@caupulican/pi-adaptative 0.81.42 → 0.81.43

package/dist/core/tool-recovery-stats.js.map

@@ -1 +1 @@
-{"version":3,"file":"tool-recovery-stats.js","sourceRoot":"","sources":["../../src/core/tool-recovery-stats.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EACN,UAAU,EACV,SAAS,EACT,WAAW,EACX,YAAY,EACZ,UAAU,EACV,QAAQ,EACR,UAAU,EACV,aAAa,GACb,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAM1C,OAAO,EACN,iCAAiC,GAEjC,MAAM,gCAAgC,CAAC;AAExC,MAAM,eAAe,GAAG,KAAK,CAAC;AAC9B,MAAM,eAAe,GAAG,KAAK,CAAC;AAC9B,MAAM,qBAAqB,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;AACvD,MAAM,gBAAgB,GAAG,WAAW,CAAC;AAgCrC,MAAM,UAAU,sCAAsC,GAAgC;IACrF,OAAO;QACN,KAAK,EAAE,CAAC;QACR,QAAQ,EAAE,CAAC;QACX,OAAO,EAAE,CAAC;QACV,YAAY,EAAE,EAAE;QAChB,cAAc,EAAE,EAAE;QAClB,MAAM,EAAE,EAAE,IAAI,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE;QACrC,gBAAgB,EAAE,EAAE,OAAO,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE;QACzD,aAAa,EAAE,EAAE;KACjB,CAAC;AAAA,CACF;AAED,SAAS,gBAAgB,CAAC,MAA+B,EAAE,GAAW,EAAU;IAC/E,IAAI,MAAM,CAAC,GAAG,CAAC,KAAK,SAAS,IAAI,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,MAAM,GAAG,eAAe,GAAG,CAAC;QAAE,OAAO,GAAG,CAAC;IAC9F,OAAO,gBAAgB,CAAC;AAAA,CACxB;AAED,MAAM,UAAU,mCAAmC,CAClD,KAAkC,EAClC,MAAoC,EAC7B;IACP,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IAC3B,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC;IACvC,MAAM,gBAAgB,GAAG,MAAM,CAAC,gBAAgB,IAAI,SAAS,CAAC;IAC9D,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;IAC1B,KAAK,CAAC,gBAAgB,CAAC,gBAAgB,CAAC,IAAI,CAAC,CAAC;IAC9C,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,MAAM,CAAC,YAAY,EAAE,GAAG,MAAM,CAAC,cAAc,CAAC,CAAC,CAAC;IAC1E,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,YAAY,EAAE,CAAC;QACxC,MAAM,GAAG,GAAG,gBAAgB,CAAC,KAAK,CAAC,YAAY,EAAE,IAAI,CAAC,CAAC;QACvD,KAAK,CAAC,YAAY,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;IAC9D,CAAC;IACD,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,cAAc,EAAE,CAAC;QAC5C,MAAM,GAAG,GAAG,gBAAgB,CAAC,KAAK,CAAC,cAAc,EAAE,MAAM,CAAC,CAAC;QAC3D,KAAK,CAAC,cAAc,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,cAAc,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;IAClE,CAAC;IACD,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QAC1B,MAAM,GAAG,GAAG,gBAAgB,CAC3B,KAAK,CAAC,aAAa,EACnB,GAAG,MAAM,CAAC,QAAQ,IAAI,SAAS,IAAI,MAAM,CAAC,KAAK,IAAI,SAAS,IAAI,IAAI,EAAE,CACtE,CAAC;QACF,IAAI,QAAQ,GAAG,KAAK,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC;QACxC,IAAI,CAAC,QAAQ,EAAE,CAAC;YACf,QAAQ,GAAG;gBACV,gBAAgB,EAAE,CAAC;gBACnB,eAAe,EAAE,CAAC;gBAClB,qBAAqB,EAAE,CAAC;gBACxB,kBAAkB,EAAE,CAAC;gBACrB,kBAAkB,EAAE,CAAC;aACrB,CAAC;YACF,KAAK,CAAC,aAAa,CAAC,GAAG,CAAC,GAAG,QAAQ,CAAC;QACrC,CAAC;QACD,IAAI,MAAM,KAAK,MAAM;YAAE,QAAQ,CAAC,gBAAgB,EAAE,CAAC;;YAC9C,QAAQ,CAAC,eAAe,EAAE,CAAC;IACjC,CAAC;IACD,IAAI,MAAM,CAAC,OAAO,KAAK,UAAU,EAAE,CAAC;QACnC,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,cAAc,EAAE,CAAC;YAC5C,MAAM,GAAG,GAAG,gBAAgB,CAC3B,KAAK,CAAC,aAAa,EACnB,GAAG,MAAM,CAAC,QAAQ,IAAI,SAAS,IAAI,MAAM,CAAC,KAAK,IAAI,SAAS,IAAI,MAAM,EAAE,CACxE,CAAC;YACF,IAAI,QAAQ,GAAG,KAAK,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC;YACxC,IAAI,CAAC,QAAQ,EAAE,CAAC;gBACf,QAAQ,GAAG;oBACV,gBAAgB,EAAE,CAAC;oBACnB,eAAe,EAAE,CAAC;oBAClB,qBAAqB,EAAE,CAAC;oBACxB,kBAAkB,EAAE,CAAC;oBACrB,kBAAkB,EAAE,CAAC;iBACrB,CAAC;gBACF,KAAK,CAAC,aAAa,CAAC,GAAG,CAAC,GAAG,QAAQ,CAAC;YACrC,CAAC;YACD,IAAI,gBAAgB,KAAK,WAAW;gBAAE,QAAQ,CAAC,qBAAqB,EAAE,CAAC;YACvE,IAAI,gBAAgB,KAAK,QAAQ;gBAAE,QAAQ,CAAC,kBAAkB,EAAE,CAAC;YACjE,IAAI,gBAAgB,KAAK,SAAS;gBAAE,QAAQ,CAAC,kBAAkB,EAAE,CAAC;QACnE,CAAC;IACF,CAAC;AAAA,CACD;AAED,MAAM,UAAU,6BAA6B,CAC5C,MAAuE,EAC9D;IACT,MAAM,MAAM,GAAG,GAAG,MAAM,CAAC,SAAS,GAAG,CAAC;IACtC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,MAAM,CAAC;QAAE,OAAO,CAAC,CAAC,CAAC;IACnD,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;IAC9D,OAAO,MAAM,CAAC,aAAa,CAAC,QAAQ,CAAC,IAAI,QAAQ,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AAAA,CACvE;AAED,SAAS,SAAS,CAAC,YAAoB,EAAE,SAAiB,EAAU;IACnE,MAAM,EAAE,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IAC7E,OAAO,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,qBAAqB,EAAE,GAAG,EAAE,OAAO,CAAC,CAAC;AAAA,CACxE;AAED,SAAS,aAAa,CAAC,KAAc,EAAmB;IACvD,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC;AAAA,CACzE;AAED,SAAS,aAAa,CAAC,KAAc,EAAmC;IACvE,OAAO,CAAC,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;AAAA,CACzF;AAED,SAAS,YAAY,CAAC,KAAc,EAAwC;IAC3E,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IACtD,MAAM,KAAK,GAAG,KAA6C,CAAC;IAC5D,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC;QAAE,OAAO,KAAK,CAAC;IACjH,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,cAAc,CAAC;QAAE,OAAO,KAAK,CAAC;IAC7F,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,gBAAgB,CAAC;QAAE,OAAO,KAAK,CAAC;IACzF,IAAI,CAAC,KAAK,CAAC,aAAa,IAAI,OAAO,KAAK,CAAC,aAAa,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IAClF,OAAO,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC,KAAK,CAC9C,CAAC,KAAK,EAAE,EAAE,CACT,CAAC,CAAC,KAAK;QACP,aAAa,CAAC,KAAK,CAAC,gBAAgB,CAAC;QACrC,aAAa,CAAC,KAAK,CAAC,eAAe,CAAC;QACpC,aAAa,CAAC,KAAK,CAAC,qBAAqB,CAAC;QAC1C,aAAa,CAAC,KAAK,CAAC,kBAAkB,CAAC;QACvC,aAAa,CAAC,KAAK,CAAC,kBAAkB,CAAC,CACxC,CAAC;AAAA,CACF;AAED,SAAS,gBAAgB,CAAC,KAAc,EAAE,SAAiB,EAAuC;IACjG,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IACtD,MAAM,MAAM,GAAG,KAA4C,CAAC;IAC5D,OAAO,CACN,MAAM,CAAC,OAAO,KAAK,CAAC;QACpB,MAAM,CAAC,SAAS,KAAK,SAAS;QAC9B,aAAa,CAAC,MAAM,CAAC,kBAAkB,CAAC;QACxC,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAC1B,CAAC;AAAA,CACF;AAED,MAAM,UAAU,8BAA8B,CAC7C,YAAoB,EACpB,SAAiB,EACwB;IACzC,MAAM,IAAI,GAAG,SAAS,CAAC,YAAY,EAAE,SAAS,CAAC,CAAC;IAChD,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,OAAO,SAAS,CAAC;IACxC,IAAI,CAAC;QACJ,MAAM,MAAM,GAAY,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC;QAChE,OAAO,gBAAgB,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC;IACjE,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,SAAS,CAAC;IAClB,CAAC;AAAA,CACD;AAED,SAAS,eAAe,CAAC,SAAiB,EAAE,WAAmB,EAAQ;IACtE,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACvB,MAAM,KAAK,GAAG,WAAW,CAAC,SAAS,CAAC;SAClC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;SACxC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC;QACd,MAAM,IAAI,GAAG,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;QACnC,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,QAAQ,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,CAAC;IAAA,CACjD,CAAC;SACD,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC;IACtD,IAAI,QAAQ,GAAG,KAAK,CAAC,MAAM,CAAC;IAC5B,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QAC1B,IAAI,IAAI,CAAC,IAAI,KAAK,WAAW;YAAE,SAAS;QACxC,IAAI,QAAQ,IAAI,eAAe,IAAI,GAAG,GAAG,IAAI,CAAC,OAAO,IAAI,qBAAqB;YAAE,SAAS;QACzF,IAAI,CAAC;YACJ,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACtB,QAAQ,EAAE,CAAC;QACZ,CAAC;QAAC,MAAM,CAAC,CAAA,CAAC;IACX,CAAC;AAAA,CACD;AAED,SAAS,+BAA+B,CAAC,YAAoB,EAAE,OAAmC,EAAQ;IACzG,MAAM,IAAI,GAAG,SAAS,CAAC,YAAY,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC;IACxD,MAAM,SAAS,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAChC,SAAS,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC1C,MAAM,YAAY,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;IACvC,MAAM,aAAa,GAAG,GAAG,IAAI,IAAI,OAAO,CAAC,GAAG,MAAM,CAAC;IACnD,aAAa,CAAC,aAAa,EAAE,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,EAAE,OAAO,CAAC,CAAC;IAC/D,UAAU,CAAC,aAAa,EAAE,IAAI,CAAC,CAAC;IAChC,IAAI,YAAY;QAAE,eAAe,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;AAAA,CACnD;AAED,MAAM,UAAU,gCAAgC,CAAC,YAAoB,EAAE,MAAuC,EAAQ;IACrH,IAAI,OAAO,GAAG,8BAA8B,CAAC,YAAY,EAAE,MAAM,CAAC,SAAS,CAAC,CAAC;IAC7E,IAAI,CAAC,OAAO,EAAE,CAAC;QACd,OAAO,GAAG;YACT,OAAO,EAAE,CAAC;YACV,SAAS,EAAE,MAAM,CAAC,SAAS;YAC3B,kBAAkB,EAAE,CAAC,CAAC;YACtB,KAAK,EAAE,sCAAsC,EAAE;SAC/C,CAAC;QACF,IAAI,CAAC;YACJ,KAAK,MAAM,IAAI,IAAI,YAAY,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;gBACpE,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC;oBAAE,SAAS;gBACvC,MAAM,MAAM,GAAY,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBACzC,IAAI,CAAC,iCAAiC,CAAC,MAAM,CAAC,IAAI,MAAM,CAAC,SAAS,KAAK,MAAM,CAAC,SAAS;oBAAE,SAAS;gBAClG,MAAM,QAAQ,GAAG,6BAA6B,CAAC,MAAM,CAAC,CAAC;gBACvD,IAAI,QAAQ,IAAI,OAAO,CAAC,kBAAkB;oBAAE,SAAS;gBACrD,mCAAmC,CAAC,OAAO,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;gBAC3D,OAAO,CAAC,kBAAkB,GAAG,QAAQ,CAAC;YACvC,CAAC;QACF,CAAC;QAAC,MAAM,CAAC;YACR,4DAA4D;QAC7D,CAAC;IACF,CAAC;IACD,MAAM,QAAQ,GAAG,6BAA6B,CAAC,MAAM,CAAC,CAAC;IACvD,IAAI,QAAQ,GAAG,OAAO,CAAC,kBAAkB,EAAE,CAAC;QAC3C,mCAAmC,CAAC,OAAO,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;QAC3D,OAAO,CAAC,kBAAkB,GAAG,QAAQ,CAAC;IACvC,CAAC;IACD,+BAA+B,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;AAAA,CACvD","sourcesContent":["import { createHash } from \"node:crypto\";\nimport {\n\texistsSync,\n\tmkdirSync,\n\treaddirSync,\n\treadFileSync,\n\trenameSync,\n\tstatSync,\n\tunlinkSync,\n\twriteFileSync,\n} from \"node:fs\";\nimport { dirname, join } from \"node:path\";\nimport type {\n\tToolArgumentExecutionOutcome,\n\tToolArgumentTeachState,\n\tToolArgumentValidationTelemetryEvent,\n} from \"@caupulican/pi-ai\";\nimport {\n\tisToolArgumentValidationLogRecord,\n\ttype ToolArgumentValidationLogRecord,\n} from \"./tool-recovery-log-records.ts\";\n\nconst MAX_DETAIL_KEYS = 1_000;\nconst MAX_STATS_FILES = 1_000;\nconst MAX_STATS_FILE_AGE_MS = 90 * 24 * 60 * 60 * 1000;\nconst OTHER_DETAIL_KEY = \"__other__\";\n\nexport interface ToolArgumentValidationTeachEfficacy {\n\trecurrenceBefore: number;\n\trecurrenceAfter: number;\n\trepairedThenSucceeded: number;\n\trepairedThenFailed: number;\n\trepairedThenNotRun: number;\n}\n\nexport interface ToolArgumentValidationStats {\n\tclean: number;\n\trepaired: number;\n\tbounced: number;\n\tfailureModes: Record<string, number>;\n\trepairsApplied: Record<string, number>;\n\ttaught: Record<ToolArgumentTeachState, number>;\n\texecutionOutcome: Record<ToolArgumentExecutionOutcome, number>;\n\tteachEfficacy: Record<string, ToolArgumentValidationTeachEfficacy>;\n}\n\nexport interface ToolArgumentValidationRecord extends ToolArgumentValidationTelemetryEvent {\n\tversion: 1;\n}\n\ninterface PersistedToolRecoveryStats {\n\tversion: 1;\n\tsessionId: string;\n\tlastRecordSequence: number;\n\tstats: ToolArgumentValidationStats;\n}\n\nexport function createEmptyToolArgumentValidationStats(): ToolArgumentValidationStats {\n\treturn {\n\t\tclean: 0,\n\t\trepaired: 0,\n\t\tbounced: 0,\n\t\tfailureModes: {},\n\t\trepairsApplied: {},\n\t\ttaught: { none: 0, note: 0, rule: 0 },\n\t\texecutionOutcome: { not_run: 0, succeeded: 0, failed: 0 },\n\t\tteachEfficacy: {},\n\t};\n}\n\nfunction boundedDetailKey(record: Record<string, unknown>, key: string): string {\n\tif (record[key] !== undefined || Object.keys(record).length < MAX_DETAIL_KEYS - 1) return key;\n\treturn OTHER_DETAIL_KEY;\n}\n\nexport function consumeToolArgumentValidationRecord(\n\tstats: ToolArgumentValidationStats,\n\trecord: ToolArgumentValidationRecord,\n): void {\n\tstats[record.outcome] += 1;\n\tconst taught = record.taught ?? \"none\";\n\tconst executionOutcome = record.executionOutcome ?? \"not_run\";\n\tstats.taught[taught] += 1;\n\tstats.executionOutcome[executionOutcome] += 1;\n\tconst modes = new Set([...record.failureModes, ...record.repairsApplied]);\n\tfor (const mode of record.failureModes) {\n\t\tconst key = boundedDetailKey(stats.failureModes, mode);\n\t\tstats.failureModes[key] = (stats.failureModes[key] ?? 0) + 1;\n\t}\n\tfor (const repair of record.repairsApplied) {\n\t\tconst key = boundedDetailKey(stats.repairsApplied, repair);\n\t\tstats.repairsApplied[key] = (stats.repairsApplied[key] ?? 0) + 1;\n\t}\n\tfor (const mode of modes) {\n\t\tconst key = boundedDetailKey(\n\t\t\tstats.teachEfficacy,\n\t\t\t`${record.provider ?? \"unknown\"}/${record.model ?? \"unknown\"}:${mode}`,\n\t\t);\n\t\tlet efficacy = stats.teachEfficacy[key];\n\t\tif (!efficacy) {\n\t\t\tefficacy = {\n\t\t\t\trecurrenceBefore: 0,\n\t\t\t\trecurrenceAfter: 0,\n\t\t\t\trepairedThenSucceeded: 0,\n\t\t\t\trepairedThenFailed: 0,\n\t\t\t\trepairedThenNotRun: 0,\n\t\t\t};\n\t\t\tstats.teachEfficacy[key] = efficacy;\n\t\t}\n\t\tif (taught === \"none\") efficacy.recurrenceBefore++;\n\t\telse efficacy.recurrenceAfter++;\n\t}\n\tif (record.outcome === \"repaired\") {\n\t\tfor (const repair of record.repairsApplied) {\n\t\t\tconst key = boundedDetailKey(\n\t\t\t\tstats.teachEfficacy,\n\t\t\t\t`${record.provider ?? \"unknown\"}/${record.model ?? \"unknown\"}:${repair}`,\n\t\t\t);\n\t\t\tlet efficacy = stats.teachEfficacy[key];\n\t\t\tif (!efficacy) {\n\t\t\t\tefficacy = {\n\t\t\t\t\trecurrenceBefore: 0,\n\t\t\t\t\trecurrenceAfter: 0,\n\t\t\t\t\trepairedThenSucceeded: 0,\n\t\t\t\t\trepairedThenFailed: 0,\n\t\t\t\t\trepairedThenNotRun: 0,\n\t\t\t\t};\n\t\t\t\tstats.teachEfficacy[key] = efficacy;\n\t\t\t}\n\t\t\tif (executionOutcome === \"succeeded\") efficacy.repairedThenSucceeded++;\n\t\t\tif (executionOutcome === \"failed\") efficacy.repairedThenFailed++;\n\t\t\tif (executionOutcome === \"not_run\") efficacy.repairedThenNotRun++;\n\t\t}\n\t}\n}\n\nexport function getToolRecoveryRecordSequence(\n\trecord: Pick<ToolArgumentValidationLogRecord, \"recordId\" | \"sessionId\">,\n): number {\n\tconst prefix = `${record.sessionId}:`;\n\tif (!record.recordId.startsWith(prefix)) return -1;\n\tconst sequence = Number(record.recordId.slice(prefix.length));\n\treturn Number.isSafeInteger(sequence) && sequence >= 0 ? sequence : -1;\n}\n\nfunction statsPath(eventLogPath: string, sessionId: string): string {\n\tconst id = createHash(\"sha256\").update(sessionId).digest(\"hex\").slice(0, 24);\n\treturn join(dirname(eventLogPath), \"tool-recovery-stats\", `${id}.json`);\n}\n\nfunction isFiniteCount(value: unknown): value is number {\n\treturn typeof value === \"number\" && Number.isFinite(value) && value >= 0;\n}\n\nfunction isCountRecord(value: unknown): value is Record<string, number> {\n\treturn !!value && typeof value === \"object\" && Object.values(value).every(isFiniteCount);\n}\n\nfunction isStatsShape(value: unknown): value is ToolArgumentValidationStats {\n\tif (!value || typeof value !== \"object\") return false;\n\tconst stats = value as Partial<ToolArgumentValidationStats>;\n\tif (!isFiniteCount(stats.clean) || !isFiniteCount(stats.repaired) || !isFiniteCount(stats.bounced)) return false;\n\tif (!isCountRecord(stats.failureModes) || !isCountRecord(stats.repairsApplied)) return false;\n\tif (!isCountRecord(stats.taught) || !isCountRecord(stats.executionOutcome)) return false;\n\tif (!stats.teachEfficacy || typeof stats.teachEfficacy !== \"object\") return false;\n\treturn Object.values(stats.teachEfficacy).every(\n\t\t(value) =>\n\t\t\t!!value &&\n\t\t\tisFiniteCount(value.recurrenceBefore) &&\n\t\t\tisFiniteCount(value.recurrenceAfter) &&\n\t\t\tisFiniteCount(value.repairedThenSucceeded) &&\n\t\t\tisFiniteCount(value.repairedThenFailed) &&\n\t\t\tisFiniteCount(value.repairedThenNotRun),\n\t);\n}\n\nfunction isPersistedStats(value: unknown, sessionId: string): value is PersistedToolRecoveryStats {\n\tif (!value || typeof value !