@tangle-network/agent-bench 0.3.6 → 0.3.8
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +11 -0
- package/HARNESS.md +43 -0
- package/dist/adapters.js +24 -24
- package/dist/benchmarks/_harness.d.ts +1 -1
- package/dist/benchmarks/_harness.js +1 -1
- package/dist/benchmarks/aec-bench.js +2 -2
- package/dist/benchmarks/agentbench.js +2 -2
- package/dist/benchmarks/appworld.js +2 -2
- package/dist/benchmarks/bfcl.js +2 -2
- package/dist/benchmarks/commit0.js +2 -2
- package/dist/benchmarks/crag.js +2 -2
- package/dist/benchmarks/dabstep.js +2 -2
- package/dist/benchmarks/enterpriseops-gym.js +2 -2
- package/dist/benchmarks/finresearchbench.js +2 -2
- package/dist/benchmarks/humaneval.d.ts +10 -1
- package/dist/benchmarks/humaneval.js +5 -3
- package/dist/benchmarks/nomiracl.js +2 -2
- package/dist/benchmarks/open-rag-bench.js +2 -2
- package/dist/benchmarks/programbench.js +2 -2
- package/dist/benchmarks/ragbench.js +2 -2
- package/dist/benchmarks/swe-bench.js +2 -2
- package/dist/benchmarks/t2-ragbench.js +2 -2
- package/dist/benchmarks/tau-bench-shared.js +2 -2
- package/dist/benchmarks/tau2-bench.js +3 -3
- package/dist/benchmarks/tau3-banking.js +3 -3
- package/dist/benchmarks/terminal-bench.js +2 -2
- package/dist/benchmarks/toollm.js +2 -2
- package/dist/benchmarks/webarena-verified.js +2 -2
- package/dist/{chunk-CKUVRZ2T.js → chunk-3U5TXJZS.js} +2 -2
- package/dist/{chunk-PPYSEKFM.js → chunk-5H5XV76F.js} +73 -15
- package/dist/chunk-5H5XV76F.js.map +1 -0
- package/dist/{chunk-YCGY7UIZ.js → chunk-7GRVHU22.js} +2 -2
- package/dist/{chunk-Z7ML6L77.js → chunk-HWST3SED.js} +2 -2
- package/dist/{chunk-SYDW647C.js → chunk-IA2FBTWC.js} +2 -2
- package/dist/{chunk-R67DFVLO.js → chunk-IFVINJ4B.js} +2 -2
- package/dist/{chunk-R36V2VP7.js → chunk-IZ5M6OAC.js} +2 -2
- package/dist/{chunk-ODT47UAY.js → chunk-K3BQGZCT.js} +2 -2
- package/dist/{chunk-IFAV6KEM.js → chunk-KP5KD6EN.js} +2 -2
- package/dist/{chunk-ZEWMTR5M.js → chunk-MQMRLGOG.js} +2 -2
- package/dist/{chunk-TSWPNOYM.js → chunk-NQG5XDSB.js} +2 -2
- package/dist/{chunk-7WSD27QQ.js → chunk-PB64GYIG.js} +2 -2
- package/dist/{chunk-HBSWHQNJ.js → chunk-RCYQEFNX.js} +3 -3
- package/dist/{chunk-J3KDJNX2.js → chunk-RH5F53JT.js} +2 -2
- package/dist/{chunk-UAIOHCUK.js → chunk-SFLA7OH3.js} +3 -3
- package/dist/{chunk-Y6O2OCUO.js → chunk-SHM6MRRF.js} +2 -2
- package/dist/{chunk-KDIKRJGB.js → chunk-SHYIRB7I.js} +2 -2
- package/dist/{chunk-HHXFIHXC.js → chunk-SVR2LKYI.js} +2 -2
- package/dist/{chunk-5SBJCB6W.js → chunk-V7AEBY6U.js} +22 -22
- package/dist/{chunk-LRRD7NAG.js → chunk-WSKWVEQB.js} +18 -2
- package/dist/chunk-WSKWVEQB.js.map +1 -0
- package/dist/{chunk-2PVVP7GN.js → chunk-XKEFIFIC.js} +2 -2
- package/dist/{chunk-JRWWGMK7.js → chunk-XYA4XSNU.js} +2 -2
- package/dist/{chunk-X5YKXC6V.js → chunk-YSMEKBTD.js} +2 -2
- package/dist/{chunk-2XU6OGEN.js → chunk-Z4TZ76N7.js} +2 -2
- package/dist/index.js +24 -24
- package/package.json +6 -5
- package/scripts/run-package-tests.mjs +30 -8
- package/scripts/verify-packed-consumer.mjs +1 -1
- package/scripts/verify-pier-agent.mts +1 -0
- package/src/benchmarks/_harness.ts +20 -2
- package/src/benchmarks/humaneval.test.mts +122 -0
- package/src/benchmarks/humaneval.ts +100 -27
- package/src/david-attribution.mts +78 -0
- package/src/david-goliath.mts +149 -0
- package/src/hev-improve.mts +25 -6
- package/src/humaneval-object-ablation.mts +201 -0
- package/src/live-improve-campaign-mbpp.mts +641 -0
- package/src/live-improve-campaign.mts +500 -0
- package/src/mbpp-structural.mts +12 -7
- package/src/quant-arena/README.md +144 -0
- package/src/quant-arena/backtest.test.mts +135 -0
- package/src/quant-arena/backtest.ts +218 -0
- package/src/quant-arena/data.test.mts +44 -0
- package/src/quant-arena/data.ts +141 -0
- package/src/quant-arena/driver.test.mts +253 -0
- package/src/quant-arena/driver.ts +219 -0
- package/src/quant-arena/fixtures/data/PROVENANCE.md +26 -0
- package/src/quant-arena/fixtures/data/holdout/IDX.csv +523 -0
- package/src/quant-arena/fixtures/data/holdout/S01.csv +523 -0
- package/src/quant-arena/fixtures/data/holdout/S02.csv +523 -0
- package/src/quant-arena/fixtures/data/holdout/S03.csv +523 -0
- package/src/quant-arena/fixtures/data/holdout/S04.csv +523 -0
- package/src/quant-arena/fixtures/data/holdout/S05.csv +523 -0
- package/src/quant-arena/fixtures/data/holdout/S06.csv +523 -0
- package/src/quant-arena/fixtures/data/holdout/S07.csv +523 -0
- package/src/quant-arena/fixtures/data/holdout/S08.csv +523 -0
- package/src/quant-arena/fixtures/data/holdout/S09.csv +523 -0
- package/src/quant-arena/fixtures/data/holdout/S10.csv +523 -0
- package/src/quant-arena/fixtures/data/insample/IDX.csv +2087 -0
- package/src/quant-arena/fixtures/data/insample/S01.csv +2087 -0
- package/src/quant-arena/fixtures/data/insample/S02.csv +2087 -0
- package/src/quant-arena/fixtures/data/insample/S03.csv +2087 -0
- package/src/quant-arena/fixtures/data/insample/S04.csv +2087 -0
- package/src/quant-arena/fixtures/data/insample/S05.csv +2087 -0
- package/src/quant-arena/fixtures/data/insample/S06.csv +2087 -0
- package/src/quant-arena/fixtures/data/insample/S07.csv +2087 -0
- package/src/quant-arena/fixtures/data/insample/S08.csv +2087 -0
- package/src/quant-arena/fixtures/data/insample/S09.csv +2087 -0
- package/src/quant-arena/fixtures/data/insample/S10.csv +2087 -0
- package/src/quant-arena/fixtures/demo-campaign/cost-ledger.jsonl +16 -0
- package/src/quant-arena/fixtures/demo-campaign/notebook.jsonl +5 -0
- package/src/quant-arena/fixtures/demo-campaign/rollout-manifest.json +171 -0
- package/src/quant-arena/fixtures/demo-campaign/strategies/cand-001-default-author/strategy.ts +119 -0
- package/src/quant-arena/fixtures/demo-campaign/strategies/cand-002-default-author/strategy.ts +119 -0
- package/src/quant-arena/fixtures/demo-campaign/strategies/cand-003-quant-researcher/strategy.ts +105 -0
- package/src/quant-arena/fixtures/demo-campaign/strategies/cand-004-quant-researcher/strategy.ts +102 -0
- package/src/quant-arena/fixtures/demo-campaign-v2/cost-ledger.jsonl +4 -0
- package/src/quant-arena/fixtures/demo-campaign-v2/notebook.jsonl +2 -0
- package/src/quant-arena/fixtures/demo-campaign-v2/rollout-manifest.json +84 -0
- package/src/quant-arena/fixtures/demo-campaign-v2/strategies/cand-001-quant-researcher/strategy.ts +117 -0
- package/src/quant-arena/holdout-certify.mts +206 -0
- package/src/quant-arena/holdout-certify.test.mts +82 -0
- package/src/quant-arena/leak-audit.test.mts +79 -0
- package/src/quant-arena/leak-audit.ts +95 -0
- package/src/quant-arena/make-fixtures.mts +161 -0
- package/src/quant-arena/multiplicity.test.mts +68 -0
