@claude-flow/cli 3.10.4 → 3.10.6
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/dist/src/benchmarks/capability-tasks.d.ts +126 -0
- package/dist/src/benchmarks/capability-tasks.d.ts.map +1 -0
- package/dist/src/benchmarks/capability-tasks.js +147 -0
- package/dist/src/benchmarks/capability-tasks.js.map +1 -0
- package/dist/src/benchmarks/gaia-agent-contrastive.d.ts +53 -0
- package/dist/src/benchmarks/gaia-agent-contrastive.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-agent-contrastive.js +177 -0
- package/dist/src/benchmarks/gaia-agent-contrastive.js.map +1 -0
- package/dist/src/benchmarks/gaia-agent-gemini.d.ts +51 -0
- package/dist/src/benchmarks/gaia-agent-gemini.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-agent-gemini.js +352 -0
- package/dist/src/benchmarks/gaia-agent-gemini.js.map +1 -0
- package/dist/src/benchmarks/gaia-agent-planning.smoke.d.ts +18 -0
- package/dist/src/benchmarks/gaia-agent-planning.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-agent-planning.smoke.js +253 -0
- package/dist/src/benchmarks/gaia-agent-planning.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-agent.d.ts +198 -0
- package/dist/src/benchmarks/gaia-agent.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-agent.js +651 -0
- package/dist/src/benchmarks/gaia-agent.js.map +1 -0
- package/dist/src/benchmarks/gaia-causal-memory.d.ts +133 -0
- package/dist/src/benchmarks/gaia-causal-memory.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-causal-memory.js +281 -0
- package/dist/src/benchmarks/gaia-causal-memory.js.map +1 -0
- package/dist/src/benchmarks/gaia-causal-memory.smoke.d.ts +22 -0
- package/dist/src/benchmarks/gaia-causal-memory.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-causal-memory.smoke.js +300 -0
- package/dist/src/benchmarks/gaia-causal-memory.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-claude-p.d.ts +105 -0
- package/dist/src/benchmarks/gaia-claude-p.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-claude-p.js +269 -0
- package/dist/src/benchmarks/gaia-claude-p.js.map +1 -0
- package/dist/src/benchmarks/gaia-claude-p.smoke.d.ts +20 -0
- package/dist/src/benchmarks/gaia-claude-p.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-claude-p.smoke.js +168 -0
- package/dist/src/benchmarks/gaia-claude-p.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-codeagent-runner.py +556 -0
- package/dist/src/benchmarks/gaia-codeagent.d.ts +129 -0
- package/dist/src/benchmarks/gaia-codeagent.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-codeagent.js +590 -0
- package/dist/src/benchmarks/gaia-codeagent.js.map +1 -0
- package/dist/src/benchmarks/gaia-codeagent.smoke.d.ts +33 -0
- package/dist/src/benchmarks/gaia-codeagent.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-codeagent.smoke.js +241 -0
- package/dist/src/benchmarks/gaia-codeagent.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-convergence.d.ts +138 -0
- package/dist/src/benchmarks/gaia-convergence.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-convergence.js +260 -0
- package/dist/src/benchmarks/gaia-convergence.js.map +1 -0
- package/dist/src/benchmarks/gaia-convergence.smoke.d.ts +19 -0
- package/dist/src/benchmarks/gaia-convergence.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-convergence.smoke.js +246 -0
- package/dist/src/benchmarks/gaia-convergence.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-critic.d.ts +123 -0
- package/dist/src/benchmarks/gaia-critic.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-critic.js +312 -0
- package/dist/src/benchmarks/gaia-critic.js.map +1 -0
- package/dist/src/benchmarks/gaia-critic.smoke.d.ts +21 -0
- package/dist/src/benchmarks/gaia-critic.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-critic.smoke.js +327 -0
- package/dist/src/benchmarks/gaia-critic.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-dag-pilot.d.ts +22 -0
- package/dist/src/benchmarks/gaia-dag-pilot.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-dag-pilot.js +158 -0
- package/dist/src/benchmarks/gaia-dag-pilot.js.map +1 -0
- package/dist/src/benchmarks/gaia-dag.d.ts +110 -0
- package/dist/src/benchmarks/gaia-dag.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-dag.js +658 -0
- package/dist/src/benchmarks/gaia-dag.js.map +1 -0
- package/dist/src/benchmarks/gaia-dag.smoke.d.ts +21 -0
- package/dist/src/benchmarks/gaia-dag.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-dag.smoke.js +290 -0
- package/dist/src/benchmarks/gaia-dag.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-decomposer.d.ts +125 -0
- package/dist/src/benchmarks/gaia-decomposer.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-decomposer.js +350 -0
- package/dist/src/benchmarks/gaia-decomposer.js.map +1 -0
- package/dist/src/benchmarks/gaia-decomposer.smoke.d.ts +21 -0
- package/dist/src/benchmarks/gaia-decomposer.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-decomposer.smoke.js +228 -0
- package/dist/src/benchmarks/gaia-decomposer.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-e2e-smoke.d.ts +27 -0
- package/dist/src/benchmarks/gaia-e2e-smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-e2e-smoke.js +136 -0
- package/dist/src/benchmarks/gaia-e2e-smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-ensemble-pilot.d.ts +17 -0
- package/dist/src/benchmarks/gaia-ensemble-pilot.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-ensemble-pilot.js +106 -0
- package/dist/src/benchmarks/gaia-ensemble-pilot.js.map +1 -0
- package/dist/src/benchmarks/gaia-ensemble.d.ts +103 -0
- package/dist/src/benchmarks/gaia-ensemble.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-ensemble.js +462 -0
- package/dist/src/benchmarks/gaia-ensemble.js.map +1 -0
- package/dist/src/benchmarks/gaia-extract.smoke.d.ts +45 -0
- package/dist/src/benchmarks/gaia-extract.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-extract.smoke.js +242 -0
- package/dist/src/benchmarks/gaia-extract.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-gemini-pilot.d.ts +17 -0
- package/dist/src/benchmarks/gaia-gemini-pilot.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-gemini-pilot.js +168 -0
- package/dist/src/benchmarks/gaia-gemini-pilot.js.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/features.d.ts +46 -0
- package/dist/src/benchmarks/gaia-hardness/features.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/features.js +170 -0