== \"object\") return false;\n\tconst record = value as Partial<PersistedToolRecoveryStats>;\n\treturn (\n\t\trecord.version === 1 &&\n\t\trecord.sessionId === sessionId &&\n\t\tisFiniteCount(record.lastRecordSequence) &&\n\t\tisStatsShape(record.stats)\n\t);\n}\n\nexport function readPersistedToolRecoveryStats(\n\teventLogPath: string,\n\tsessionId: string,\n): PersistedToolRecoveryStats | undefined {\n\tconst path = statsPath(eventLogPath, sessionId);\n\tif (!existsSync(path)) return undefined;\n\ttry {\n\t\tconst parsed: unknown = JSON.parse(readFileSync(path, \"utf-8\"));\n\t\treturn isPersistedStats(parsed, sessionId) ? parsed : undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nfunction pruneStatsFiles(directory: string, currentPath: string): void {\n\tconst now = Date.now();\n\tconst files = readdirSync(directory)\n\t\t.filter((name) => name.endsWith(\".json\"))\n\t\t.map((name) => {\n\t\t\tconst path = join(directory, name);\n\t\t\treturn { path, mtimeMs: statSync(path).mtimeMs };\n\t\t})\n\t\t.sort((left, right) => left.mtimeMs - right.mtimeMs);\n\tlet retained = files.length;\n\tfor (const file of files) {\n\t\tif (file.path === currentPath) continue;\n\t\tif (retained <= MAX_STATS_FILES && now - file.mtimeMs <= MAX_STATS_FILE_AGE_MS) continue;\n\t\ttry {\n\t\t\tunlinkSync(file.path);\n\t\t\tretained--;\n\t\t} catch {}\n\t}\n}\n\nfunction writePersistedToolRecoveryStats(eventLogPath: string, summary: PersistedToolRecoveryStats): void {\n\tconst path = statsPath(eventLogPath, summary.sessionId);\n\tconst directory = dirname(path);\n\tmkdirSync(directory, { recursive: true });\n\tconst isNewSummary = !existsSync(path);\n\tconst temporaryPath = `${path}.${process.pid}.tmp`;\n\twriteFileSync(temporaryPath, JSON.stringify(summary), \"utf-8\");\n\trenameSync(temporaryPath, path);\n\tif (isNewSummary) pruneStatsFiles(directory, path);\n}\n\nexport function updatePersistedToolRecoveryStats(eventLogPath: string, record: ToolArgumentValidationLogRecord): void {\n\tlet summary = readPersistedToolRecoveryStats(eventLogPath, record.sessionId);\n\tif (!summary) {\n\t\tsummary = {\n\t\t\tversion: 1,\n\t\t\tsessionId: record.sessionId,\n\t\t\tlastRecordSequence: -1,\n\t\t\tstats: createEmptyToolArgumentValidationStats(),\n\t\t};\n\t\ttry {\n\t\t\tfor (const line of readFileSync(eventLogPath, \"utf-8\").split(\"\\n\")) {\n\t\t\t\tif (line.trim().length === 0) continue;\n\t\t\t\tconst parsed: unknown = JSON.parse(line);\n\t\t\t\tif (!isToolArgumentValidationLogRecord(parsed) || parsed.sessionId !== record.sessionId) continue;\n\t\t\t\tconst sequence = getToolRecoveryRecordSequence(parsed);\n\t\t\t\tif (sequence <= summary.lastRecordSequence) continue;\n\t\t\t\tconsumeToolArgumentValidationRecord(summary.stats, parsed);\n\t\t\t\tsummary.lastRecordSequence = sequence;\n\t\t\t}\n\t\t} catch {\n\t\t\t// Fall through and at least account for the current record.\n\t\t}\n\t}\n\tconst sequence = getToolRecoveryRecordSequence(record);\n\tif (sequence > summary.lastRecordSequence) {\n\t\tconsumeToolArgumentValidationRecord(summary.stats, record);\n\t\tsummary.lastRecordSequence = sequence;\n\t}\n\twritePersistedToolRecoveryStats(eventLogPath, summary);\n}\n"]}
