{
  "schemaVersion": 1,
  "studyId": "felan-extensions-2026-09",
  "status": "complete",
  "createdAt": "2026-09-04T12:02:05Z",
  "repository": {
    "commit": "f7530cac88fb0d86e95bb713040d66df6abcadd3",
    "cleanBeforeFreeze": true,
    "configSha256": "7a847181e54b618d92aa672bf7a5a6c29eceecacda0af7c4600ddbdfd600c170",
    "configFrozenRelativePath": "sources/provenance/project/felan-extension-evals.yaml",
    "activeConfigRelativePath": "felan-extension-evals.yaml",
    "configRelativePathAtFreeze": "harness-evals.yaml"
  },
  "frameworkAtFreeze": {
    "dependencySpec": "^0.2.15",
    "localLinkActive": true,
    "linkedPackageVersion": "0.2.13",
    "linkedGitCommit": "d5aaab2afc5f9536a86bbf8ffefd67def1c5c002"
  },
  "canonicalFiles": [
    "README.md",
    "README-snippet.md",
    "index.html",
    "results.json",
    "results.csv",
    "manifest.json",
    "SHA256SUMS"
  ],
  "sourcePolicy": {
    "sourceCopies": "Selected batch manifest, redacted source evidence, frozen configuration, aggregate configuration, reverification manifests, and recalculation evidence are preserved under sources/; duplicate framework report copies are omitted.",
    "excludedSensitiveArtifacts": [
      "credentials",
      "auth stores",
      "model stores",
      "raw workspaces",
      "full transcripts",
      "full session JSONL"
    ],
    "subagentsCostProjection": "Cost-only descendant aggregates and raw source hashes are retained without conversation content."
  },
  "primarySelections": [
    {
      "benchmarkId": "subagents",
      "selectedBatchId": "20260903-052334-7809",
      "sourceBatchId": "20260903-052334-7809",
      "reverificationId": null,
      "evidenceType": "recalculated-direct",
      "report": "benchmarks/subagents/results.html",
      "publishedOutcomes": {
        "quality": "available",
        "cost": "recalculated"
      },
      "recalculationEvidence": "sources/provenance/subagents-cost-recalculation.json"
    },
    {
      "benchmarkId": "markitdown-cost",
      "selectedBatchId": "20260903-045947-5857",
      "sourceBatchId": "20260902-112808-33e8",
      "reverificationId": "20260903-045947-56b5",
      "evidenceType": "offline-reverification",
      "report": "benchmarks/markitdown-cost/results.json"
    },
    {
      "benchmarkId": "output-style-concise",
      "selectedBatchId": "20260903-045947-ba0e",
      "sourceBatchId": "20260902-121757-89d1",
      "reverificationId": "20260903-045947-56b5",
      "evidenceType": "offline-reverification",
      "report": "benchmarks/output-style-concise/results.json"
    },
    {
      "benchmarkId": "prewalk",
      "selectedBatchId": "20260903-044517-8452",
      "sourceBatchId": "20260902-121757-b16d",
      "reverificationId": "20260903-044517-d1f1",
      "evidenceType": "offline-reverification",
      "report": "benchmarks/prewalk/results.json"
    },
    {
      "benchmarkId": "rtk",
      "selectedBatchId": "20260903-045947-e11d",
      "sourceBatchId": "20260902-121847-5fa4",
      "reverificationId": "20260903-045947-56b5",
      "evidenceType": "offline-reverification",
      "report": "benchmarks/rtk/results.json"
    },
    {
      "benchmarkId": "codebase-memory",
      "selectedBatchId": "20260904-124556-6ffe",
      "sourceBatchId": "20260904-124556-6ffe",
      "reverificationId": null,
      "evidenceType": "derived-redacted-local-batch",
      "report": "benchmarks/codebase-memory/results.json",
      "evidence": "sources/provenance/codebase-memory.json",
      "publicationScope": {
        "cases": 3,
        "runs": 18,
        "trialsPerArmPerCase": 3
      }
    }
  ],
  "knownLimitations": [
    "Subagents cost is a documented recalculation that adds billed child-session usage to the native parent-session totals.",
    "Headline resource outcomes use ratioOfReducedSums: median trial value within each case, then a sum across cases. Macro-mean case gains and ranges remain diagnostics.",
    "The Codebase Memory result covers all 18 retained attempts across three cases; its exact source harness-evals.yaml was not supplied.",
    "RTK candidate aggregate quality is 83.3% (5/6); its detailed framework report also retains the reducer-derived arm value.",
    "Detailed benchmark CSV attempt rows omit the final empty column; the files remain parseable and the canonical top-level results.csv is rectangular."