- package/dist/src/benchmarks/gaia-hardness/features.js.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.d.ts +105 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.js +260 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.js.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.smoke.d.ts +20 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.smoke.js +235 -0
- package/dist/src/benchmarks/gaia-hardness/predictor.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/train-data-loader.d.ts +51 -0
- package/dist/src/benchmarks/gaia-hardness/train-data-loader.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-hardness/train-data-loader.js +179 -0
- package/dist/src/benchmarks/gaia-hardness/train-data-loader.js.map +1 -0
- package/dist/src/benchmarks/gaia-judge.d.ts +88 -0
- package/dist/src/benchmarks/gaia-judge.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-judge.js +437 -0
- package/dist/src/benchmarks/gaia-judge.js.map +1 -0
- package/dist/src/benchmarks/gaia-loader.d.ts +87 -0
- package/dist/src/benchmarks/gaia-loader.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-loader.js +326 -0
- package/dist/src/benchmarks/gaia-loader.js.map +1 -0
- package/dist/src/benchmarks/gaia-mode-router.d.ts +67 -0
- package/dist/src/benchmarks/gaia-mode-router.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-mode-router.js +126 -0
- package/dist/src/benchmarks/gaia-mode-router.js.map +1 -0
- package/dist/src/benchmarks/gaia-mode-router.smoke.d.ts +15 -0
- package/dist/src/benchmarks/gaia-mode-router.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-mode-router.smoke.js +131 -0
- package/dist/src/benchmarks/gaia-mode-router.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/file_read.d.ts +35 -0
- package/dist/src/benchmarks/gaia-tools/file_read.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/file_read.js +403 -0
- package/dist/src/benchmarks/gaia-tools/file_read.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.d.ts +126 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.js +225 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.d.ts +24 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.js +276 -0
- package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.d.ts +45 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.js +222 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.smoke.d.ts +24 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.smoke.js +169 -0
- package/dist/src/benchmarks/gaia-tools/image_describe.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/index.d.ts +32 -0
- package/dist/src/benchmarks/gaia-tools/index.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/index.js +36 -0
- package/dist/src/benchmarks/gaia-tools/index.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/pdf_read.d.ts +27 -0
- package/dist/src/benchmarks/gaia-tools/pdf_read.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/pdf_read.js +226 -0
- package/dist/src/benchmarks/gaia-tools/pdf_read.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.d.ts +31 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.js +94 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.smoke.d.ts +13 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.smoke.js +98 -0
- package/dist/src/benchmarks/gaia-tools/python_exec.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/types.d.ts +62 -0
- package/dist/src/benchmarks/gaia-tools/types.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/types.js +12 -0
- package/dist/src/benchmarks/gaia-tools/types.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/visit_webpage.d.ts +52 -0
- package/dist/src/benchmarks/gaia-tools/visit_webpage.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/visit_webpage.js +304 -0
- package/dist/src/benchmarks/gaia-tools/visit_webpage.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.d.ts +61 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.js +217 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.smoke.d.ts +26 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.smoke.js +152 -0
- package/dist/src/benchmarks/gaia-tools/web_browse.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_search.d.ts +30 -0
- package/dist/src/benchmarks/gaia-tools/web_search.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_search.js +210 -0
- package/dist/src/benchmarks/gaia-tools/web_search.js.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_search.smoke.d.ts +24 -0
- package/dist/src/benchmarks/gaia-tools/web_search.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-tools/web_search.smoke.js +311 -0
- package/dist/src/benchmarks/gaia-tools/web_search.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-visit-webpage.smoke.d.ts +25 -0
- package/dist/src/benchmarks/gaia-visit-webpage.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-visit-webpage.smoke.js +304 -0
- package/dist/src/benchmarks/gaia-visit-webpage.smoke.js.map +1 -0
- package/dist/src/benchmarks/gaia-voting.d.ts +88 -0
- package/dist/src/benchmarks/gaia-voting.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-voting.js +297 -0
- package/dist/src/benchmarks/gaia-voting.js.map +1 -0
- package/dist/src/benchmarks/gaia-voting.smoke.d.ts +20 -0
- package/dist/src/benchmarks/gaia-voting.smoke.d.ts.map +1 -0
- package/dist/src/benchmarks/gaia-voting.smoke.js +332 -0
- package/dist/src/benchmarks/gaia-voting.smoke.js.map +1 -0
- package/dist/src/commands/gaia-bench.d.ts +40 -0
- package/dist/src/commands/gaia-bench.d.ts.map +1 -0
- package/dist/src/commands/gaia-bench.js +597 -0
- package/dist/src/commands/gaia-bench.js.map +1 -0
- package/dist/src/commands/index.d.ts.map +1 -1
- package/dist/src/commands/index.js +2 -0
- package/dist/src/commands/index.js.map +1 -1
- package/dist/src/commands/init.d.ts.map +1 -1
- package/dist/src/commands/init.js +41 -4
- package/dist/src/commands/init.js.map +1 -1
- package/dist/src/commands/performance-capability.d.ts +30 -0
- package/dist/src/commands/performance-capability.d.ts.map +1 -0
- package/dist/src/commands/performance-capability.js +349 -0
- package/dist/src/commands/performance-capability.js.map +1 -0
- package/dist/src/commands/route.d.ts.map +1 -1
- package/dist/src/commands/route.js +9 -0
- package/dist/src/commands/route.js.map +1 -1
- package/dist/src/init/executor.d.ts.map +1 -1
- package/dist/src/init/executor.js +25 -6
- package/dist/src/init/executor.js.map +1 -1
- package/dist/src/init/mcp-generator.d.ts.map +1 -1
- package/dist/src/init/mcp-generator.js +9 -4