- package/src/quant-arena/multiplicity.ts +87 -0
- package/src/quant-arena/nautilus-certify.ts +31 -0
- package/src/quant-arena/oms.ts +90 -0
- package/src/quant-arena/profiles/quant-researcher.profile.json +7 -0
- package/src/quant-arena/python/pyproject.toml +8 -0
- package/src/quant-arena/python/uv.lock +1297 -0
- package/src/quant-arena/python/vbt-worker.py +192 -0
- package/src/quant-arena/quant-loop.mts +813 -0
- package/src/quant-arena/quant-loop.test.mts +75 -0
- package/src/quant-arena/strategies/buy-hold-index/strategy.ts +11 -0
- package/src/quant-arena/strategies/equal-weight/strategy.ts +20 -0
- package/src/quant-arena/strategies/sma-crossover/strategy.ts +42 -0
- package/src/quant-arena/types.ts +133 -0
- package/src/quant-arena/vbt-client.ts +321 -0
- package/src/quant-arena/vbt-parity.test.mts +183 -0
- package/src/quant-arena/windows.test.mts +45 -0
- package/src/quant-arena/windows.ts +54 -0
- package/src/rollout-ledger/backfill-swe-arena.mts +606 -0
- package/src/rollout-ledger/backfill-swe-arena.test.mts +338 -0
- package/src/rollout-ledger/settle-capture.mts +442 -0
- package/src/rollout-ledger/settle-capture.test.mts +270 -0
- package/src/stream-observe.py +45 -0
- package/src/stream-observe.tpl.html +247 -0
- package/src/supervisor-arena.mts +816 -0
- package/src/swe-arena/activation.mts +228 -0
- package/src/swe-arena/activation.test.mts +303 -0
- package/src/swe-arena/analyze.ts +211 -0
- package/src/swe-arena/arms.ts +804 -0
- package/src/swe-arena/bootstrap-meta.mts +188 -0
- package/src/swe-arena/bootstrap-meta.test.mts +51 -0
- package/src/swe-arena/briefing.mts +217 -0
- package/src/swe-arena/briefing.test.mts +178 -0
- package/src/swe-arena/calibrate.ts +217 -0
- package/src/swe-arena/capabilities.mts +76 -0
- package/src/swe-arena/capabilities.test.mts +57 -0
- package/src/swe-arena/capacity.ts +194 -0
- package/src/swe-arena/cell-evidence.mts +437 -0
- package/src/swe-arena/cell-evidence.test.mts +248 -0
- package/src/swe-arena/diagnosis-ensemble.test.mts +210 -0
- package/src/swe-arena/diagnosis-ensemble.ts +520 -0
- package/src/swe-arena/execution.test.mts +1170 -0
- package/src/swe-arena/factory-command-container.ts +284 -0
- package/src/swe-arena/factory-judge-child.mts +228 -0
- package/src/swe-arena/factory.test.mts +643 -0
- package/src/swe-arena/fixtures/analyze.py +80 -0
- package/src/swe-arena/fixtures/excludes.txt +8 -0
- package/src/swe-arena/fixtures/factory/agent-eval-309/calibration.md +51 -0
- package/src/swe-arena/fixtures/factory/agent-eval-309/manifest.json +29 -0
- package/src/swe-arena/fixtures/factory/agent-eval-309/spec.md +64 -0
- package/src/swe-arena/fixtures/factory/agent-runtime-232/calibration.md +48 -0
- package/src/swe-arena/fixtures/factory/agent-runtime-232/manifest.json +29 -0
- package/src/swe-arena/fixtures/factory/agent-runtime-232/spec.md +48 -0
- package/src/swe-arena/fixtures/factory/loops-28/calibration.md +47 -0
- package/src/swe-arena/fixtures/factory/loops-28/manifest.json +30 -0
- package/src/swe-arena/fixtures/factory/loops-28/spec.md +50 -0
- package/src/swe-arena/fixtures/gen1-salvage/README.md +45 -0
- package/src/swe-arena/fixtures/gen1-salvage/cand0-e6d7361.diff +116 -0
- package/src/swe-arena/fixtures/gen1-salvage/cand1-76a8590.diff +293 -0
- package/src/swe-arena/fixtures/holdout-preregister.log +12 -0
- package/src/swe-arena/fixtures/holdout.json +44 -0
- package/src/swe-arena/fixtures/instances.json +146 -0
- package/src/swe-arena/fixtures/ledger.jsonl +12 -0
- package/src/swe-arena/fixtures/patches/pallets__flask-5014.solo.patch +36 -0
- package/src/swe-arena/fixtures/patches/pydata__xarray-4687.sup.patch +33 -0
- package/src/swe-arena/fixtures/rejudge.jsonl +15 -0
- package/src/swe-arena/fixtures/rematch.jsonl +3 -0
- package/src/swe-arena/fixtures/rematch2.jsonl +3 -0
- package/src/swe-arena/fixtures/rematch3.jsonl +3 -0
- package/src/swe-arena/fixtures/run-report/README.md +43 -0
- package/src/swe-arena/fixtures/run-report/factory-agent-eval-309-FSUP0.json +173 -0
- package/src/swe-arena/fixtures/run-report/factory-agent-eval-309-FSUP0.md +100 -0
- package/src/swe-arena/fixtures/run-report/gen3-rollup.json +551 -0
- package/src/swe-arena/fixtures/run-report/gen3-rollup.md +64 -0
- package/src/swe-arena/fixtures/sup-journal-true.json +19 -0
- package/src/swe-arena/fixtures/verify/astropy__astropy-13033.sh +48 -0
- package/src/swe-arena/fixtures/verify/django__django-11532.sh +50 -0
- package/src/swe-arena/fixtures/verify/matplotlib__matplotlib-20826.sh +76 -0
- package/src/swe-arena/fixtures/verify/pydata__xarray-4687.sh +44 -0
- package/src/swe-arena/fixtures/verify/pytest-dev__pytest-6197.sh +32 -0
- package/src/swe-arena/fixtures/verify/sphinx-doc__sphinx-9658.sh +51 -0
- package/src/swe-arena/fixtures/worker-tokens.json +42 -0
- package/src/swe-arena/fixtures.ts +237 -0
- package/src/swe-arena/gepa-seat.mts +583 -0
- package/src/swe-arena/gepa-seat.test.mts +635 -0
- package/src/swe-arena/holdout-certify.mts +408 -0
- package/src/swe-arena/holdout-certify.test.mts +160 -0
- package/src/swe-arena/judge-child.mts +37 -0
- package/src/swe-arena/ledger-orphans.mts +77 -0
- package/src/swe-arena/ledger-orphans.test.mts +147 -0
- package/src/swe-arena/lineage-record.mts +164 -0
- package/src/swe-arena/lineage-record.test.mts +115 -0
- package/src/swe-arena/manifest.mts +293 -0
- package/src/swe-arena/manifest.test.mts +169 -0
- package/src/swe-arena/materialize.ts +142 -0
- package/src/swe-arena/outer-loop.mts +2795 -0
- package/src/swe-arena/outer-loop.test.mts +696 -0
- package/src/swe-arena/parity.test.mts +87 -0
- package/src/swe-arena/premeasured-from-cells.mts +281 -0
- package/src/swe-arena/premeasured-from-cells.test.mts +180 -0
- package/src/swe-arena/proc.test.mts +174 -0
- package/src/swe-arena/proc.ts +260 -0
- package/src/swe-arena/profiles/default-author.profile.json +4 -0
- package/src/swe-arena/proposer-fanout.mts +770 -0
- package/src/swe-arena/proposer-fanout.test.mts +619 -0
- package/src/swe-arena/proposer-provenance.mts +177 -0
- package/src/swe-arena/proposer-provenance.test.mts +106 -0
- package/src/swe-arena/reconcile.ts +0 -0
- package/src/swe-arena/replay.mts +183 -0
- package/src/swe-arena/replay.test.mts +300 -0
- package/src/swe-arena/run-experiment.mts +727 -0
- package/src/swe-arena/run-report.mts +75 -0
- package/src/swe-arena/run-supervisor.mjs +297 -0
- package/src/swe-arena/run-supervisor.test.mts +500 -0
- package/src/swe-arena/score-split.mts +140 -0
- package/src/swe-arena/score-split.test.mts +123 -0
- package/src/swe-arena/serialized-judge.ts +414 -0
- package/src/swe-arena/types.ts +218 -0
- package/src/swe-code-improve.mts +328 -0