  ],
  "sourceFiles": [
    {
      "path": "sources/provenance/aggregate-gain-recalculation.json",
      "sha256": "4f7f7942521e128f177c5e20400a6d577d92a06665cfc1256c2f630b61567d7c",
      "bytes": 25793,
      "preservation": "derived-aggregate-recalculation"
    },
    {
      "path": "sources/provenance/batches/20260902-112808-33e8.json",
      "sha256": "bfea0be5ec9eeae3f81153c4ec137e7f3263f655270f2d8ea0253b6d048b2d0b",
      "bytes": 2577,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/batches/20260902-121757-89d1.json",
      "sha256": "4a5f0b6452494658300ac3f6b5ca96f7005bc6582ebaf0240f2a74d97466bc85",
      "bytes": 1938,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/batches/20260902-121757-b16d.json",
      "sha256": "25a42a3c26ea99b26e24f06b54e23ca0be77777ee8d1f5503597fbfa7225604a",
      "bytes": 1417,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/batches/20260902-121847-5fa4.json",
      "sha256": "9503c424ae9d416ae825a971358391d1bb121c7811b296f64f6bced9f87b0635",
      "bytes": 2001,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/batches/20260903-044517-8452.json",
      "sha256": "55ca33a4678dd8050dca5238d46b2094f2ae87e657e64cb6833ef41ab458c2bb",
      "bytes": 1494,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/batches/20260903-045947-5857.json",
      "sha256": "778ac99eb732f01ff9a7aa18c31a8f473178a0cfb3f986aee18c1eccc224f897",
      "bytes": 2778,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/batches/20260903-045947-ba0e.json",
      "sha256": "da5e302fb54156b8e36ea19ea3d8ba137aa455273248e7a7279d66524f66656b",
      "bytes": 2089,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/batches/20260903-045947-e11d.json",
      "sha256": "078add0aa58e9748011610e86968f3b7ae9cafad9c7c4ebb34155b23808bcc74",
      "bytes": 1630,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/batches/20260903-052334-7809.json",
      "sha256": "9c59e7c727c842d2124eb95fe5d5a5f3616124bb7e571d5a1d1026b8c52f3c2d",
      "bytes": 1122,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/batches/20260904-124556-6ffe.json",
      "sha256": "2beacb30dfaaf8d10ad13c1dacb873d41919a48ce4059a7853a85455b2538c6e",
      "bytes": 1897,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/codebase-memory.json",
      "sha256": "ff4570487872a01dc315cea42d17311703791b1a4c9cfc3083eed099d02d2ce3",
      "bytes": 48140,
      "preservation": "derived-published-evidence"
    },
    {
      "path": "sources/provenance/project/felan-extension-evals-aggregate.yaml",
      "sha256": "dd0016a08e30255826cbf724d4394d1a4a5855828e630a629035610ebb333ca9",
      "bytes": 25022,
      "preservation": "derived-aggregate-config"
    },
    {
      "path": "sources/provenance/project/felan-extension-evals.yaml",
      "sha256": "7a847181e54b618d92aa672bf7a5a6c29eceecacda0af7c4600ddbdfd600c170",
      "bytes": 24724,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/reverifications/20260903-044517-d1f1.json",
      "sha256": "1d70c174e8a4c0e033bcc842f3d79f32394e73df44ce44d02f385bee33ebfa24",
      "bytes": 120420,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/reverifications/20260903-045947-56b5.json",
      "sha256": "7909a852d47d94eefdf18fd83f8b7f8efc547310efd4c92423e5f7ad266d93c3",
      "bytes": 88804,
      "preservation": "byte-preserved-copy"
    },
    {
      "path": "sources/provenance/selected-evidence.json",
      "sha256": "907bff8d1db75bb5bc0ad42d32ca8cbabc27bcb142b9fe0f2caa0da3dab95076",
      "bytes": 30593,
      "preservation": "derived-provenance-index"
    },
    {
      "path": "sources/provenance/subagents-cost-recalculation.json",
      "sha256": "711b3265612210842bbdda9ff549a0de663d9693a8e4ef95006250b7dab600e3",
      "bytes": 7742,
      "preservation": "derived-cost-recalculation"
    }
  ],
  "presentationAnnotations": [