- package/dist/src/init/mcp-generator.js.map +1 -1
- package/dist/src/init/statusline-generator.d.ts.map +1 -1
- package/dist/src/init/statusline-generator.js +25 -2
- package/dist/src/init/statusline-generator.js.map +1 -1
- package/dist/src/mcp-tools/agentdb-tools.d.ts.map +1 -1
- package/dist/src/mcp-tools/agentdb-tools.js +20 -3
- package/dist/src/mcp-tools/agentdb-tools.js.map +1 -1
- package/dist/src/mcp-tools/system-tools.d.ts.map +1 -1
- package/dist/src/mcp-tools/system-tools.js +12 -1
- package/dist/src/mcp-tools/system-tools.js.map +1 -1
- package/dist/src/memory/memory-bridge.d.ts.map +1 -1
- package/dist/src/memory/memory-bridge.js +34 -0
- package/dist/src/memory/memory-bridge.js.map +1 -1
- package/dist/tsconfig.tsbuildinfo +1 -1
- package/package.json +1 -1
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"gaia-critic.smoke.js","sourceRoot":"","sources":["../../../src/benchmarks/gaia-critic.smoke.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,EACL,YAAY,EACZ,sBAAsB,GAGvB,MAAM,kBAAkB,CAAC;AAI1B,8EAA8E;AAC9E,eAAe;AACf,8EAA8E;AAE9E,IAAI,MAAM,GAAG,CAAC,CAAC;AACf,IAAI,MAAM,GAAG,CAAC,CAAC;AAEf,SAAS,MAAM,CAAC,SAAkB,EAAE,OAAe;IACjD,IAAI,SAAS,EAAE,CAAC;QACd,OAAO,CAAC,GAAG,CAAC,WAAW,OAAO,EAAE,CAAC,CAAC;QAClC,MAAM,EAAE,CAAC;IACX,CAAC;SAAM,CAAC;QACN,OAAO,CAAC,KAAK,CAAC,WAAW,OAAO,EAAE,CAAC,CAAC;QACpC,MAAM,EAAE,CAAC;IACX,CAAC;AACH,CAAC;AAED,SAAS,WAAW,CAAI,MAAS,EAAE,QAAW,EAAE,OAAe;IAC7D,MAAM,CAAC,MAAM,KAAK,QAAQ,EAAE,GAAG,OAAO,cAAc,MAAM,CAAC,QAAQ,CAAC,SAAS,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;AAClG,CAAC;AAED,sCAAsC;AACtC,MAAM,gBAAgB,GAAiB;IACrC,OAAO,EAAE,WAAW;IACpB,KAAK,EAAE,CAAC;IACR,QAAQ,EAAE,gCAAgC;IAC1C,YAAY,EAAE,OAAO;IACrB,SAAS,EAAE,IAAI;IACf,SAAS,EAAE,IAAI;CAChB,CAAC;AAEF,wDAAwD;AACxD,SAAS,eAAe,CAAC,WAA0B,EAAE,KAAK,GAAG,CAAC;IAC5D,OAAO;QACL,UAAU,EAAE,gBAAgB,CAAC,OAAO;QACpC,WAAW;QACX,KAAK;QACL,eAAe,EAAE,EAAE,UAAU,EAAE,CAAC,EAAE;QAClC,gBAAgB,EAAE,GAAG;QACrB,iBAAiB,EAAE,EAAE;QACrB,MAAM,EAAE,IAAI;KACb,CAAC;AACJ,CAAC;AAiBD,SAAS,gBAAgB,CAAC,SAAyB;IACjD,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,MAAM,QAAQ,GAAG,UAAU,CAAC,KAAK,CAAC;IAElC,8DAA8D;IAC7D,UAAkB,CAAC,KAAK,GAAG,KAAK,EAAE,IAAa,EAAE,KAAmB,EAAqB,EAAE;QAC1F,MAAM,IAAI,GAAG,SAAS,CAAC,IAAI,CAAC,GAAG,CAAC,SAAS,EAAE,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC;QAClE,SAAS,EAAE,CAAC;QAEZ,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE,CAAC;YACb,OAAO;gBACL,EAAE,EAAE,KAAK;gBACT,MAAM,EAAE,IAAI,CAAC,MAAM,IAAI,GAAG;gBAC1B,IAAI,EAAE,KAAK,IAAI,EAAE,CAAC,IAAI,CAAC,IAAI,IAAI,gBAAgB;gBAC/C,IAAI,EAAE,KAAK,IAAI,EAAE,GAAG,MAAM,IAAI,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC;aAC1B,CAAC;QAC3B,CAAC;QAED,OAAO;YACL,EAAE,EAAE,IAAI;YACR,MAAM,EAAE,GAAG;YACX,IAAI,EAAE,KAAK,IAAI,EAAE,CAAC,IAAI,CAAC,IAAI,IAAI,EAAE;YACjC,IAAI,EAAE,KAAK,IAAI,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,IAAI,EAAE,CAAC;SAC3B,CAAC;IAC3B,CAAC,CAAC;IAEF,OAAO,GAAG,EAAE,GAAG,UAAU,CAAC,KAAK,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;AAChD,CAAC;AAED;;GAEG;AACH,SAAS,kBAAkB,CACzB,OAA0B,EAC1B,SAAiB,EACjB,iBAAiB,GAAG,EAAE;IAEtB,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,EAAE,OAAO,EAAE,SAAS,EAAE,iBAAiB,EAAE,CAAC,CAAC;IAC1E,OAAO;QACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC;QAC1C,KAAK,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,aAAa,EAAE,EAAE,EAAE;KAChD,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;GAYG;AAEH,8EAA8E;AAC9E,2CAA2C;AAC3C,8EAA8E;AAE9E,KAAK,UAAU,gBAAgB;IAC7B,OAAO,CAAC,GAAG,CAAC,4CAA4C,CAAC,CAAC;IAE1D,MAAM,OAAO,GAAG,gBAAgB,CAAC;QAC/B,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,kBAAkB,CAAC,MAAM,EAAE,oBAAoB,CAAC,EAAE;KACrE,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,MAAM,OAAO,GAAkB,MAAM,YAAY,CAC/C,gBAAgB,EAChB,OAAO,EACP,EAAE,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,MAAM,EAAE,sBAAsB,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,EAC7E,EAAE,KAAK,EAAE,mBAAmB,EAAE,MAAM,EAAE,UAAU,EAAE,CACnD,CAAC;QAEF,WAAW,CAAC,OAAO,CAAC,OAAO,EAAE,MAAM,EAAE,mBAAmB,CAAC,CAAC;QAC1D,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,wBAAwB,CAAC,CAAC;QAC/D,MAAM,CAAC,OAAO,CAAC,OAAO,IAAI,CAAC,EAAE,yBAAyB,CAAC,CAAC;QACxD,MAAM,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,eAAe,CAAC,CAAC;IAC1C,CAAC;YAAS,CAAC;QACT,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED,8EAA8E;AAC9E,uDAAuD;AACvD,8EAA8E;AAE9E,KAAK,UAAU,gBAAgB;IAC7B,OAAO,CAAC,GAAG,CAAC,wDAAwD,CAAC,CAAC;IAEtE,MAAM,OAAO,GAAG,gBAAgB,CAAC;QAC/B;YACE,EAAE,EAAE,IAAI;YACR,IAAI,EAAE,kBAAkB,CACtB,MAAM,EACN,+BAA+B,EAC/B,OAAO,CACR;SACF;KACF,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,MAAM,OAAO,GAAkB,MAAM,YAAY,CAC/C,gBAAgB,EAChB,MAAM,EACN,EAAE,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,MAAM,EAAE,mBAAmB,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,EAC1E,EAAE,KAAK,EAAE,mBAAmB,EAAE,MAAM,EAAE,UAAU,EAAE,CACnD,CAAC;QAEF,WAAW,CAAC,OAAO,CAAC,OAAO,EAAE,MAAM,EAAE,mBAAmB,CAAC,CAAC;QAC1D,MAAM,CAAC,CAAC,OAAO,CAAC,iBAAiB,IAAI,EAAE,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,gCAAgC,CAAC,CAAC;QACvF,WAAW,CAAC,OAAO,CAAC,iBAAiB,EAAE,OAAO,EAAE,8BAA8B,CAAC,CAAC;IAClF,CAAC;YAAS,CAAC;QACT,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED,8EAA8E;AAC9E,0DAA0D;AAC1D,8EAA8E;AAE9E,KAAK,UAAU,sBAAsB;IACnC,OAAO,CAAC,GAAG,CAAC,0EAA0E,CAAC,CAAC;IAExF,oDAAoD;IACpD,uEAAuE;IACvE,0EAA0E;IAC1E,EAAE;IACF,wDAAwD;IACxD,mFAAmF;IACnF,mEAAmE;IACnE,wFAAwF;IACxF,mEAAmE;IACnE,EAAE;IACF,+EAA+E;IAC/E,4EAA4E;IAE5E,MAAM,iBAAiB,GAAG;QACxB,EAAE,EAAE,QAAQ;QACZ,IAAI,EAAE,SAAS;QACf,IAAI,EAAE,WAAW;QACjB,WAAW,EAAE,UAAU;QACvB,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,oBAAoB,EAAE,CAAC;QACvD,KAAK,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,aAAa,EAAE,EAAE,EAAE;QAC/C,KAAK,EAAE,kBAAkB;KAC1B,CAAC;IACF,MAAM,sBAAsB,GAAG;QAC7B,EAAE,EAAE,QAAQ;QACZ,IAAI,EAAE,SAAS;QACf,IAAI,EAAE,WAAW;QACjB,WAAW,EAAE,UAAU;QACvB,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,yBAAyB,EAAE,CAAC;QAC5D,KAAK,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,aAAa,EAAE,EAAE,EAAE;QAC/C,KAAK,EAAE,kBAAkB;KAC1B,CAAC;IAEF,MAAM,OAAO,GAAG,gBAAgB,CAAC;QAC/B,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,iBAAiB,EAAE,EAA2B,eAAe;QAC/E,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,kBAAkB,CAAC,MAAM,EAAE,aAAa,EAAE,OAAO,CAAC,EAAE,EAAG,WAAW;QACpF,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,sBAAsB,EAAE,EAAsB,uBAAuB;QACvF,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,kBAAkB,CAAC,MAAM,EAAE,cAAc,EAAE,OAAO,CAAC,EAAE,EAAE,WAAW;KACrF,