- package/src/swe-emit-patch.mts +104 -0
- package/src/swe-improve.mts +232 -0
- package/src/swe-jail.ts +2 -2
- package/src/swe-local-proof.mts +169 -0
- package/src/swe-repro-calibrate.mts +446 -0
- package/src/swe-stream.mts +1497 -0
- package/src/swe-structural.mts +245 -837
- package/dist/chunk-LRRD7NAG.js.map +0 -1
- package/dist/chunk-PPYSEKFM.js.map +0 -1
- /package/dist/{chunk-CKUVRZ2T.js.map → chunk-3U5TXJZS.js.map} +0 -0
- /package/dist/{chunk-YCGY7UIZ.js.map → chunk-7GRVHU22.js.map} +0 -0
- /package/dist/{chunk-Z7ML6L77.js.map → chunk-HWST3SED.js.map} +0 -0
- /package/dist/{chunk-SYDW647C.js.map → chunk-IA2FBTWC.js.map} +0 -0
- /package/dist/{chunk-R67DFVLO.js.map → chunk-IFVINJ4B.js.map} +0 -0
- /package/dist/{chunk-R36V2VP7.js.map → chunk-IZ5M6OAC.js.map} +0 -0
- /package/dist/{chunk-ODT47UAY.js.map → chunk-K3BQGZCT.js.map} +0 -0
- /package/dist/{chunk-IFAV6KEM.js.map → chunk-KP5KD6EN.js.map} +0 -0
- /package/dist/{chunk-ZEWMTR5M.js.map → chunk-MQMRLGOG.js.map} +0 -0
- /package/dist/{chunk-TSWPNOYM.js.map → chunk-NQG5XDSB.js.map} +0 -0
- /package/dist/{chunk-7WSD27QQ.js.map → chunk-PB64GYIG.js.map} +0 -0
- /package/dist/{chunk-HBSWHQNJ.js.map → chunk-RCYQEFNX.js.map} +0 -0
- /package/dist/{chunk-J3KDJNX2.js.map → chunk-RH5F53JT.js.map} +0 -0
- /package/dist/{chunk-UAIOHCUK.js.map → chunk-SFLA7OH3.js.map} +0 -0
- /package/dist/{chunk-Y6O2OCUO.js.map → chunk-SHM6MRRF.js.map} +0 -0
- /package/dist/{chunk-KDIKRJGB.js.map → chunk-SHYIRB7I.js.map} +0 -0
- /package/dist/{chunk-HHXFIHXC.js.map → chunk-SVR2LKYI.js.map} +0 -0
- /package/dist/{chunk-5SBJCB6W.js.map → chunk-V7AEBY6U.js.map} +0 -0
- /package/dist/{chunk-2PVVP7GN.js.map → chunk-XKEFIFIC.js.map} +0 -0
- /package/dist/{chunk-JRWWGMK7.js.map → chunk-XYA4XSNU.js.map} +0 -0
- /package/dist/{chunk-X5YKXC6V.js.map → chunk-YSMEKBTD.js.map} +0 -0
- /package/dist/{chunk-2XU6OGEN.js.map → chunk-Z4TZ76N7.js.map} +0 -0
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/benchmarks/_harness.ts"],"sourcesContent":["/**\n * Shared code-bench harness. The \"stage the artifact → run the benchmark's own\n * evaluator in an external process (mkdtemp / execFile / .venv python / Docker)\n * → read its JSON report → { resolved, score }\" spine that swe-bench,\n * terminal-bench, commit0, programbench and aec-bench all need. Factored out so\n * the Docker/venv/report-reading logic lives in ONE place instead of being\n * copy-pasted per adapter.\n *\n * It owns NO benchmark policy: each adapter passes the argv for its evaluator\n * and a `parseReport` that maps that evaluator's report JSON → a BenchScore. The\n * harness owns process spawning, temp-dir lifecycle, large-buffer/timeout config\n * and fail-loud diagnostics.\n *\n * Fail-loud: an absent .venv / harness / Docker daemon THROWS from `preflight`\n * (the adapter passes the import line + the exact fix). A staged run that exits\n * nonzero throws with the captured stderr — never a fabricated score.\n */\n\nimport { execFile, spawn } from 'node:child_process'\nimport { createHash } from 'node:crypto'\nimport {\n cp,\n lstat,\n mkdir,\n mkdtemp,\n readFile,\n readlink,\n readdir,\n rename,\n rm,\n writeFile,\n} from 'node:fs/promises'\nimport { tmpdir } from 'node:os'\nimport { basename, dirname, isAbsolute, join, relative, resolve, sep } from 'node:path'\nimport { fileURLToPath } from 'node:url'\nimport { promisify } from 'node:util'\nimport type {\n BenchScore,\n JudgeArtifactFileReceipt,\n JudgeArtifactReceipt,\n} from './types'\n\nconst execFileAsync = promisify(execFile)\n\n/** Repo root for the bench package (…/bench), so `.venv` and `fixtures` resolve. */\nexport const benchRoot = fileURLToPath(new URL('../..', import.meta.url))\n\n/** Resolve the shared interpreter without requiring an installed package to contain a venv. */\nexport function resolveBenchPython(\n env: Readonly<{ AGENT_BENCH_PYTHON?: string }> = process.env,\n root: string = benchRoot,\n): string {\n const configured = env.AGENT_BENCH_PYTHON\n if (configured === undefined) return join(root, '.venv', 'bin', 'python')\n if (!isAbsolute(configured)) {\n throw new Error('AGENT_BENCH_PYTHON must be an absolute path')\n }\n return configured\n}\n\n/** The shared interpreter every Python-backed evaluator runs through. */\nexport const venvPython = resolveBenchPython()\n\n/** Interpreter for a NAMED isolated venv (e.g. `.venv-commit0`). Benches whose pip\n * deps conflict with the shared `.venv` (commit0 downgrades pydantic/sqlalchemy)\n * get their own venv and pass its python explicitly — keeping the shared one clean. */\nexport const venvPythonAt = (venvDir: string): string => join(benchRoot, venvDir, 'bin', 'python')\n/** Report/transcript reads are large; 256 MiB matches the SWE harness budget. */\nexport const bigBuffer = 1024 * 1024 * 256\n\n/** Path to a named executable inside the bench venv (e.g. `venvBin('tb')`). */\nexport function venvBin(name: string): string {\n return join(benchRoot, '.venv', 'bin', name)\n}\n\n/**\n * Run the bench venv python with an inline script (`-c`); return stdout. Throws\n * (with stderr) on a nonzero exit — the loaders rely on this to fail loud rather\n * than parse a partial dump.\n */\nexport async function runVenvPython(\n script: string,\n args: string[] = [],\n timeoutMs = 0,\n python: string = venvPython,\n): Promise<string> {\n const { stdout } = await execFileAsync(python, ['-c', script, ...args], {\n maxBuffer: bigBuffer,\n timeout: timeoutMs,\n })\n return stdout\n}\n\n/**\n * Preflight a python-backed harness: import the module(s) and (optionally) ping\n * Docker, all inside the bench venv. On failure THROWS the captured error joined\n * to the adapter's `fix` guidance — the contract every code-bench preflight wants.\n */\nexport async function preflightVenvImports(opts: {\n /** Module names to `import` (e.g. ['swebench']); '' entries are ignored. */\n modules: string[]\n /** Also `docker.from_env().ping()` — true for Docker-backed evaluators. */\n requireDocker?: boolean\n /** Actionable remediation appended to the thrown message. */\n fix: string\n /** Override the interpreter (e.g. an isolated `.venv-commit0`). Default: shared `.venv`. */\n python?: string\n}): Promise<void> {\n const imports = opts.modules.filter((m) => m.length > 0)\n const lines = [...imports.map((m) => `import ${m}`)]\n if (opts.requireDocker) lines.push('import docker', 'docker.from_env().ping()')\n lines.push(\"print('ok')\")\n try {\n await runVenvPython(lines.join('\\n'), [], 0, opts.python ?? venvPython)\n } catch (err) {\n const msg = err instanceof Error ? err.message : String(err)\n throw new Error(`${msg}\\n${opts.fix}`)\n }\n}\n\n/**\n * Run a bench-local python driver script (in the bench venv) while piping\n * `input` to its stdin, returning stdout. The driver's verdict JSON is its LAST\n * stdout line; callers parse that and inspect an `error` field (fail loud).