+{"version":3,"file":"tool-recovery-stats.js","sourceRoot":"","sources":["../../src/core/tool-recovery-stats.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EACN,UAAU,EACV,SAAS,EACT,WAAW,EACX,YAAY,EACZ,UAAU,EACV,QAAQ,EACR,UAAU,EACV,aAAa,GACb,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAM1C,OAAO,EACN,iCAAiC,GAEjC,MAAM,gCAAgC,CAAC;AAExC,MAAM,eAAe,GAAG,KAAK,CAAC;AAC9B,MAAM,eAAe,GAAG,KAAK,CAAC;AAC9B,MAAM,qBAAqB,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;AACvD,MAAM,gBAAgB,GAAG,WAAW,CAAC;AAgCrC,MAAM,UAAU,sCAAsC;IACrD,OAAO;QACN,KAAK,EAAE,CAAC;QACR,QAAQ,EAAE,CAAC;QACX,OAAO,EAAE,CAAC;QACV,YAAY,EAAE,EAAE;QAChB,cAAc,EAAE,EAAE;QAClB,MAAM,EAAE,EAAE,IAAI,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE;QACrC,gBAAgB,EAAE,EAAE,OAAO,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE;QACzD,aAAa,EAAE,EAAE;KACjB,CAAC;AACH,CAAC;AAED,SAAS,gBAAgB,CAAC,MAA+B,EAAE,GAAW;IACrE,IAAI,MAAM,CAAC,GAAG,CAAC,KAAK,SAAS,IAAI,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,MAAM,GAAG,eAAe,GAAG,CAAC;QAAE,OAAO,GAAG,CAAC;IAC9F,OAAO,gBAAgB,CAAC;AACzB,CAAC;AAED,SAAS,gBAAgB,CAAC,KAAkC,EAAE,GAAW;IACxE,MAAM,QAAQ,GAAG,KAAK,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC;IAC1C,IAAI,QAAQ;QAAE,OAAO,QAAQ,CAAC;IAC9B,MAAM,OAAO,GAAwC;QACpD,gBAAgB,EAAE,CAAC;QACnB,eAAe,EAAE,CAAC;QAClB,qBAAqB,EAAE,CAAC;QACxB,kBAAkB,EAAE,CAAC;QACrB,kBAAkB,EAAE,CAAC;KACrB,CAAC;IACF,KAAK,CAAC,aAAa,CAAC,GAAG,CAAC,GAAG,OAAO,CAAC;IACnC,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,mCAAmC,CAClD,KAAkC,EAClC,MAAoC;IAEpC,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IAC3B,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC;IACvC,MAAM,gBAAgB,GAAG,MAAM,CAAC,gBAAgB,IAAI,SAAS,CAAC;IAC9D,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;IAC1B,KAAK,CAAC,gBAAgB,CAAC,gBAAgB,CAAC,IAAI,CAAC,CAAC;IAC9C,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,MAAM,CAAC,YAAY,EAAE,GAAG,MAAM,CAAC,cAAc,CAAC,CAAC,CAAC;IAC1E,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,YAAY,EAAE,CAAC;QACxC,MAAM,GAAG,GAAG,gBAAgB,CAAC,KAAK,CAAC,YAAY,EAAE,IAAI,CAAC,CAAC;QACvD,KAAK,CAAC,YAAY,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;IAC9D,CAAC;IACD,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,cAAc,EAAE,CAAC;QAC5C,MAAM,GAAG,GAAG,gBAAgB,CAAC,KAAK,CAAC,cAAc,EAAE,MAAM,CAAC,CAAC;QAC3D,KAAK,CAAC,cAAc,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,cAAc,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;IAClE,CAAC;IACD,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QAC1B,MAAM,GAAG,GAAG,gBAAgB,CAC3B,KAAK,CAAC,aAAa,EACnB,GAAG,MAAM,CAAC,QAAQ,IAAI,SAAS,IAAI,MAAM,CAAC,KAAK,IAAI,SAAS,IAAI,IAAI,EAAE,CACtE,CAAC;QACF,MAAM,QAAQ,GAAG,gBAAgB,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;QAC9C,IAAI,MAAM,KAAK,MAAM;YAAE,QAAQ,CAAC,gBAAgB,EAAE,CAAC;;YAC9C,QAAQ,CAAC,eAAe,EAAE,CAAC;IACjC,CAAC;IACD,IAAI,MAAM,CAAC,OAAO,KAAK,UAAU,EAAE,CAAC;QACnC,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,cAAc,EAAE,CAAC;YAC5C,MAAM,GAAG,GAAG,gBAAgB,CAC3B,KAAK,CAAC,aAAa,EACnB,GAAG,MAAM,CAAC,QAAQ,IAAI,SAAS,IAAI,MAAM,CAAC,KAAK,IAAI,SAAS,IAAI,MAAM,EAAE,CACxE,CAAC;YACF,MAAM,QAAQ,GAAG,gBAAgB,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;YAC9C,IAAI,gBAAgB,KAAK,WAAW;gBAAE,QAAQ,CAAC,qBAAqB,EAAE,CAAC;YACvE,IAAI,gBAAgB,KAAK,QAAQ;gBAAE,QAAQ,CAAC,kBAAkB,EAAE,CAAC;YACjE,IAAI,gBAAgB,KAAK,SAAS;gBAAE,QAAQ,CAAC,kBAAkB,EAAE,CAAC;QACnE,CAAC;IACF,CAAC;AACF,CAAC