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,MAAM,sBAAsB,CAAC,gBAAgB,EAAE;YAC5D,YAAY,EAAE,IAAI;YAClB,MAAM,EAAE,UAAU;YAClB,aAAa,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,UAAU,EAAE,CAAC,EAAE;SACrD,CAAC,CAAC;QAEH,WAAW,CAAC,MAAM,CAAC,gBAAgB,EAAE,CAAC,EAAE,uBAAuB,CAAC,CAAC;QACjE,WAAW,CAAC,MAAM,CAAC,cAAc,CAAC,MAAM,EAAE,CAAC,EAAE,+BAA+B,CAAC,CAAC;QAC9E,WAAW,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,uBAAuB,CAAC,CAAC;QAC/E,WAAW,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,wBAAwB,CAAC,CAAC;QAChF,yCAAyC;QACzC,MAAM,CAAC,MAAM,CAAC,WAAW,KAAK,IAAI,EAAE,mDAAmD,CAAC,CAAC;IAC3F,CAAC;YAAS,CAAC;QACT,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED,8EAA8E;AAC9E,mEAAmE;AACnE,8EAA8E;AAE9E,KAAK,UAAU,qBAAqB;IAClC,OAAO,CAAC,GAAG,CAAC,kEAAkE,CAAC,CAAC;IAEhF,MAAM,OAAO,GAAG,gBAAgB,CAAC;QAC/B;YACE,EAAE,EAAE,IAAI;YACR,IAAI,EAAE,kBAAkB,CAAC,WAAW,EAAE,qCAAqC,CAAC;SAC7E;KACF,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,MAAM,OAAO,GAAkB,MAAM,YAAY,CAC/C,gBAAgB,EAChB,OAAO,EACP,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,EACvB,EAAE,KAAK,EAAE,mBAAmB,EAAE,MAAM,EAAE,UAAU,EAAE,CACnD,CAAC;QAEF,WAAW,CAAC,OAAO,CAAC,OAAO,EAAE,WAAW,EAAE,wBAAwB,CAAC,CAAC;QAEpE,iEAAiE;QACjE,4EAA4E;QAC5E,yEAAyE;QACzE,0CAA0C;QAC1C,MAAM,kBAAkB,GAAG;YACzB,EAAE,EAAE,QAAQ;YACZ,IAAI,EAAE,SAAS;YACf,IAAI,EAAE,WAAW;YACjB,WAAW,EAAE,UAAU;YACvB,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,qBAAqB,EAAE,CAAC;YACxD,KAAK,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,aAAa,EAAE,EAAE,EAAE;YAC/C,KAAK,EAAE,kBAAkB;SAC1B,CAAC;QAEF,MAAM,QAAQ,GAAG,gBAAgB,CAAC;YAChC,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,kBAAkB,EAAE,EAAI,aAAa;YACvD,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,kBAAkB,CAAC,WAAW,EAAE,gBAAgB,CAAC,EAAE,EAAE,SAAS;SACjF,CAAC,CAAC;QAEH,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,sBAAsB,CAAC,gBAAgB,EAAE;gBAC5D,YAAY,EAAE,IAAI;gBAClB,MAAM,EAAE,UAAU;gBAClB,aAAa,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,UAAU,EAAE,CAAC,EAAE;aACrD,CAAC,CAAC;YAEH,WAAW,CAAC,MAAM,CAAC,gBAAgB,EAAE,CAAC,EAAE,kCAAkC,CAAC,CAAC;YAC5E,WAAW,CAAC,MAAM,CAAC,cAAc,CAAC,MAAM,EAAE,CAAC,EAAE,8BAA8B,CAAC,CAAC;YAC7E,WAAW,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,WAAW,EAAE,sBAAsB,CAAC,CAAC;QACrF,CAAC;gBAAS,CAAC;YACT,QAAQ,EAAE,CAAC;QACb,CAAC;IACH,CAAC;YAAS,CAAC;QACT,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED,8EAA8E;AAC9E,kDAAkD;AAClD,8EAA8E;AAE9E,KAAK,UAAU,sBAAsB;IACnC,OAAO,CAAC,GAAG,CAAC,gFAAgF,CAAC,CAAC;IAE9F,MAAM,OAAO,GAAG,gBAAgB,CAAC;QAC/B,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,IAAI,EAAE,YAAY,EAAE;KAC/C,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,MAAM,OAAO,GAAkB,MAAM,YAAY,CAC/C,gBAAgB,EAChB,OAAO,EACP,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,EACvB,EAAE,KAAK,EAAE,mBAAmB,EAAE,MAAM,EAAE,UAAU,EAAE,CACnD,CAAC;QAEF,2EAA2E;QAC3E,WAAW,CAAC,OAAO,CAAC,OAAO,EAAE,WAAW,EAAE,qCAAqC,CAAC,CAAC;QACjF,MAAM,CAAC,OAAO,CAAC,KAAK,KAAK,IAAI,EAAE,mBAAmB,CAAC,CAAC;QACpD,MAAM,CAAC,OAAO,CAAC,OAAO,KAAK,CAAC,EAAE,uBAAuB,CAAC,CAAC;QACvD,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,QAAQ,CAAC,QAAQ,CAAC,EAAE,4BAA4B,CAAC,CAAC;IAC7E,CAAC;YAAS,CAAC;QACT,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED,8EAA8E;AAC9E,uDAAuD;AACvD,8EAA8E;AAE9E,KAAK,UAAU,mBAAmB;IAChC,OAAO,CAAC,GAAG,CAAC,yEAAyE,CAAC,CAAC;IAEvF,+DAA+D;IAC/D,MAAM,aAAa,GAAG;QACpB,OAAO,EAAE,CAAC;gBACR,IAAI,EAAE,MAAM;gBACZ,IAAI,EAAE,8JAA8J;aACrK,CAAC;QACF,KAAK,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,aAAa,EAAE,EAAE,EAAE;KAChD,CAAC;IAEF,MAAM,OAAO,GAAG,gBAAgB,CAAC;QAC/B,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,aAAa,EAAE;KAClC,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,MAAM,OAAO,GAAkB,MAAM,YAAY,CAC/C,gBAAgB,EAChB,MAAM,EACN,EAAE,KAAK,EAAE,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,EACvB,EAAE,KAAK,EAAE,mBAAmB,EAAE,MAAM,EAAE,UAAU,EAAE,CACnD,CAAC;QAEF,gEAAgE;QAChE,WAAW,CAAC,OAAO,CAAC,OAAO,EAAE,MAAM,EAAE,yCAAyC,CAAC,CAAC;QAChF,MAAM,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,qCAAqC,CAAC,CAAC;IAChE,CAAC;YAAS,CAAC;QACT,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED,8EAA8E;AAC9E,gBAAgB;AAChB,8EAA8E;AAE9E,KAAK,UAAU,IAAI;IACjB,OAAO,CAAC,GAAG,CAAC,mDAAmD,CAAC,CAAC;IACjE,OAAO,CAAC,GAAG,CAAC,uDAAuD,CAAC,CAAC;IAErE,IAAI,CAAC;QACH,MAAM,gBAAgB,EAAE,CAAC;QACzB,MAAM,gBAAgB,EAAE,CAAC;QACzB,MAAM,sBAAsB,EAAE,CAAC;QAC/B,MAAM,qBAAqB,EAAE,CAAC;QAC9B,MAAM,sBAAsB,EAAE,CAAC;QAC/B,MAAM,mBAAmB,EAAE,CAAC;IAC9B,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,OAAO,CAAC,KAAK,CAAC,iCAAiC,EAAE,GAAG,CAAC,CAAC;QACtD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;IAED,OAAO,CAAC,GAAG,CAAC,kBAAkB,MAAM,YAAY,MAAM,aAAa,CAAC,CAAC;IAErE,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QACf,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;AACH,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE;IACjB,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IACnB,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC,CAAC,CAAC"}
|
|
@@ -0,0 +1,22 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* GAIA DAG 5-Question Pilot Runner — iter 64
|
|
3
|
+
*
|
|
4
|
+
* Runs 5 specific GAIA L1 questions through the Co-Sight DAG harness
|
|
5
|
+
* and compares results against single-Sonnet baseline (from iter 63b).
|
|
6
|
+
*
|
|
7
|
+
* Pilot question mix (selected from iter 63b failures):
|
|
8
|
+
* 1. 5d0080cb — calculation (fish bag volume from academic paper, multi-hop)
|
|
9
|
+
* 2. cffe0e32 — reasoning puzzle (Secret Santa assignment chain)
|
|
10
|
+
* 3. ec09fa32 — riddle (game show, requires careful reasoning)
|
|
11
|
+
* 4. 46719c30 — retrieval (paper authors -> their other publications)
|
|
12
|
+
* 5. b816bfce — retrieval+reasoning (journal name from Norse mythology)
|
|
13
|
+
*
|
|
14
|
+
* Cost cap: $2.00
|
|
15
|
+
*
|
|
16
|
+
* Usage:
|
|
17
|
+
* node dist/src/benchmarks/gaia-dag-pilot.js
|
|
18
|
+
*
|
|
19
|
+
* Refs: ADR-139, iter 64, #2156
|
|
20
|
+
*/
|
|
21
|
+
export {};
|
|
22
|
+
//# sourceMappingURL=gaia-dag-pilot.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"gaia-dag-pilot.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/gaia-dag-pilot.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG"}
|
|
@@ -0,0 +1,158 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* GAIA DAG 5-Question Pilot Runner — iter 64
|
|
3
|
+
*
|
|
4
|
+
* Runs 5 specific GAIA L1 questions through the Co-Sight DAG harness
|
|
5
|
+
* and compares results against single-Sonnet baseline (from iter 63b).