\n *\n * Uses spawn + an explicit `stdin.end(input)` rather than promisify(execFile)'s\n * `input` option, because that option is NOT honored by async execFile — stdin\n * is left open and a driver that does `sys.stdin.read()` blocks forever. The\n * artifact-piping judges (commit0, appworld) MUST go through this.\n */\nexport function runVenvScriptStdin(\n scriptPath: string,\n args: string[],\n input: string,\n opts: { cwd?: string; timeoutMs?: number; python?: string } = {},\n): Promise<string> {\n return new Promise<string>((resolve, reject) => {\n const child = spawn(opts.python ?? venvPython, [scriptPath, ...args], {\n cwd: opts.cwd ?? benchRoot,\n ...(opts.timeoutMs ? { timeout: opts.timeoutMs } : {}),\n })\n let stdout = ''\n let stderr = ''\n let bytes = 0\n child.stdout.on('data', (c: Buffer) => {\n bytes += c.length\n if (bytes <= bigBuffer) stdout += c.toString('utf8')\n })\n child.stderr.on('data', (c: Buffer) => {\n stderr += c.toString('utf8')\n })\n child.on('error', (err) => reject(err))\n child.on('close', (code) => {\n if (code === 0) resolve(stdout)\n else reject(new Error((stderr || stdout || `exit ${code}`).slice(0, 1500)))\n })\n child.stdin.end(input)\n })\n}\n\nexport interface StagedRunSpec {\n /** mkdtemp prefix, e.g. 'swebench-' / 'commit0-'. */\n tmpPrefix: string\n /**\n * Write the artifact + any harness inputs into the temp dir. Returns nothing;\n * `argv`/`cwd` consume `dir` to point the evaluator at what was written.\n */\n stage(dir: string): Promise<void>\n /** The external evaluator to spawn. `bin` defaults to the bench venv python. */\n bin?: string\n /** argv for the evaluator, computed from the temp `dir`. */\n argv(dir: string): string[]\n /** Working directory for the evaluator. Defaults to the temp `dir`. */\n cwd?(dir: string): string\n /** Hard timeout for the evaluator (ms); 0 = none. */\n timeoutMs?: number\n /**\n * Read the evaluator's report(s) out of `dir` and map to a BenchScore. Throws\n * if the expected report is absent/malformed (fail loud — no default score).\n */\n parseReport(dir: string): Promise<BenchScore>\n /**\n * Copy the complete evaluator directory plus raw process stdout/stderr to this\n * caller-owned directory before cleanup. The destination must not exist.\n */\n capture?: StagedRunCaptureSpec\n /** Keep the temp dir on disk (debugging). Default false → always cleaned up. */\n keepTmp?: boolean\n}\n\nexport interface StagedRunCaptureSpec {\n /** Destination for `evaluator/`, `process/`, and the hashed `receipt.json`. */\n destination: string\n}\n\n/** A staged run failed after any requested evidence was durably retained. */\nexport class StagedJudgeError extends Error {\n readonly judgeArtifacts?: JudgeArtifactReceipt\n\n constructor(message: string, judgeArtifacts?: JudgeArtifactReceipt, options?: ErrorOptions) {\n super(message, options)\n this.name = 'StagedJudgeError'\n this.judgeArtifacts = judgeArtifacts\n }\n}\n\nfunction sha256(bytes: Uint8Array): `sha256:${string}` {\n return `sha256:${createHash('sha256').update(bytes).digest('hex')}`\n}\n\nfunction portablePath(path: string): string {\n return path.split(sep).join('/')\n}\n\nfunction compareText(left: string, right: string): number {\n return left < right ? -1 : left > right ? 1 : 0\n}\n\nasync function collectArtifactFiles(\n root: string,\n current: string,\n): Promise<JudgeArtifactFileReceipt[]> {\n const absolute = join(root, current)\n const entries = await readdir(absolute, { withFileTypes: true })\n const files: JudgeArtifactFileReceipt[] = []\n for (const entry of entries.sort((left, right) => compareText(left.name, right.name))) {\n const relativePath = join(current, entry.name)\n const path = join(root, relativePath)\n if (entry.isDirectory()) {\n files.push(...await collectArtifactFiles(root, relativePath))\n continue\n }\n if (entry.isFile()) {\n const bytes = await readFile(path)\n files.push({\n path: portablePath(relativePath),\n byteLength: bytes.byteLength,\n sha256: sha256(bytes),\n kind: 'file',\n })\n continue\n }\n if (entry.isSymbolicLink()) {\n const targetBytes = await readlink(path, { encoding: 'buffer' })\n files.push({\n path: portablePath(relativePath),\n byteLength: targetBytes.byteLength,\n sha256: sha256(targetBytes),\n kind: 'symlink',\n })\n continue\n }\n throw new Error(`staged judge capture does not support ${relativePath}`)\n }\n return files\n}\n\nfunction isWithin(parent: string, candidate: string): boolean {\n const path = relative(parent, candidate)\n return path === '' || (!path.startsWith(`..${sep}`) && path !== '..' && !isAbsolute(path))\n}\n\nasync function assertDestinationAbsent(destination: string): Promise<void> {\n try {\n await lstat(destination)\n } catch (error) {\n if ((error as NodeJS.ErrnoException).code === 'ENOENT') return\n throw error\n }\n throw new Error(`staged judge capture destination already exists: ${destination}`)\n}\n\nasync function captureStagedRun(\n sourceDirectory: string,\n spec: StagedRunCaptureSpec,\n processOutput: Readonly<{ stdout: Buffer; stderr: Buffer }>,\n evaluatorSucceeded: boolean,\n): Promise<JudgeArtifactReceipt> {\n const source = resolve(sourceDirectory)\n const destination = resolve(spec.destination)\n if (isWithin(source, destination)) {\n throw new Error('staged judge capture destination must be outside the evaluator directory')\n }\n await mkdir(dirname(destination), { recursive: true })\n await assertDestinationAbsent(destination)\n const staging = await mkdtemp(join(dirname(destination), `.