21f3447eda814eb",
|
|
"hypotheses.md": "43b2c53a2ab34a77c64019406840c3b7821208817edc5926d8c5a6b598e32fff",
|
|
"preflight/epic-sequence-golden.py": "9a0be0b0cdccc94181791d57a6f9c0af126f6b05da22bc0ff5995d9bdb6c89d1",
|
|
"preflight/epic-sequence-naive.py": "2b5db110732e6ea4881c2cd5167f9b9a91ca677f421d907a1baf018bbc0612df",
|
|
"preflight/feat-patch-golden.py": "3d97700c0311883a0412dfcade5ccb82b9e1580d9c2e46ec8720140b784b3752",
|
|
"preflight/feat-patch-naive.py": "f5be0bca9d6540158e7a59a39a9655946864b8198626bf590be504706dcacc69",
|
|
"preflight/issue-reference-golden.py": "6be990afa89cce704e6aa0ba16a329a985f90aacca1c2265a342846b3e3023b9",
|
|
"preflight/issue-reference-naive.py": "597e0c4f8963788747c295cf0253c367002959b80e2c1899b9c13eb1e4f4ece4",
|
|
"preflight/refactor-events-golden.py": "7ddfba5bfd472dbc469758e2aada14e120f71c0f613e89c187d2fb65e0b3153f",
|
|
"preflight/refactor-events-naive.py": "1075d745aa2b701d75adb5c15387f1e39acfe39c62b01fd7128ce288e90e12da",
|
|
"preflight/stale-reference-v2-golden.py": "54b91c4f4fdefab21a30b5a80c7102fb6613f6aa1c5f5ea806066a0cf308415e",
|
|
"preflight/stale-reference-v2-naive.py": "4403c7b33da0b215dc3cbcd334b77507e9a8fe884c732a64b799db8ce2908fa3",
|
|
"preflight/unrelated-display-golden.py": "af741c7d1da7d620b51ad32b0d722310aaaccb722aa401e6b07982b9884321ac",
|
|
"preflight/unrelated-display-naive.py": "a066548343bf5d8513561beef10c281e4aa362e9c7deb34016e5d0f720413bcf",
|
|
"regression/test_seed_integrity.py": "aac084a69f2f7c5da1f6d949ea03d9ab189af5b31abdda13b701652b51c2f043"
|
|
}
|
|
}
|