|
|
6
|
+
*
|
|
7
|
+
* Pilot question mix (selected from iter 63b failures):
|
|
8
|
+
* 1. 5d0080cb — calculation (fish bag volume from academic paper, multi-hop)
|
|
9
|
+
* 2. cffe0e32 — reasoning puzzle (Secret Santa assignment chain)
|
|
10
|
+
* 3. ec09fa32 — riddle (game show, requires careful reasoning)
|
|
11
|
+
* 4. 46719c30 — retrieval (paper authors -> their other publications)
|
|
12
|
+
* 5. b816bfce — retrieval+reasoning (journal name from Norse mythology)
|
|
13
|
+
*
|
|
14
|
+
* Cost cap: $2.00
|
|
15
|
+
*
|
|
16
|
+
* Usage:
|
|
17
|
+
* node dist/src/benchmarks/gaia-dag-pilot.js
|
|
18
|
+
*
|
|
19
|
+
* Refs: ADR-139, iter 64, #2156
|
|
20
|
+
*/
|
|
21
|
+
import { loadGaia } from './gaia-loader.js';
|
|
22
|
+
import { runDagPilot } from './gaia-dag.js';
|
|
23
|
+
// ---------------------------------------------------------------------------
|
|
24
|
+
// Pilot question IDs (5 questions from iter 63b failures)
|
|
25
|
+
// ---------------------------------------------------------------------------
|
|
26
|
+
const PILOT_TASK_IDS = [
|
|
27
|
+
'5d0080cb-90d7-4712-bc33-848150e917d3', // calculation: fish bag volume
|
|
28
|
+
'cffe0e32-c9a6-4c52-9877-78ceb4aaa9fb', // reasoning: Secret Santa
|
|
29
|
+
'ec09fa32-d03f-4bf8-84b0-1f16922c3ae4', // riddle: game show bees
|
|
30
|
+
'46719c30-f4c3-4cad-be07-d5cb21eee6bb', // retrieval: paper authors + their work
|
|
31
|
+
'b816bfce-3d80-4913-a07d-69b752ce6377', // retrieval: journal from Norse mythology
|
|
32
|
+
];
|
|
33
|
+
// Baseline: single-Sonnet results from iter 63b
|
|
34
|
+
const BASELINE = {
|
|
35
|
+
'5d0080cb-90d7-4712-bc33-848150e917d3': { correct: false, answer: null },
|
|
36
|
+
'cffe0e32-c9a6-4c52-9877-78ceb4aaa9fb': { correct: false, answer: null },
|
|
37
|
+
'ec09fa32-d03f-4bf8-84b0-1f16922c3ae4': { correct: false, answer: null },
|
|
38
|
+
'46719c30-f4c3-4cad-be07-d5cb21eee6bb': { correct: false, answer: null },
|
|
39
|
+
'b816bfce-3d80-4913-a07d-69b752ce6377': { correct: false, answer: 'cuddly' }, // close but wrong
|
|
40
|
+
};
|
|
41
|
+
const COST_CAP_USD = 2.00;
|
|
42
|
+
async function main() {
|
|
43
|
+
console.log('\n=== GAIA DAG 5-Question Pilot (iter 64) ===\n');
|
|
44
|
+
console.log('Questions: 5 (from iter 63b failures)');
|
|
45
|
+
console.log(`Planner: ${process.env['PLAN_MODEL'] ?? 'claude-sonnet-4-6'}`);
|
|
46
|
+
console.log(`Actor: ${process.env['ACT_MODEL'] ?? 'gemini-2.5-pro'}`);
|
|
47
|
+
console.log(`Cost cap: $${COST_CAP_USD}\n`);
|
|
48
|
+
// Load full GAIA L1 dataset and filter to pilot questions
|
|
49
|
+
console.log('Loading GAIA L1 dataset...');
|
|
50
|
+
let allQuestions;
|
|
51
|
+
try {
|
|
52
|
+
allQuestions = await loadGaia({ level: 1 });
|
|
53
|
+
}
|
|
54
|
+
catch (err) {
|
|
55
|
+
console.error('Failed to load GAIA dataset:', err instanceof Error ? err.message : String(err));
|
|
56
|
+
process.exit(1);
|
|
57
|
+
}
|
|
58
|
+
const pilotQuestions = PILOT_TASK_IDS.map((id) => {
|
|
59
|
+
const q = allQuestions.find((q) => q.task_id === id);
|
|
60
|
+
if (!q) {
|
|
61
|
+
console.warn(`WARNING: Question ${id} not found in dataset`);
|
|
62
|
+
}
|
|
63
|
+
return q;
|
|
64
|
+
}).filter(Boolean);
|
|
65
|
+
if (pilotQuestions.length === 0) {
|
|
66
|
+
console.error('No pilot questions found. Check HF token and dataset availability.');
|
|
67
|
+
process.exit(1);
|
|
68
|
+
}
|
|
69
|
+
console.log(`Loaded ${pilotQuestions.length} pilot questions\n`);
|
|
70
|
+
for (const q of pilotQuestions) {
|
|
71
|
+
console.log(` ${q.task_id.slice(0, 8)}: ${q.question.slice(0, 70)}...`);
|
|
72
|
+
console.log(` expected: "${q.final_answer}"`);
|
|
73
|
+
if (q.file_path)
|
|
74
|
+
console.log(` file: ${q.file_path}`);
|
|
75
|
+
}
|
|
76
|
+
console.log('');
|
|
77
|
+
// Run DAG pilot
|
|
78
|
+
const result = await runDagPilot(pilotQuestions, {});
|
|
79
|
+
// Check cost cap
|
|
80
|
+
if (result.totalCostUsd > COST_CAP_USD) {
|
|
81
|
+
console.warn(`\nWARNING: Cost $${result.totalCostUsd.toFixed(4)} exceeded cap $${COST_CAP_USD}\n`);
|
|
82
|
+
}
|
|
83
|
+
// Report results
|
|
84
|
+
console.log('\n=== DAG Pilot Results ===\n');
|
|
85
|
+
console.log(`Score: ${result.correct}/${result.total} (${(result.accuracy * 100).toFixed(1)}%)`);
|
|
86
|
+
console.log(`Avg steps/question: ${result.avgStepsPerQuestion.toFixed(1)}`);
|
|
87
|
+
console.log(`Total cost: $${result.totalCostUsd.toFixed(4)}`);
|
|
88
|
+
console.log(`Projected 53Q cost: $${result.projectedCost53Q.toFixed(2)}`);
|
|
89
|
+
console.log(`Mean wall time: ${(result.meanWallMs / 1000).toFixed(1)}s\n`);
|
|
90
|
+
console.log('Per-question breakdown:');
|
|
91
|
+
const baselineCorrect = Object.values(BASELINE).filter((b) => b.correct).length;
|
|
92
|
+
let dagBetter = 0;
|
|
93
|
+
let dagWorse = 0;
|
|
94
|
+
for (const q of result.perQuestion) {
|
|
95
|
+
const baseline = BASELINE[q.taskId];
|
|
96
|
+
const baselineResult = baseline?.correct ? 'PASS' : 'FAIL';
|
|
97
|
+
const dagResult = q.correct ? 'PASS' : 'FAIL';
|
|
98
|
+
const change = !baseline?.correct && q.correct ? ' (+RECOVERED)' :
|
|
99
|
+
baseline?.correct && !q.correct ? ' (-REGRESSED)' : '';
|
|
100
|
+
if (!baseline?.correct && q.correct)
|
|
101
|
+
dagBetter++;
|
|
102
|
+
if (baseline?.correct && !q.correct)
|
|
103
|
+
dagWorse++;
|
|
104
|
+
console.log(` ${dagResult} [was ${baselineResult}]${change} ${q.taskId.slice(0, 8)}: got="${q.got?.slice(0, 40) ?? 'null'}" expected="${q.expected.slice(0, 40)}"`);
|
|
105
|
+