${basename(destination)}-capture-`))\n try {\n await cp(source, join(staging, 'evaluator'), {\n recursive: true,\n errorOnExist: true,\n force: false,\n preserveTimestamps: true,\n verbatimSymlinks: true,\n })\n await mkdir(join(staging, 'process'))\n await writeFile(join(staging, 'process', 'stdout.bin'), processOutput.stdout)\n await writeFile(join(staging, 'process', 'stderr.bin'), processOutput.stderr)\n\n const files = [\n ...await collectArtifactFiles(staging, 'evaluator'),\n ...await collectArtifactFiles(staging, 'process'),\n ].sort((left, right) => compareText(left.path, right.path))\n const byteLength = files.reduce((total, file) => total + file.byteLength, 0)\n const treeBytes = Buffer.from(\n files\n .map((file) => `${file.path}\\0${file.kind}\\0${file.byteLength}\\0${file.sha256}\\n`)\n .join(''),\n 'utf8',\n )\n const receipt: JudgeArtifactReceipt = {\n schema: 'agent-bench/judge-artifacts/v1',\n directory: destination,\n evaluatorDirectory: join(destination, 'evaluator'),\n manifestPath: join(destination, 'receipt.json'),\n evaluatorSucceeded,\n files,\n fileCount: files.length,\n byteLength,\n treeSha256: sha256(treeBytes),\n }\n await writeFile(join(staging, 'receipt.json'), `${JSON.stringify(receipt, null, 2)}\\n`)\n await rename(staging, destination)\n return receipt\n } catch (error) {\n await rm(staging, { recursive: true, force: true }).catch(() => {})\n throw error\n }\n}\n\nfunction processBytes(value: unknown): Buffer {\n if (Buffer.isBuffer(value)) return value\n if (value === undefined || value === null) return Buffer.alloc(0)\n return Buffer.from(String(value), 'utf8')\n}\n\n/**\n * The shared judge body: mkdtemp → stage → spawn evaluator → parseReport →\n * cleanup. The evaluator's stdout/stderr is surfaced on failure; the temp dir is\n * always removed in `finally` unless `keepTmp`.\n */\nexport async function runStagedJudge(spec: StagedRunSpec): Promise<BenchScore> {\n const dir = await mkdtemp(join(tmpdir(), spec.tmpPrefix))\n let stdout: Buffer<ArrayBufferLike> = Buffer.alloc(0)\n let stderr: Buffer<ArrayBufferLike> = Buffer.alloc(0)\n let evaluatorSucceeded = false\n let score: BenchScore | undefined\n let failure: unknown\n try {\n try {\n await spec.stage(dir)\n const bin = spec.bin ?? venvPython\n const argv = spec.argv(dir)\n try {\n const output = await execFileAsync(bin, argv, {\n cwd: spec.cwd ? spec.cwd(dir) : dir,\n encoding: 'buffer',\n maxBuffer: bigBuffer,\n ...(spec.timeoutMs ? { timeout: spec.timeoutMs } : {}),\n })\n stdout = processBytes(output.stdout)\n stderr = processBytes(output.stderr)\n evaluatorSucceeded = true\n } catch (err) {\n const e = err as { stderr?: unknown; stdout?: unknown; message?: string }\n stdout = processBytes(e.stdout)\n stderr = processBytes(e.stderr)\n const detail = processBytes(e.stderr ?? e.stdout ?? e.message ?? String(err))\n .toString('utf8')\n .slice(0, 2000)\n throw new Error(`${spec.tmpPrefix.replace(/-$/, '')} evaluator failed (${bin} ${argv.join(' ')}):\\n${detail}`)\n }\n score = await spec.parseReport(dir)\n } catch (err) {\n failure = err\n }\n } finally {\n let judgeArtifacts: JudgeArtifactReceipt | undefined\n if (spec.capture) {\n try {\n judgeArtifacts = await captureStagedRun(\n dir,\n spec.capture,\n { stdout, stderr },\n evaluatorSucceeded,\n )\n } catch (captureError) {\n failure = new Error(\n `staged judge failed to capture evaluator artifacts: ${captureError instanceof Error ? captureError.message : captureError}`,\n { cause: failure ?? captureError },\n )\n }\n }\n if (!spec.keepTmp) await rm(dir, { recursive: true, force: true }).catch(() => {})\n if (failure) {\n throw new StagedJudgeError(\n failure instanceof Error ? failure.message : String(failure),\n judgeArtifacts,\n { cause: failure },\n )\n }\n if (!score) throw new StagedJudgeError('staged judge completed without a score', judgeArtifacts)\n return judgeArtifacts ? { ...score, judgeArtifacts } : score\n }\n}\n\n/** Read + JSON.parse a report file from a staged run; throws with the path on failure. */\nexport async function readJsonReport<T>(path: string): Promise<T> {\n let raw: string\n try {\n raw = await readFile(path, 'utf8')\n } catch (err) {\n throw new Error(`expected report not written: ${path} (${err instanceof Error ? err.message : err})`)\n }\n try {\n return JSON.parse(raw) as T\n } catch (err) {\n throw new Error(`report not valid JSON at ${path}: ${err instanceof Error ? err.message : err}`)\n }\n}\n\n/** Write a UTF-8 file into a staged dir (artifact / preds.json / attempt.sh / …). */\nexport async function stageFile(path: string, content: string): Promise<void> {\n await writeFile(path, content)\n}\n\n/** Sanitize an instance id into a filesystem/run-id-safe token. */\nexport function safeRunId(prefix: string, id: string): string {\n return `${prefix}-${id}`.replace(/[^a-zA-Z0-9_.-]/g, '_')\n}\n"],"mappings":";AAkBA,SAAS,UAAU,aAAa;AAChC,SAAS,kBAAkB;AAC3B;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP,SAAS,cAAc;AACvB,SAAS,UAAU,SAAS,YAAY,MAAM,UAAU,SAAS,WAAW;AAC5E,SAAS,qBAAqB;AAC9B,SAAS,iBAAiB;AAO1B,IAAM,gBAAgB,UAAU,QAAQ;AAGjC,IAAM,YAAY,cAAc,IAAI,IAAI,SAAS,YAAY,GAAG,CAAC;AAGjE,SAAS,mBACd,MAAiD,QAAQ,KACzD,OAAe,WACP;AACR,QAAM,aAAa,IAAI;AACvB,MAAI,eAAe,OAAW,QAAO,KAAK,MAAM,SAAS,OAAO,QAAQ;AACxE,MAAI,CAAC,WAAW,UAAU,GAAG;AAC3B,UAAM,IAAI,MAAM,6CAA6C;AAAA,EAC/D;AACA,SAAO;AACT;AAGO,IAAM,aAAa,mBAAmB;AAKtC,IAAM,eAAe,CAAC,YAA4B,KAAK,WAAW,SAAS,OAAO,QAAQ;AAE1F,IAAM,YAAY,OAAO,OAAO;AAGhC,SAAS,QAAQ,MAAsB;AAC5C,SAAO,KAAK,WAAW,SAAS,OAAO,IAAI;AAC7C;AAOA,eAAsB,cACpB,QACA,OAAiB,CAAC,GAClB,YAAY,GACZ,SAAiB,YACA;AACjB,QAAM,EAAE,OAAO,IAAI,MAAM,cAAc,QAAQ,CAAC,MAAM,QAAQ,GAAG,IAAI,GAAG;AAAA,IACtE,WAAW;AAAA,IACX,SAAS;AAAA,EACX,CAAC;AACD,SAAO;AACT;AAOA,eAAsB,qBAAqB,MASzB;AAChB,QAAM,UAAU,KAAK,QAAQ,OAAO,CAAC,MAAM,EAAE,SAAS,CAAC;AACvD,QAAM,QAAQ,CAAC,GAAG,QAAQ,IAAI,CAAC,MAAM,UAAU,CAAC,EAAE,CAAC;AACnD,MAAI,KAAK,cAAe,OAAM,KAAK,iBAAiB,0BAA0B;AAC9E,QAAM,KAAK,aAAa;AACxB,MAAI;AACF,UAAM,cAAc,MAAM,KAAK,IAAI,GAAG,CAAC,GAAG,GAAG,KAAK,UAAU,UAAU;AAAA,EACxE,SAAS,KAAK;AACZ,UAAM,MAAM,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAC3D,UAAM,IAAI,MAAM,GAAG,GAAG;AAAA,EAAK,KAAK,GAAG,EAAE;AAAA,EACvC;AACF;AAYO,SAAS,mBACd,YACA,MACA,OACA,OAA8D,CAAC,GAC9C;AACjB,SAAO,IAAI,QAAgB,CAACA,UAAS,WAAW;AAC9C,UAAM,QAAQ,MAAM,KAAK,UAAU,YAAY,CAAC,YAAY,GAAG,IAAI,GAAG;AAAA,MACpE,KAAK,KAAK,OAAO;AAAA,MACjB,GAAI,KAAK,YAAY,EAAE,SAAS,KAAK,UAAU,IAAI,CAAC;AAAA,IACtD,CAAC;AACD,QAAI,SAAS;AACb,QAAI,SAAS;AACb,QAAI,QAAQ;AACZ,UAAM,OAAO,GAAG,QAAQ,CAAC,MAAc;AACrC,eAAS,EAAE;AACX,UAAI,SAAS,UAAW,WAAU,EAAE,SAAS,MAAM;AAAA,IACrD,CAAC;AACD,UAAM,OAAO,GAAG,QAAQ,CAAC,MAAc;AACrC,gBAAU,EAAE,SAAS,MAAM;AAAA,IAC7B,CAAC;AACD,UAAM,GAAG,SAAS,CAAC,QAAQ,OAAO,GAAG,CAAC;AACtC,UAAM,GAAG,SAAS,CAAC,SAAS;AAC1B,UAAI,SAAS,EAAG,CAAAA,SAAQ,MAAM;AAAA,UACzB,QAAO,IAAI,OAAO,UAAU,UAAU,QAAQ,IAAI,IAAI,MAAM,GAAG,IAAI,CAAC,CAAC;AAAA,IAC5E,CAAC;AACD,UAAM,MAAM,IAAI,KAAK;AAAA,EACvB,CAAC;AACH;AAsCO,IAAM,mBAAN,cAA+B,MAAM;AAAA,EACjC;AAAA,EAET,YAAY,SAAiB,gBAAuC,SAAwB;AAC1F,UAAM,SAAS,OAAO;AACtB,SAAK,OAAO;AACZ,SAAK,iBAAiB;AAAA,EACxB;AACF;AAEA,SAAS,OAAO,OAAuC;AACrD,SAAO,UAAU,WAAW,QAAQ,EAAE,OAAO,KAAK,EAAE,OAAO,KAAK,CAAC;AACnE;AAEA,SAAS,aAAa,MAAsB;AAC1C,SAAO,KAAK,MAAM,GAAG,EAAE,KAAK,GAAG;AACjC;AAEA,SAAS,YAAY,MAAc,OAAuB;AACxD,SAAO,OAAO,QAAQ,KAAK,OAAO,QAAQ,IAAI;AAChD;AAEA,eAAe,qBACb,MACA,SACqC;AACrC,QAAM,WAAW,KAAK,MAAM,OAAO;AACnC,QAAM,UAAU,MAAM,QAAQ,UAAU,EAAE,eAAe,KAAK,CAAC;AAC/D,QAAM,QAAoC,CAAC;AAC3C,aAAW,SAAS,QAAQ,KAAK,CAAC,MAAM,UAAU,YAAY,KAAK,MAAM,MAAM,IAAI,CAAC,GAAG;AACrF,UAAM,eAAe,KAAK,SAAS,MAAM,IAAI;AAC7C,UAAM,OAAO,KAAK,MAAM,YAAY;AACpC,QAAI,MAAM,YAAY,GAAG;AACvB,YAAM,KAAK,GAAG,MAAM,