    {
      "benchmarkId": "rtk",
      "scope": "Quality aggregation disclosure",
      "reason": "Candidate passed 5/6 attempts while the reducer-derived arm value is 1.",
      "report": "benchmarks/rtk/results.json"
    },
    {
      "benchmarkId": "subagents",
      "scope": "Descendant-inclusive cost detail",
      "reason": "The native report omits child-session cost.",
      "report": "benchmarks/subagents/results.html"
    },
    {
      "benchmarkId": "all-comparable",
      "scope": "Headline aggregation",
      "reason": "Primary and secondary additive resource outcomes use the ratio of median-reduced case sums; macro case gains remain diagnostics.",
      "evidence": "sources/provenance/aggregate-gain-recalculation.json"
    },
    {
      "benchmarkId": "codebase-memory",
      "scope": "Published evidence scope",
      "reason": "The final current result covers all 18 retained attempts across three cases; raw sensitive run artifacts remain excluded.",
      "report": "benchmarks/codebase-memory/results.html"
    }
  ],
  "designSystem": {
    "name": "Felan AI Design System",
    "sourceCommit": "7b26e2675debe55717725f11c32f0772eb3b4b7b",
    "integration": "Copied canonical styles, tokens, and logo into a report-local asset directory.",
    "files": [
      {
        "packagedPath": "assets/felan/styles.css",
        "sha256": "632c0d7f22e19646b27a0ccc164ec6e8bdd37cfa0ae65b37bf9a4d3b5682a141",
        "sourcePathFromProjectRoot": "../design-system/project/styles.css"
      },
      {
        "packagedPath": "assets/felan/colors_and_type.css",
        "sha256": "7cfc5b8b934bbe33a7f8037b8caeba8b14d2867e498a2f994b329dfa5dd25c8e",
        "sourcePathFromProjectRoot": "../design-system/project/colors_and_type.css"
      },
      {
        "packagedPath": "assets/felan/felan-logo.svg",
        "sha256": "7a8eb8b8fb6ea8298428e3e0e4cfcc53b406a17d2ed6332d7fefcba92edcb6bf",
        "sourcePathFromProjectRoot": "../design-system/project/assets/felan-logo.svg"
      }
    ],
    "sourceRepositoryPathFromProjectRoot": "../design-system"
  },
  "manualCorrections": [
    {
      "benchmarkId": "subagents",
      "scope": "Candidate cost only",
      "method": "Add child assistant billed cost to each parent run before benchmark aggregation.",
      "evidence": "sources/provenance/subagents-cost-recalculation.json",
      "descendantSessions": 9,
      "descendantAssistantRequests": 75
    }
  ],
  "readmeSnippet": {
    "path": "README-snippet.md",
    "target": "felan-ai/felan README",
    "linkBase": "https://felan-ai.github.io/harness-bench/results/2026-09-felan-extensions/",
    "requires": "GitHub Pages publishing the harness-bench main branch from the repository root"
  },
  "publicationAggregation": {
    "mode": "ratioOfReducedSums",
    "trials": "median",
    "evidence": "sources/provenance/aggregate-gain-recalculation.json",
    "activeConfig": "sources/provenance/project/felan-extension-evals-aggregate.yaml",
    "legacyDiagnostics": "Macro mean case-level gains and ranges remain in detailed report JSON.",
    "scope": "Six primary benchmark comparisons"
  },
  "derivedReports": [
    "benchmarks/markitdown-cost/results.json",
    "benchmarks/markitdown-cost/results.csv",
    "benchmarks/markitdown-cost/results.html",
    "benchmarks/output-style-concise/results.json",
    "benchmarks/output-style-concise/results.csv",
    "benchmarks/output-style-concise/results.html",
    "benchmarks/prewalk/results.json",
    "benchmarks/prewalk/results.csv",
    "benchmarks/prewalk/results.html",
    "benchmarks/rtk/results.json",
    "benchmarks/rtk/results.csv",
    "benchmarks/rtk/results.html",
    "benchmarks/codebase-memory/results.json",
    "benchmarks/codebase-memory/results.csv",
    "benchmarks/codebase-memory/results.html"
  ]
}