console.log(` steps=${q.steps} (completed=${q.completedSteps}, blocked=${q.blockedSteps}) plannerCycles=${q.plannerCycles} cost=$${q.costUsd.toFixed(4)} wall=${(q.wallMs / 1000).toFixed(1)}s`);
|
|
106
|
+
}
|
|
107
|
+
console.log(`\nComparison vs single-Sonnet (iter 63b):`);
|
|
108
|
+
console.log(` Baseline: ${baselineCorrect}/5 (${(baselineCorrect * 20).toFixed(0)}%)`);
|
|
109
|
+
console.log(` DAG: ${result.correct}/5 (${(result.accuracy * 100).toFixed(1)}%)`);
|
|
110
|
+
console.log(` Recovered: ${dagBetter} questions (single-Sonnet failed, DAG solved)`);
|
|
111
|
+
console.log(` Regressed: ${dagWorse} questions (single-Sonnet passed, DAG failed)`);
|
|
112
|
+
console.log('\n=== Gate Verdict ===');
|
|
113
|
+
const multiStepExecuted = result.perQuestion.some((q) => q.steps > 1);
|
|
114
|
+
const avgSteps = result.avgStepsPerQuestion;
|
|
115
|
+
if (result.correct >= 4 && multiStepExecuted) {
|
|
116
|
+
console.log(`PROCEED to iter 65 full 53Q run.`);
|
|
117
|
+
console.log(` Reason: DAG ${result.correct}/5, avg ${avgSteps.toFixed(1)} steps/Q, multi-step plans executing.`);
|
|
118
|
+
}
|
|
119
|
+
else if (result.correct >= 3 && avgSteps >= 2) {
|
|
120
|
+
console.log(`PROCEED with CAUTION to iter 65 — diagnose blocking first.`);
|
|
121
|
+
console.log(` Reason: DAG ${result.correct}/5, avg ${avgSteps.toFixed(1)} steps/Q.`);
|
|
122
|
+
}
|
|
123
|
+
else {
|
|
124
|
+
console.log(`DIAGNOSE before iter 65.`);
|
|
125
|
+
console.log(` Reason: DAG ${result.correct}/5, avg ${avgSteps.toFixed(1)} steps/Q.`);
|
|
126
|
+
if (!multiStepExecuted) {
|
|
127
|
+
console.log(` Issue: Plans collapsed to single step — check planner prompt.`);
|
|
128
|
+
}
|
|
129
|
+
}
|
|
130
|
+
// Write JSON result
|
|
131
|
+
const jsonOut = {
|
|
132
|
+
run_type: 'dag-5q-pilot',
|
|
133
|
+
iter: 64,
|
|
134
|
+
timestamp: new Date().toISOString(),
|
|
135
|
+
planModel: process.env['PLAN_MODEL'] ?? 'claude-sonnet-4-6',
|
|
136
|
+
actModel: process.env['ACT_MODEL'] ?? 'gemini-2.5-pro',
|
|
137
|
+
summary: {
|
|
138
|
+
total: result.total,
|
|
139
|
+
correct: result.correct,
|
|
140
|
+
accuracy: result.accuracy,
|
|
141
|
+
avgStepsPerQuestion: result.avgStepsPerQuestion,
|
|
142
|
+
totalCostUsd: result.totalCostUsd,
|
|
143
|
+
projectedCost53Q: result.projectedCost53Q,
|
|
144
|
+
meanWallMs: result.meanWallMs,
|
|
145
|
+
},
|
|
146
|
+
baseline: { correct: baselineCorrect, total: 5, source: 'iter-63b' },
|
|
147
|
+
perQuestion: result.perQuestion,
|
|
148
|
+
};
|
|
149
|
+
const outPath = `/Users/cohen/Projects/ruflo/docs/benchmarks/runs/gaia-l1-iter64-dag-5q-pilot.json`;
|
|
150
|
+
const fs = await import('node:fs');
|
|
151
|
+
fs.writeFileSync(outPath, JSON.stringify(jsonOut, null, 2));
|
|
152
|
+
console.log(`\nResults written to: ${outPath}\n`);
|
|
153
|
+
}
|
|
154
|
+
main().catch((err) => {
|
|
155
|
+
console.error('Pilot failed:', err);
|
|
156
|
+
process.exit(1);
|
|
157
|
+
});
|
|
158
|
+
//# sourceMappingURL=gaia-dag-pilot.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"gaia-dag-pilot.js","sourceRoot":"","sources":["../../../src/benchmarks/gaia-dag-pilot.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,WAAW,EAAkB,MAAM,eAAe,CAAC;AAG5D,8EAA8E;AAC9E,0DAA0D;AAC1D,8EAA8E;AAE9E,MAAM,cAAc,GAAG;IACrB,sCAAsC,EAAE,+BAA+B;IACvE,sCAAsC,EAAE,0BAA0B;IAClE,sCAAsC,EAAE,yBAAyB;IACjE,sCAAsC,EAAE,wCAAwC;IAChF,sCAAsC,EAAE,0CAA0C;CACnF,CAAC;AAEF,gDAAgD;AAChD,MAAM,QAAQ,GAAgE;IAC5E,sCAAsC,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE;IACxE,sCAAsC,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE;IACxE,sCAAsC,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE;IACxE,sCAAsC,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE;IACxE,sCAAsC,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,EAAE,kBAAkB;CACjG,CAAC;AAEF,MAAM,YAAY,GAAG,IAAI,CAAC;AAE1B,KAAK,UAAU,IAAI;IACjB,OAAO,CAAC,GAAG,CAAC,iDAAiD,CAAC,CAAC;IAC/D,OAAO,CAAC,GAAG,CAAC,uCAAuC,CAAC,CAAC;IACrD,OAAO,CAAC,GAAG,CAAC,YAAY,OAAO,CAAC,GAAG,CAAC,YAAY,CAAC,IAAI,mBAAmB,EAAE,CAAC,CAAC;IAC5E,OAAO,CAAC,GAAG,CAAC,YAAY,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC,IAAI,gBAAgB,EAAE,CAAC,CAAC;IACxE,OAAO,CAAC,GAAG,CAAC,cAAc,YAAY,IAAI,CAAC,CAAC;IAE5C,0DAA0D;IAC1D,OAAO,CAAC,GAAG,CAAC,4BAA4B,CAAC,CAAC;IAC1C,IAAI,YAAY,CAAC;IACjB,IAAI,CAAC;QACH,YAAY,GAAG,MAAM,QAAQ,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC,CAAC;IAC9C,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,OAAO,CAAC,KAAK,CAAC,8BAA8B,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;QAChG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;IAED,MAAM,cAAc,GAAG,cAAc,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE;QAC/C,MAAM,CAAC,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,EAAE,CAAC,CAAC;QACrD,IAAI,CAAC,CAAC,EAAE,CAAC;YACP,OAAO,CAAC,IAAI,CAAC,qBAAqB,EAAE,uBAAuB,CAAC,CAAC;QAC/D,CAAC;QACD,OAAO,CAAC,CAAC;IACX,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAwB,CAAC;IAE1C,IAAI,cAAc,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAChC,OAAO,CAAC,KAAK,CAAC,oEAAoE,CAAC,CAAC;QACpF,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;IAED,OAAO,CAAC,GAAG,CAAC,UAAU,cAAc,CAAC,MAAM,oBAAoB,CAAC,CAAC;IACjE,KAAK,MAAM,CAAC,IAAI,cAAc,EAAE,CAAC;QAC/B,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC;QACzE,OAAO,CAAC,GAAG,CAAC,yBAAyB,CAAC,CAAC,YAAY,GAAG,CAAC,CAAC;QACxD,IAAI,CAAC,CAAC,SAAS;YAAE,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC;IAClE,CAAC;IACD,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAEhB,gBAAgB;IAChB,MAAM,MAAM,GAAmB,MAAM,WAAW,CAAC,cAAc,EAAE,EAAE,CAAC,CAAC;IAErE,iBAAiB;IACjB,IAAI,MAAM,CAAC,YAAY,GAAG,YAAY,EAAE,CAAC;QACvC,OAAO,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,kBAAkB,YAAY,IAAI,CAAC,CAAC;IACrG,CAAC;IAED,iBAAiB;IACjB,OAAO,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;IAC7C,OAAO,CAAC,GAAG,CAAC,UAAU,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,KAAK,KAAK,CAAC,MAAM,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IACjG,OAAO,CAAC,GAAG,CAAC,uBAAuB,MAAM,CAAC,mBAAmB,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAC5E,OAAO,CAAC,GAAG,CAAC,gBAAgB,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAC9D,OAAO,CAAC,GAAG,CAAC,wBAAwB,MAAM,CAAC,gBAAgB,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAC1E,OAAO,CAAC,GAAG,CAAC,mBAAmB,CAAC,MAAM,CAAC,UAAU,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC;IAE3E,OAAO,CAAC,GAAG,CAAC,yBAAyB,CAAC,CAAC;IACvC,MAAM,eAAe,GAAG,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC;IAChF,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,QAAQ,GAAG,CAAC,CAAC;IAEjB,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,WAAW,EAAE,CAAC;QACnC,MAAM,QAAQ,GAAG,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;QACpC,MAAM,cAAc,GAAG,QAAQ,EAAE,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;QAC3D,MAAM,SAAS,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;QAC9C,MAAM,MAAM,GAAG,CAAC,QAAQ,EAAE,OAAO,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC;YACnD,QAAQ,EAAE,OAAO,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,EAAE,CAAC;QACtE,IAAI,CAAC,QAAQ,EAAE,OAAO,IAAI,CAAC,CAAC,OAAO;YAAE,SAAS,EAAE,CAAC;QACjD,IAAI,QAAQ,EAAE,OAAO,IAAI,CAAC,CAAC,CAAC,OAAO;YAAE,QAAQ,EAAE,CAAC;QAEhD,OAAO,CAAC,GAAG,CAAC,KAAK,SAAS,SAAS,cAAc,IAAI,MAAM,IAAI,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,UAAU,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,MAAM,eAAe,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC;QACrK,OAAO,CAAC,GAAG,CAAC,kBAAkB,CAAC,CAAC,KAAK,eAAe,CAAC,CAAC,cAAc,aAAa,CAAC,CAAC,YAAY,mBAAmB,CAAC,CAAC,aAAa,UAAU,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;IAC3M,CAAC;IAED,OAAO,CAAC,GAAG,CAAC,2CAA2C,CAAC,CAAC;IACzD,OAAO,CAAC,GAAG,CAAC,eAAe,eAAe,OAAO,CAAC,eAAe,GAAG,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IACxF,OAAO,CAAC,GAAG,CAAC,eAAe,MAAM,CAAC,OAAO,OAAO,CAAC,MAAM,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IACxF,OAAO,CAAC,GAAG,CAAC,gBAAgB,SAAS,+CAA+C,CAAC,CAAC;IACtF,OAAO,CAAC,GAAG,CAAC,gBAAgB,QAAQ,+CAA+C,CAAC,CAAC;IAErF,OAAO,CAAC,GAAG,CAAC,wBAAwB,CAAC,CAAC;IACtC,MAAM,iBAAiB,GAAG,MAAM,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;IACtE,MAAM,QAAQ,GAAG,MAAM,CAAC,mBAAmB,CAAC;IAE5C,IAAI,MAAM,CAAC,OAAO,IAAI,CAAC,IAAI,iBAAiB,EAAE,CAAC;QAC7C,OAAO,CAAC,GAAG,CAAC,kCAAkC,CAAC,CAAC;QAChD,OAAO,CAAC,GAAG,CAAC,iBAAiB,MAAM,CAAC,OAAO,WAAW,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,uCAAuC,CAAC,CAAC;IACpH,CAAC;SAAM,IAAI,MAAM,CAAC,OAAO,IAAI,CAAC,IAAI,QAAQ,IAAI,CAAC,EAAE,CAAC;QAChD,OAAO,CAAC,GAAG,CAAC,4DAA4D,CAAC,CAAC;QAC1E,OAAO,CAAC,GAAG,CAAC,iBAAiB,MAAM,CAAC,OAAO,WAAW,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC;IACxF,CAAC;SAAM,CAAC;QACN,OAAO,CAAC,GAAG,CAAC,0BAA0B,CAAC,CAAC;QACxC,OAAO,CAAC,GAAG,CAAC,iBAAiB,MAAM,CAAC,OAAO,WAAW,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC;QACtF,IAAI,CAAC,iBAAiB,EAAE,CAAC;YACvB,OAAO,CAAC,GAAG,CAAC,iEAAiE,CAAC,CAAC;QACjF,CAAC;IACH,CAAC;IAED,oBAAoB;IACpB,MAAM,OAAO,GAAG;QACd,QAAQ,EAAE,cAAc;QACxB,IAAI,EAAE,EAAE;QACR,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACnC,SAAS,EAAE,OAAO,CAAC,GAAG,CAAC,YAAY,CAAC,IAAI,mBAAmB;QAC3D,QAAQ,EAAE,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC,IAAI,gBAAgB;QACtD,OAAO,EAAE;YACP,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,OAAO,EAAE,MAAM,CAAC,OAAO;YACvB,QAAQ,EAAE,MAAM,CAAC,QAAQ;YACzB,mBAAmB,EAAE,MAAM,CAAC,mBAAmB;YAC/C,YAAY,EAAE,MAAM,CAAC,YAAY;YACjC,gBAAgB,EAAE,MAAM,CAAC,gBAAgB;YACzC,UAAU,EAAE,MAAM,CAAC,UAAU;SAC9B;QACD,QAAQ,EAAE,EAAE,OAAO,EAAE,eAAe,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,UAAU,EAAE;QACpE,WAAW,EAAE,MAAM,CAAC,WAAW;KAChC,CAAC;IAEF,MAAM,OAAO,GAAG,mFAAmF,CAAC;IACpG,MAAM,EAAE,GAAG,MAAM,MAAM,CAAC,SAAS,CAAC,CAAC;IACnC,EAAE,CAAC,aAAa,CAAC,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IAC5D,OAAO,CAAC,GAAG,CAAC,yBAAyB,OAAO,IAAI,CAAC,CAAC;AACpD,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;IACnB,OAAO,CAAC,KAAK,CAAC,eAAe,EAAE,GAAG,CAAC,CAAC;IACpC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC,CAAC,CAAC"}
|
|
@@ -0,0 +1,110 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* GAIA DAG Harness — Co-Sight Architecture Port (ADR-139 Addendum)
|
|
3
|
+
*
|
|
4
|
+
* Ports the ZTE-AICloud/Co-Sight DAG orchestration pattern (Apache 2.0,
|
|
5
|
+
* arXiv 2510.21557) into the ruflo GAIA harness.
|
|
6
|
+
*
|
|
7
|
+
* Architecture:
|
|
8
|
+
* 1. PLAN — Claude Sonnet 4.6 reads the question and emits a DAG of 3-7
|
|
9
|
+
* steps: {id, description, depends_on: [], suggested_tool}.
|
|
10
|
+
* Claude-aware prompt: "3-5 steps max, direct answer when clear."
|
|
11
|
+
* 2. EXECUTE — Loop while ready steps exist (deps satisfied).
|
|
12
|
+
* Run all ready steps in PARALLEL (Promise.all, cap ≤5).
|
|
13
|
+
* Each step = a Gemini 2.5 Pro actor with the full tool suite.
|
|
14
|
+
* Actor marks step completed/blocked + writes step_notes.
|
|
15
|
+
* Blocked steps trigger planner re_plan before next cycle.
|
|
16
|
+
* 3. FINALIZE — Planner reads all step_notes → produces final answer
|
|
17
|
+
* using T2 extraction cascade from gaia-agent.ts.
|
|
18
|
+
* 4. CAMV — Async credibility labeling per step (stubbed, iter 65).