qBAAqB,MAAM,YAAY,CAAC;AAC5D;AAAA,IACF;AACA,QAAI,MAAM,OAAO,GAAG;AAClB,YAAM,QAAQ,MAAM,SAAS,IAAI;AACjC,YAAM,KAAK;AAAA,QACT,MAAM,aAAa,YAAY;AAAA,QAC/B,YAAY,MAAM;AAAA,QAClB,QAAQ,OAAO,KAAK;AAAA,QACpB,MAAM;AAAA,MACR,CAAC;AACD;AAAA,IACF;AACA,QAAI,MAAM,eAAe,GAAG;AAC1B,YAAM,cAAc,MAAM,SAAS,MAAM,EAAE,UAAU,SAAS,CAAC;AAC/D,YAAM,KAAK;AAAA,QACT,MAAM,aAAa,YAAY;AAAA,QAC/B,YAAY,YAAY;AAAA,QACxB,QAAQ,OAAO,WAAW;AAAA,QAC1B,MAAM;AAAA,MACR,CAAC;AACD;AAAA,IACF;AACA,UAAM,IAAI,MAAM,yCAAyC,YAAY,EAAE;AAAA,EACzE;AACA,SAAO;AACT;AAEA,SAAS,SAAS,QAAgB,WAA4B;AAC5D,QAAM,OAAO,SAAS,QAAQ,SAAS;AACvC,SAAO,SAAS,MAAO,CAAC,KAAK,WAAW,KAAK,GAAG,EAAE,KAAK,SAAS,QAAQ,CAAC,WAAW,IAAI;AAC1F;AAEA,eAAe,wBAAwB,aAAoC;AACzE,MAAI;AACF,UAAM,MAAM,WAAW;AAAA,EACzB,SAAS,OAAO;AACd,QAAK,MAAgC,SAAS,SAAU;AACxD,UAAM;AAAA,EACR;AACA,QAAM,IAAI,MAAM,oDAAoD,WAAW,EAAE;AACnF;AAEA,eAAe,iBACb,iBACA,MACA,eACA,oBAC+B;AAC/B,QAAM,SAAS,QAAQ,eAAe;AACtC,QAAM,cAAc,QAAQ,KAAK,WAAW;AAC5C,MAAI,SAAS,QAAQ,WAAW,GAAG;AACjC,UAAM,IAAI,MAAM,0EAA0E;AAAA,EAC5F;AACA,QAAM,MAAM,QAAQ,WAAW,GAAG,EAAE,WAAW,KAAK,CAAC;AACrD,QAAM,wBAAwB,WAAW;AACzC,QAAM,UAAU,MAAM,QAAQ,KAAK,QAAQ,WAAW,GAAG,IAAI,SAAS,WAAW,CAAC,WAAW,CAAC;AAC9F,MAAI;AACF,UAAM,GAAG,QAAQ,KAAK,SAAS,WAAW,GAAG;AAAA,MAC3C,WAAW;AAAA,MACX,cAAc;AAAA,MACd,OAAO;AAAA,MACP,oBAAoB;AAAA,MACpB,kBAAkB;AAAA,IACpB,CAAC;AACD,UAAM,MAAM,KAAK,SAAS,SAAS,CAAC;AACpC,UAAM,UAAU,KAAK,SAAS,WAAW,YAAY,GAAG,cAAc,MAAM;AAC5E,UAAM,UAAU,KAAK,SAAS,WAAW,YAAY,GAAG,cAAc,MAAM;AAE5E,UAAM,QAAQ;AAAA,MACZ,GAAG,MAAM,qBAAqB,SAAS,WAAW;AAAA,MAClD,GAAG,MAAM,qBAAqB,SAAS,SAAS;AAAA,IAClD,EAAE,KAAK,CAAC,MAAM,UAAU,YAAY,KAAK,MAAM,MAAM,IAAI,CAAC;AAC1D,UAAM,aAAa,MAAM,OAAO,CAAC,OAAO,SAAS,QAAQ,KAAK,YAAY,CAAC;AAC3E,UAAM,YAAY,OAAO;AAAA,MACvB,MACG,IAAI,CAAC,SAAS,GAAG,KAAK,IAAI,KAAK,KAAK,IAAI,KAAK,KAAK,UAAU,KAAK,KAAK,MAAM;AAAA,CAAI,EAChF,KAAK,EAAE;AAAA,MACV;AAAA,IACF;AACA,UAAM,UAAgC;AAAA,MACpC,QAAQ;AAAA,MACR,WAAW;AAAA,MACX,oBAAoB,KAAK,aAAa,WAAW;AAAA,MACjD,cAAc,KAAK,aAAa,cAAc;AAAA,MAC9C;AAAA,MACA;AAAA,MACA,WAAW,MAAM;AAAA,MACjB;AAAA,MACA,YAAY,OAAO,SAAS;AAAA,IAC9B;AACA,UAAM,UAAU,KAAK,SAAS,cAAc,GAAG,GAAG,KAAK,UAAU,SAAS,MAAM,CAAC,CAAC;AAAA,CAAI;AACtF,UAAM,OAAO,SAAS,WAAW;AACjC,WAAO;AAAA,EACT,SAAS,OAAO;AACd,UAAM,GAAG,SAAS,EAAE,WAAW,MAAM,OAAO,KAAK,CAAC,EAAE,MAAM,MAAM;AAAA,IAAC,CAAC;AAClE,UAAM;AAAA,EACR;AACF;AAEA,SAAS,aAAa,OAAwB;AAC5C,MAAI,OAAO,SAAS,KAAK,EAAG,QAAO;AACnC,MAAI,UAAU,UAAa,UAAU,KAAM,QAAO,OAAO,MAAM,CAAC;AAChE,SAAO,OAAO,KAAK,OAAO,KAAK,GAAG,MAAM;AAC1C;AAOA,eAAsB,eAAe,MAA0C;AAC7E,QAAM,MAAM,MAAM,QAAQ,KAAK,OAAO,GAAG,KAAK,SAAS,CAAC;AACxD,MAAI,SAAkC,OAAO,MAAM,CAAC;AACpD,MAAI,SAAkC,OAAO,MAAM,CAAC;AACpD,MAAI,qBAAqB;AACzB,MAAI;AACJ,MAAI;AACJ,MAAI;AACF,QAAI;AACF,YAAM,KAAK,MAAM,GAAG;AACpB,YAAM,MAAM,KAAK,OAAO;AACxB,YAAM,OAAO,KAAK,KAAK,GAAG;AAC1B,UAAI;AACF,cAAM,SAAS,MAAM,cAAc,KAAK,MAAM;AAAA,UAC5C,KAAK,KAAK,MAAM,KAAK,IAAI,GAAG,IAAI;AAAA,UAChC,UAAU;AAAA,UACV,WAAW;AAAA,UACX,GAAI,KAAK,YAAY,EAAE,SAAS,KAAK,UAAU,IAAI,CAAC;AAAA,QACtD,CAAC;AACD,iBAAS,aAAa,OAAO,MAAM;AACnC,iBAAS,aAAa,OAAO,MAAM;AACnC,6BAAqB;AAAA,MACvB,SAAS,KAAK;AACZ,cAAM,IAAI;AACV,iBAAS,aAAa,EAAE,MAAM;AAC9B,iBAAS,aAAa,EAAE,MAAM;AAC9B,cAAM,SAAS,aAAa,EAAE,UAAU,EAAE,UAAU,EAAE,WAAW,OAAO,GAAG,CAAC,EACzE,SAAS,MAAM,EACf,MAAM,GAAG,GAAI;AAChB,cAAM,IAAI,MAAM,GAAG,KAAK,UAAU,QAAQ,MAAM,EAAE,CAAC,sBAAsB,GAAG,IAAI,KAAK,KAAK,GAAG,CAAC;AAAA,EAAO,MAAM,EAAE;AAAA,MAC/G;AACA,cAAQ,MAAM,KAAK,YAAY,GAAG;AAAA,IACpC,SAAS,KAAK;AACZ,gBAAU;AAAA,IACZ;AAAA,EACF,UAAE;AACA,QAAI;AACJ,QAAI,KAAK,SAAS;AAChB,UAAI;AACF,yBAAiB,MAAM;AAAA,UACrB;AAAA,UACA,KAAK;AAAA,UACL,EAAE,QAAQ,OAAO;AAAA,UACjB;AAAA,QACF;AAAA,MACF,SAAS,cAAc;AACrB,kBAAU,IAAI;AAAA,UACZ,uDAAuD,wBAAwB,QAAQ,aAAa,UAAU,YAAY;AAAA,UAC1H,EAAE,OAAO,WAAW,aAAa;AAAA,QACnC;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,KAAK,QAAS,OAAM,GAAG,KAAK,EAAE,WAAW,MAAM,OAAO,KAAK,CAAC,EAAE,MAAM,MAAM;AAAA,IAAC,CAAC;AACjF,QAAI,SAAS;AACX,YAAM,IAAI;AAAA,QACR,mBAAmB,QAAQ,QAAQ,UAAU,OAAO,OAAO;AAAA,QAC3D;AAAA,QACA,EAAE,OAAO,QAAQ;AAAA,MACnB;AAAA,IACF;AACA,QAAI,CAAC,MAAO,OAAM,IAAI,iBAAiB,0CAA0C,cAAc;AAC/F,WAAO,iBAAiB,EAAE,GAAG,OAAO,eAAe,IAAI;AAAA,EACzD;AACF;AAGA,eAAsB,eAAkB,MAA0B;AAChE,MAAI;AACJ,MAAI;AACF,UAAM,MAAM,SAAS,MAAM,MAAM;AAAA,EACnC,SAAS,KAAK;AACZ,UAAM,IAAI,MAAM,gCAAgC,IAAI,KAAK,eAAe,QAAQ,IAAI,UAAU,GAAG,GAAG;AAAA,EACtG;AACA,MAAI;AACF,WAAO,KAAK,MAAM,GAAG;AAAA,EACvB,SAAS,KAAK;AACZ,UAAM,IAAI,MAAM,4BAA4B,IAAI,KAAK,eAAe,QAAQ,IAAI,UAAU,GAAG,EAAE;AAAA,EACjG;AACF;AAGA,eAAsB,UAAU,MAAc,SAAgC;AAC5E,QAAM,UAAU,MAAM,OAAO;AAC/B;AAGO,SAAS,UAAU,QAAgB,IAAoB;AAC5D,SAAO,GAAG,MAAM,IAAI,EAAE,GAAG,QAAQ,oBAAoB,GAAG;AAC1D;","names":["resolve"]}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/benchmarks/humaneval.ts"],"sourcesContent":["/**\n * HumanEval adapter — the deployable-checker domain as a `BenchmarkAdapter`, so the\n * gate runner (`runGate`) can A/B the STEERING regime on it: a real rollout\n * through the `Supervisor` that self-corrects across rounds, vs blind\n * random@k. This is the experiment `humaneval-gate.mts` names as \"the next one\" —\n * the gate measures SELECTION over stateless single completions; this measures\n * whether observe→steer (self-correction) beats blind compute at equal k.\n *\n * Worker artifact = the model's reply (a Python function, fenced or raw). The\n * DETERMINISTIC judge runs the candidate against the task's own `test` in an\n * isolated `--network=none` python container — exit 0 = pass. No gold\n * `canonical_solution` is ever shown to the model; `goldArtifact` returns it only\n * to self-verify the judge before spending tokens.\n *\n * The primitives (loader / extractor / Docker checker) live here and are reused by\n * `humaneval-gate.mts` — one home, no duplication.\n */\n\nimport { execFile } from 'node:child_process'\nimport { mkdtempSync, readFileSync, rmSync, writeFileSync } from 'node:fs'\nimport { tmpdir } from 'node:os'\nimport { join } from 'node:path'\nimport { gunzipSync } from 'node:zlib'\nimport type { BenchmarkAdapter, BenchScore, BenchTask, LoadOptions } from './types'\n\nconst humanevalUrl = 'https://github.com/openai/human-eval/raw/master/data/HumanEval.jsonl.gz'\nconst dockerImage = 'python:3.12-slim'\nconst dockerTimeoutMs = Number(process.env.DOCKER_TIMEOUT_MS ?? 20000)\n\nexport interface HumanEvalTask {\n taskId: string\n prompt: string\n test: string\n entryPoint: string\n /** Reference solution body — used ONLY to self-verify the judge, never shown to the worker. */\n canonicalSolution?: string\n}\n\n/** Pull the 164-task HumanEval JSONL.gz and parse it. Fail loud on a non-OK fetch\n * or a malformed line — a silently-short task set would poison the gate. `offset`\n * selects a deeper slice (the later tasks are harder) so the worker has a\n * correctable middle band rather than a saturated easy prefix. */\nexport async function loadHumanEval(limit: number, offset = 0): Promise<HumanEvalTask[]> {\n // Prefer a locally-cached .jsonl.gz (HUMANEVAL_GZ) — the GitHub raw URL rate-limits\n // (429) under repeated runs. Fall back to the network fetch when unset.