|
|
19
|
+
*
|
|
20
|
+
* Role assignment (env-configurable):
|
|
21
|
+
* PLAN_MODEL = claude-sonnet-4-6 (default)
|
|
22
|
+
* ACT_MODEL = gemini-2.5-pro (default)
|
|
23
|
+
* VISION_MODEL = gemini-2.5-pro (default, same as act)
|
|
24
|
+
*
|
|
25
|
+
* CLI:
|
|
26
|
+
* gaia-bench run --mode=dag --model claude-sonnet-4-6
|
|
27
|
+
*
|
|
28
|
+
* Cost: planner ~$0.02/Q + actors ~$0.03/Q = ~$0.05/Q (vs single-Sonnet ~$0.075)
|
|
29
|
+
*
|
|
30
|
+
* Refs: ADR-139, github.com/ZTE-AICloud/Co-Sight, arXiv 2510.21557, #2156
|
|
31
|
+
*/
|
|
32
|
+
import { GaiaQuestion } from './gaia-loader.js';
|
|
33
|
+
import { GaiaToolCatalogue } from './gaia-tools/index.js';
|
|
34
|
+
export interface DagStep {
|
|
35
|
+
id: number;
|
|
36
|
+
description: string;
|
|
37
|
+
depends_on: number[];
|
|
38
|
+
suggested_tool?: string;
|
|
39
|
+
status: 'not_started' | 'in_progress' | 'completed' | 'blocked';
|
|
40
|
+
step_notes: string;
|
|
41
|
+
}
|
|
42
|
+
export interface DagPlan {
|
|
43
|
+
title: string;
|
|
44
|
+
question: string;
|
|
45
|
+
steps: DagStep[];
|
|
46
|
+
}
|
|
47
|
+
/**
|
|
48
|
+
* Get all steps whose dependencies are fully satisfied (completed).
|
|
49
|
+
* Mirrors Co-Sight's Plan.get_ready_steps().
|
|
50
|
+
*/
|
|
51
|
+
export declare function getReadySteps(plan: DagPlan): DagStep[];
|
|
52
|
+
export interface DagResult {
|
|
53
|
+
questionId: string;
|
|
54
|
+
finalAnswer: string | null;
|
|
55
|
+
normalisedAnswer: string;
|
|
56
|
+
plan: DagPlan;
|
|
57
|
+
totalSteps: number;
|
|
58
|
+
completedSteps: number;
|
|
59
|
+
blockedSteps: number;
|
|
60
|
+
plannerCycles: number;
|
|
61
|
+
totalInputTokens: number;
|
|
62
|
+
totalOutputTokens: number;
|
|
63
|
+
estimatedCostUsd: number;
|
|
64
|
+
wallMs: number;
|
|
65
|
+
timedOut?: boolean;
|
|
66
|
+
error?: string;
|
|
67
|
+
}
|
|
68
|
+
export interface DagOptions {
|
|
69
|
+
planModel?: string;
|
|
70
|
+
actModel?: string;
|
|
71
|
+
anthropicApiKey?: string;
|
|
72
|
+
geminiApiKey?: string;
|
|
73
|
+
catalogue?: GaiaToolCatalogue;
|
|
74
|
+
/** Disable CAMV stub (default: false — stub is always disabled in this iter). */
|
|
75
|
+
enableCamv?: boolean;
|
|
76
|
+
}
|
|
77
|
+
/**
|
|
78
|
+
* Run a GAIA question through the Co-Sight DAG harness.
|
|
79
|
+
*
|
|
80
|
+
* Steps:
|
|
81
|
+
* 1. Planner (Claude Sonnet) creates a DAG plan.
|
|
82
|
+
* 2. Execute loop: parallel actors (Gemini 2.5 Pro) run ready steps.
|
|
83
|
+
* 3. Blocked steps trigger replan (up to MAX_REPLAN_CYCLES).
|
|
84
|
+
* 4. Finalizer (Claude Sonnet) reads all step notes → final answer.
|
|
85
|
+
*/
|
|
86
|
+
export declare function runGaiaDAG(question: GaiaQuestion, options?: DagOptions): Promise<DagResult>;
|
|
87
|
+
export interface DagPilotResult {
|
|
88
|
+
correct: number;
|
|
89
|
+
total: number;
|
|
90
|
+
accuracy: number;
|
|
91
|
+
avgStepsPerQuestion: number;
|
|
92
|
+
perQuestion: Array<{
|
|
93
|
+
taskId: string;
|
|
94
|
+
question: string;
|
|
95
|
+
expected: string;
|
|
96
|
+
got: string | null;
|
|
97
|
+
correct: boolean;
|
|
98
|
+
steps: number;
|
|
99
|
+
completedSteps: number;
|
|
100
|
+
blockedSteps: number;
|
|
101
|
+
plannerCycles: number;
|
|
102
|
+
costUsd: number;
|
|
103
|
+
wallMs: number;
|
|
104
|
+
}>;
|
|
105
|
+
totalCostUsd: number;
|
|
106
|
+
projectedCost53Q: number;
|
|
107
|
+
meanWallMs: number;
|
|
108
|
+
}
|
|
109
|
+
export declare function runDagPilot(questions: GaiaQuestion[], options?: DagOptions): Promise<DagPilotResult>;
|
|
110
|
+
//# sourceMappingURL=gaia-dag.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"gaia-dag.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/gaia-dag.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AAKH,OAAO,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAChD,OAAO,EAEL,iBAAiB,EAKlB,MAAM,uBAAuB,CAAC;AAiD/B,MAAM,WAAW,OAAO;IACtB,EAAE,EAAE,MAAM,CAAC;IACX,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,EAAE,CAAC;IACrB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,MAAM,EAAE,aAAa,GAAG,aAAa,GAAG,WAAW,GAAG,SAAS,CAAC;IAChE,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,OAAO;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,OAAO,EAAE,CAAC;CAClB;AAED;;;GAGG;AACH,wBAAgB,aAAa,CAAC,IAAI,EAAE,OAAO,GAAG,OAAO,EAAE,CAQtD;AAygBD,MAAM,WAAW,SAAS;IACxB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,gBAAgB,EAAE,MAAM,CAAC;IACzB,IAAI,EAAE,OAAO,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,cAAc,EAAE,MAAM,CAAC;IACvB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,gBAAgB,EAAE,MAAM,CAAC;IACzB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,gBAAgB,EAAE,MAAM,CAAC;IACzB,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,UAAU;IACzB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,SAAS,CAAC,EAAE,iBAAiB,CAAC;IAC9B,iFAAiF;IACjF,UAAU,CAAC,EAAE,OAAO,CAAC;CACtB;AA6BD;;;;;;;;GAQG;AACH,wBAAsB,UAAU,CAC9B,QAAQ,EAAE,YAAY,EACtB,OAAO,GAAE,UAAe,GACvB,OAAO,CAAC,SAAS,CAAC,CA6IpB;AAMD,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,mBAAmB,EAAE,MAAM,CAAC;IAC5B,WAAW,EAAE,KAAK,CAAC;QACjB,MAAM,EAAE,MAAM,CAAC;QACf,QAAQ,EAAE,MAAM,CAAC;QACjB,QAAQ,EAAE,MAAM,CAAC;QACjB,GAAG,EAAE,MAAM,GAAG,IAAI,CAAC;QACnB,OAAO,EAAE,OAAO,CAAC;QACjB,KAAK,EAAE,MAAM,CAAC;QACd,cAAc,EAAE,MAAM,CAAC;QACvB,YAAY,EAAE,MAAM,CAAC;QACrB,aAAa,EAAE,MAAM,CAAC;QACtB,OAAO,EAAE,MAAM,CAAC;QAChB,MAAM,EAAE,MAAM,CAAC;KAChB,CAAC,CAAC;IACH,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,wBAAsB,WAAW,CAC/B,SAAS,EAAE,YAAY,EAAE,EACzB,OAAO,GAAE,UAAe,GACvB,OAAO,CAAC,cAAc,CAAC,CA6CzB"}
|