\n const localGz = process.env.HUMANEVAL_GZ\n let gz: Buffer\n if (localGz) {\n gz = readFileSync(localGz)\n } else {\n const res = await fetch(humanevalUrl)\n if (!res.ok) throw new Error(`HumanEval fetch HTTP ${res.status}: ${humanevalUrl}`)\n gz = Buffer.from(await res.arrayBuffer())\n }\n const text = gunzipSync(gz).toString('utf8')\n const tasks: HumanEvalTask[] = []\n for (const line of text.split('\\n')) {\n if (line.trim() === '') continue\n const d = JSON.parse(line) as {\n task_id?: string\n prompt?: string\n test?: string\n entry_point?: string\n canonical_solution?: string\n }\n if (!d.task_id || !d.prompt || !d.test || !d.entry_point) {\n throw new Error(`malformed HumanEval record: ${line.slice(0, 120)}`)\n }\n tasks.push({\n taskId: d.task_id,\n prompt: d.prompt,\n test: d.test,\n entryPoint: d.entry_point,\n ...(d.canonical_solution ? { canonicalSolution: d.canonical_solution } : {}),\n })\n }\n if (tasks.length === 0) throw new Error('HumanEval parsed to 0 tasks')\n if (offset >= tasks.length) throw new Error(`OFFSET ${offset} >= dataset size ${tasks.length}`)\n return tasks.slice(offset, offset + limit)\n}\n\nconst solveInstruction =\n 'Complete the following Python function. Output the COMPLETE function definition (signature, docstring optional, body) inside a single ```python code block. Include any imports the function needs. Do not write tests or example calls.'\n\nexport function basePrompt(task: HumanEvalTask): string {\n return `${solveInstruction}\\n\\n\\`\\`\\`python\\n${task.prompt}\\`\\`\\``\n}\n\n/** Extract the function source from a model reply: prefer a fenced ```python (or\n * bare ```) block, else fall back to the raw text. The deployable program adds the\n * prompt header (imports + signature context), so a candidate that returns only a\n * body still runs; a candidate that re-defines the function shadows the header. */\nexport function extractCode(reply: string): string {\n const fenced = reply.match(/```(?:python|py)?\\s*\\n([\\s\\S]*?)```/i)\n if (fenced && typeof fenced[1] === 'string') return fenced[1].trim()\n return reply.trim()\n}\n\n/** The deployable test program: the prompt header (imports + signature/docstring the\n * model was given), then the candidate (its def shadows the header's stub), then the\n * task's own check() suite and the call. No gold solution anywhere. */\nfunction buildProgram(task: HumanEvalTask, candidate: string): string {\n return `${task.prompt}\\n${candidate}\\n\\n${task.test}\\n\\ncheck(${task.entryPoint})\\n`\n}\n\nexport interface CheckResult {\n /** {0,1} pass-count for this candidate (1 = the check() suite passed). */\n pass: number\n /** On failure: the interpreter stderr tail (traceback / failing assertion). The\n * execution-grounded feedback a self-repair loop steers on; ignored by selection. */\n detail?: string\n}\n\n/** Run one candidate's deployable test program in an isolated container:\n * `docker run --rm --network=none -v <tmp>:/w -w /w <img> python /w/p.py`.\n * Exit 0 → pass. A docker invocation error (binary missing, daemon down, image\n * pull failure) is NOT a test failure — it throws so the harness fails loud rather\n * than scoring every candidate 0 from a broken checker. */\nlet dockerRunSeq = 0\n\nexport function runChecker(task: HumanEvalTask, candidate: string): Promise<CheckResult> {\n const dir = mkdtempSync(join(tmpdir(), 'hev-'))\n writeFileSync(join(dir, 'p.py'), buildProgram(task, candidate))\n // Unique container name so we can force-reap it regardless of the docker client's state.\n const name = `hev-${process.pid}-${dockerRunSeq++}`\n return new Promise<CheckResult>((resolvePromise, reject) => {\n let settled = false\n const cleanup = () => {\n rmSync(dir, { recursive: true, force: true })\n // `execFile`'s `timeout` kills the docker CLIENT, not the container — a hung\n // `python` would otherwise pin a CPU forever. Force-reap by name (fire-and-forget;\n // the name is unique, so no reuse race).\n execFile('docker', ['rm', '-f', name], () => {})\n }\n const finish = (res: CheckResult) => {\n if (settled) return\n settled = true\n clearTimeout(backstop)\n cleanup()\n resolvePromise(res)\n }\n const fail = (e: Error) => {\n if (settled) return\n settled = true\n clearTimeout(backstop)\n cleanup()\n reject(e)\n }\n // A hung container can leave the docker client stuck forwarding SIGTERM, so the\n // execFile callback never fires. This guarantees resolution (and reap) after the\n // timeout, independent of the callback.\n const backstop = setTimeout(() => finish({ pass: 0 }), dockerTimeoutMs + 3000)\n execFile(\n 'docker',\n [\n 'run',\n '--rm',\n '--name',\n name,\n '--network=none',\n '--cpus=1',\n '--memory=512m',\n '-v',\n `${dir}:/w:ro`,\n '-w',\n '/w',\n dockerImage,\n 'python',\n '/w/p.py',\n ],\n { timeout: dockerTimeoutMs, killSignal: 'SIGKILL', maxBuffer: 4 * 1024 * 1024 },\n (err, _stdout, stderr) => {\n if (err) {\n const e = err as NodeJS.ErrnoException & { killed?: boolean; code?: number | string }\n if (e.code === 'ENOENT') {\n fail(new Error('docker binary not found on PATH — cannot run the deployable checker'))\n return\n }\n if (/cannot connect to the docker daemon|is the docker daemon running|permission denied while trying to connect/i.test(stderr)) {\n fail(new Error(`docker daemon unreachable: ${stderr.slice(0, 200)}`))\n return\n }\n if (/(unable to find image|pull access denied|manifest unknown|error response from daemon).*(pull|repository|registry)/i.test(stderr)) {\n fail(new Error(`docker image ${dockerImage} unavailable: ${stderr.slice(0, 200)}`))\n return\n }\n // killed-by-timeout or a non-zero exit (assert failure / error) are genuine\n // test FAILURES — score 0, do not throw. Carry the stderr tail as the\n // execution-grounded failure detail (empty ⇒ timeout/SIGKILL left no output).\n finish({ pass: 0, detail: (stderr || '').slice(-600) || 'timed out (no output)' })\n return\n }\n finish({ pass: 1 })\n },\n )\n })\n}\n\n/** A HumanEval task carries its checker inputs in metadata so the deterministic\n * judge can rebuild the deployable program from a `BenchTask` alone. */\ninterface HumanEvalMeta extends Record<string, unknown> {\n promptHeader: string\n test: string\n entryPoint: string\n canonicalSolution?: string\n}\n\nfunction toBenchTask(t: HumanEvalTask): BenchTask {\n const metadata: HumanEvalMeta = {\n promptHeader: t.prompt,\n test: t.test,\n entryPoint: t.entryPoint,\n ...(t.canonicalSolution ? { canonicalSolution: t.canonicalSolution } : {}),\n }\n return { id: t.taskId, prompt: basePrompt(t), metadata }\n}\n\nfunction taskFromMeta(task: BenchTask): HumanEvalTask {\n const m = task.metadata as HumanEvalMeta | undefined\n if (!m?.promptHeader || !m.test || !m.entryPoint) {\n throw new Error(`HumanEval judge: task ${task.id} missing checker metadata`)\n }\n return { taskId: task.id, prompt: m.promptHeader, test: m.test, entryPoint: m.entryPoint }\n}\n\n/** The HumanEval `BenchmarkAdapter`. OFFSET (env) selects the correctable middle\n * band; loadTasks honors `limit`/`ids`. The judge is the Docker deployable checker. */\nexport function createHumanEvalAdapter(): BenchmarkAdapter {\n return {\n name: 'humaneval',\n async preflight() {\n // The judge is the only hard dependency; it fails loud on a missing/broken\n // docker, so a cheap presence check here gives an earlier, clearer signal.\n await new Promise<void>((resolve, reject) => {\n execFile('docker', ['version', '--format', '{{.Server.Version}}'], (err) => {\n if (err) reject(new Error('HumanEval judge needs a running Docker daemon (python:3.12-slim, --network=none)'))\n else resolve()\n })\n })\n },\n async loadTasks(opts?: LoadOptions) {\n const offset = Number(process.env.OFFSET ?? 0)\n // Pull a generous window when filtering by id, else exactly `limit` from offset.\n const all = await loadHumanEval(opts?.ids ? 164 : (opts?.limit ?? 8), offset)\n const picked = opts?.ids ? all.filter((t) => opts.ids?.includes(t.taskId)) : all\n return picked.map(toBenchTask)\n },\n async judge(task: BenchTask, artifact: string): Promise<BenchScore> {\n const { pass } = await runChecker(taskFromMeta(task), extractCode(artifact))\n return { resolved: pass === 1, score: pass, detail: pass === 1 ? 'tests passed' : 'tests failed' }\n },\n async goldArtifact(task: BenchTask) {\n const m = task.metadata as HumanEvalMeta | undefined\n const sol = m?.canonicalSolution\n // Return the COMPLETE function (signature header + canonical body), i.e. what a\n // real worker emits — NOT the body alone. The judge runs `extractCode`, whose\n // unfenced fallback is `reply.trim()`; trimming a body-only string strips its\n // leading indent and breaks it, so a body-only gold fails its own judge. A full\n // def starts at column 0, trims safely, and self-verifies.\n return sol ? `${m!.promptHeader}${sol}` : undefined\n },\n }\n}\n"],"mappings":";AAkBA,SAAS,gBAAgB;AACzB,SAAS,aAAa,cAAc,QAAQ,qBAAqB;AACjE,SAAS,cAAc;AACvB,SAAS,YAAY;AACrB,SAAS,kBAAkB;AAG3B,IAAM,eAAe;AACrB,IAAM,cAAc;AACpB,IAAM,kBAAkB,OAAO,QAAQ,IAAI,qBAAqB,GAAK;AAerE,eAAsB,cAAc,OAAe,SAAS,GAA6B;AAGvF,QAAM,UAAU,QAAQ,IAAI;AAC5B,MAAI;AACJ,MAAI,SAAS;AACX,SAAK,aAAa,OAAO;AAAA,EAC3B,OAAO;AACL,UAAM,MAAM,MAAM,MAAM,YAAY;AACpC,QAAI,CAAC,IAAI,GAAI,OAAM,IAAI,MAAM,wBAAwB,IAAI,MAAM,KAAK,YAAY,EAAE;AAClF,SAAK,OAAO,KAAK,MAAM,IAAI,YAAY,CAAC;AAAA,EAC1C;AACA,QAAM,OAAO,WAAW,EAAE,EAAE,SAAS,MAAM;AAC3C,QAAM,QAAyB,CAAC;AAChC,aAAW,QAAQ,KAAK,MAAM,IAAI,GAAG;AACnC,QAAI,KAAK,KAAK,MAAM,GAAI;AACxB,UAAM,IAAI,KAAK,MAAM,IAAI;AAOzB,QAAI,CAAC,EAAE,WAAW,CAAC,EAAE,UAAU,CAAC,EAAE,QAAQ,CAAC,EAAE,aAAa;AACxD,YAAM,IAAI,MAAM,+BAA+B,KAAK,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,IACrE;AACA,UAAM,KAAK;AAAA,MACT,QAAQ,EAAE;AAAA,MACV,QAAQ,EAAE;AAAA,MACV,MAAM,EAAE;AAAA,MACR,YAAY,EAAE;AAAA,MACd,GAAI,EAAE,qBAAqB,EAAE,mBAAmB,EAAE,mBAAmB,IAAI,CAAC;AAAA,IAC5E,CAAC;AAAA,EACH;AACA,MAAI,MAAM,WAAW,EAAG,OAAM,IAAI,MAAM,6BAA6B;AACrE,MAAI,UAAU,MAAM,OAAQ,OAAM,IAAI,MAAM,UAAU,MAAM,oBAAoB,MAAM,MAAM,EAAE;AAC9F,SAAO,MAAM,MAAM,QAAQ,SAAS,KAAK;AAC3C;AAEA,IAAM,mBACJ;AAEK,SAAS,WAAW,MAA6B;AACtD,SAAO,GAAG,gBAAgB;AAAA;AAAA;AAAA,EAAqB,KAAK,MAAM;AAC5D;AAMO,SAAS,YAAY,OAAuB;AACjD,QAAM,SAAS,MAAM,MAAM,sCAAsC;AACjE,MAAI,UAAU,OAAO,OAAO,CAAC,MAAM,SAAU,QAAO,OAAO,CAAC,EAAE,KAAK;AACnE,SAAO,MAAM,KAAK;AACpB;AAKA,SAAS,aAAa,MAAqB,WAA2B;AACpE,SAAO,GAAG,KAAK,MAAM;AAAA,EAAK,SAAS;AAAA;AAAA,EAAO,KAAK,IAAI;AAAA;AAAA,QAAa,KAAK,UAAU;AAAA;AACjF;AAeA,IAAI,eAAe;AAEZ,SAAS,WAAW,MAAqB,WAAyC;AACvF,QAAM,MAAM,YAAY,KAAK,OAAO,GAAG,MAAM,CAAC;AAC9C,gBAAc,KAAK,KAAK,MAAM,GAAG,aAAa,MAAM,SAAS,CAAC;AAE9D,QAAM,OAAO,OAAO,QAAQ,GAAG,IAAI,cAAc;AACjD,SAAO,IAAI,QAAqB,CAAC,gBAAgB,WAAW;AAC1D,QAAI,UAAU;AACd,UAAM,UAAU,MAAM;AACpB,aAAO,KAAK,EAAE,WAAW,MAAM,OAAO,KAAK,CAAC;AAI5C,eAAS,UAAU,CAAC,MAAM,MAAM,IAAI,GAAG,MAAM;AAAA,MAAC,CAAC;AAAA,IACjD;AACA,UAAM,SAAS,CAAC,QAAqB;AACnC,UAAI,QAAS;AACb,gBAAU;AACV,mBAAa,QAAQ;AACrB,cAAQ;AACR,qBAAe,GAAG;AAAA,IACpB;AACA,UAAM,OAAO,CAAC,MAAa;AACzB,UAAI,QAAS;AACb,gBAAU;AACV,mBAAa,QAAQ;AACrB,cAAQ;AACR,aAAO,CAAC;AAAA,IACV;AAIA,UAAM,WAAW,WAAW,MAAM,OAAO,EAAE,MAAM,EAAE,CAAC,GAAG,kBAAkB,GAAI;AAC7E;AAAA,MACE;AAAA,MACA;AAAA,QACE;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA,GAAG,GAAG;AAAA,QACN;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,MACF;AAAA,MACA,EAAE,SAAS,iBAAiB,YAAY,WAAW,WAAW,IAAI,OAAO,KAAK;AAAA,MAC9E,CAAC,KAAK,SAAS,WAAW;AACxB,YAAI,KAAK;AACP,gBAAM,IAAI;AACV,cAAI,EAAE,SAAS,UAAU;AACvB,iBAAK,IAAI,MAAM,0EAAqE,CAAC;AACrF;AAAA,UACF;AACA,cAAI,8GAA8G,KAAK,MAAM,GAAG;AAC9H,iBAAK,IAAI,MAAM,8BAA8B,OAAO,MAAM,GAAG,GAAG,CAAC,EAAE,CAAC;AACpE;AAAA,UACF;AACA,cAAI,qHAAqH,KAAK,MAAM,GAAG;AACrI,iBAAK,IAAI,MAAM,gBAAgB,WAAW,iBAAiB,OAAO,MAAM,GAAG,GAAG,CAAC,EAAE,CAAC;AAClF;AAAA,UACF;AAIA,iBAAO,EAAE,MAAM,GAAG,SAAS,UAAU,IAAI,MAAM,IAAI,KAAK,wBAAwB,CAAC;AACjF;AAAA,QACF;AACA,eAAO,EAAE,MAAM,EAAE,CAAC;AAAA,MACpB;AAAA,IACF;AAAA,EACF,CAAC;AACH;AAWA,SAAS,YAAY,GAA6B;AAChD,QAAM,WAA0B;AAAA,IAC9B,cAAc,EAAE;AAAA,IAChB,MAAM,EAAE;AAAA,IACR,YAAY,EAAE;AAAA,IACd,GAAI,EAAE,oBAAoB,EAAE,mBAAmB,EAAE,kBAAkB,IAAI,CAAC;AAAA,EAC1E;AACA,SAAO,EAAE,IAAI,EAAE,QAAQ,QAAQ,WAAW,CAAC,GAAG,SAAS;AACzD;AAEA,SAAS,aAAa,MAAgC;AACpD,QAAM,IAAI,KAAK;AACf,MAAI,CAAC,GAAG,gBAAgB,CAAC,EAAE,QAAQ,CAAC,EAAE,YAAY;AAChD,UAAM,IAAI,MAAM,yBAAyB,KAAK,EAAE,2BAA2B;AAAA,EAC7E;AACA,SAAO,EAAE,QAAQ,KAAK,IAAI,QAAQ,EAAE,cAAc,MAAM,EAAE,MAAM,YAAY,EAAE,WAAW;AAC3F;AAIO,SAAS,yBAA2C;AACzD,SAAO;AAAA,IACL,MAAM;AAAA,IACN,MAAM,YAAY;AAGhB,YAAM,IAAI,QAAc,CAAC,SAAS,WAAW;AAC3C,iBAAS,UAAU,CAAC,WAAW,YAAY,qBAAqB,GAAG,CAAC,QAAQ;AAC1E,cAAI,IAAK,QAAO,IAAI,MAAM,kFAAkF,CAAC;AAAA,cACxG,SAAQ;AAAA,QACf,CAAC;AAAA,MACH,CAAC;AAAA,IACH;AAAA,IACA,MAAM,UAAU,MAAoB;AAClC,YAAM,SAAS,OAAO,QAAQ,IAAI,UAAU,CAAC;AAE7C,YAAM,MAAM,MAAM,cAAc,MAAM,MAAM,MAAO,MAAM,SAAS,GAAI,MAAM;AAC5E,YAAM,SAAS,MAAM,MAAM,IAAI,OAAO,CAAC,MAAM,KAAK,KAAK,SAAS,EAAE,MAAM,CAAC,IAAI;AAC7E,aAAO,OAAO,IAAI,WAAW;AAAA,IAC/B;AAAA,IACA,MAAM,MAAM,MAAiB,UAAuC;AAClE,YAAM,EAAE,KAAK,IAAI,MAAM,WAAW,aAAa,IAAI,GAAG,YAAY,QAAQ,CAAC;AAC3E,aAAO,EAAE,UAAU,SAAS,GAAG,OAAO,MAAM,QAAQ,SAAS,IAAI,iBAAiB,eAAe;AAAA,IACnG;AAAA,IACA,MAAM,aAAa,MAAiB;AAClC,YAAM,IAAI,KAAK;AACf,YAAM,MAAM,GAAG;AAMf,aAAO,MAAM,GAAG,EAAG,YAAY,GAAG,GAAG,KAAK;AAAA,IAC5C;AAAA,EACF;AACF;","names":[]}
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|