;AAED,MAAM,UAAU,6BAA6B,CAC5C,MAAuE;IAEvE,MAAM,MAAM,GAAG,GAAG,MAAM,CAAC,SAAS,GAAG,CAAC;IACtC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,MAAM,CAAC;QAAE,OAAO,CAAC,CAAC,CAAC;IACnD,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;IAC9D,OAAO,MAAM,CAAC,aAAa,CAAC,QAAQ,CAAC,IAAI,QAAQ,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AACxE,CAAC;AAED,SAAS,SAAS,CAAC,YAAoB,EAAE,SAAiB;IACzD,MAAM,EAAE,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IAC7E,OAAO,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,qBAAqB,EAAE,GAAG,EAAE,OAAO,CAAC,CAAC;AACzE,CAAC;AAED,SAAS,aAAa,CAAC,KAAc;IACpC,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC;AAC1E,CAAC;AAED,SAAS,aAAa,CAAC,KAAc;IACpC,OAAO,CAAC,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;AAC1F,CAAC;AAED,SAAS,YAAY,CAAC,KAAc;IACnC,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IACtD,MAAM,KAAK,GAAG,KAA6C,CAAC;IAC5D,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC;QAAE,OAAO,KAAK,CAAC;IACjH,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,cAAc,CAAC;QAAE,OAAO,KAAK,CAAC;IAC7F,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,gBAAgB,CAAC;QAAE,OAAO,KAAK,CAAC;IACzF,IAAI,CAAC,KAAK,CAAC,aAAa,IAAI,OAAO,KAAK,CAAC,aAAa,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IAClF,OAAO,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC,KAAK,CAC9C,CAAC,KAAK,EAAE,EAAE,CACT,CAAC,CAAC,KAAK;QACP,aAAa,CAAC,KAAK,CAAC,gBAAgB,CAAC;QACrC,aAAa,CAAC,KAAK,CAAC,eAAe,CAAC;QACpC,aAAa,CAAC,KAAK,CAAC,qBAAqB,CAAC;QAC1C,aAAa,CAAC,KAAK,CAAC,kBAAkB,CAAC;QACvC,aAAa,CAAC,KAAK,CAAC,kBAAkB,CAAC,CACxC,CAAC;AACH,CAAC;AAED,SAAS,gBAAgB,CAAC,KAAc,EAAE,SAAiB;IAC1D,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IACtD,MAAM,MAAM,GAAG,KAA4C,CAAC;IAC5D,OAAO,CACN,MAAM,CAAC,OAAO,KAAK,CAAC;QACpB,MAAM,CAAC,SAAS,KAAK,SAAS;QAC9B,aAAa,CAAC,MAAM,CAAC,kBAAkB,CAAC;QACxC,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAC1B,CAAC;AACH,CAAC;AAED,MAAM,UAAU,8BAA8B,CAC7C,YAAoB,EACpB,SAAiB;IAEjB,MAAM,IAAI,GAAG,SAAS,CAAC,YAAY,EAAE,SAAS,CAAC,CAAC;IAChD,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,OAAO,SAAS,CAAC;IACxC,IAAI,CAAC;QACJ,MAAM,MAAM,GAAY,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC;QAChE,OAAO,gBAAgB,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC;IACjE,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,SAAS,CAAC;IAClB,CAAC;AACF,CAAC;AAED,SAAS,eAAe,CAAC,SAAiB,EAAE,WAAmB;IAC9D,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACvB,MAAM,KAAK,GAAG,WAAW,CAAC,SAAS,CAAC;SAClC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;SACxC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;QACb,MAAM,IAAI,GAAG,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;QACnC,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,QAAQ,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,CAAC;IAClD,CAAC,CAAC;SACD,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC;IACtD,IAAI,QAAQ,GAAG,KAAK,CAAC,MAAM,CAAC;IAC5B,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QAC1B,IAAI,IAAI,CAAC,IAAI,KAAK,WAAW;YAAE,SAAS;QACxC,IAAI,QAAQ,IAAI,eAAe,IAAI,GAAG,GAAG,IAAI,CAAC,OAAO,IAAI,qBAAqB;YAAE,SAAS;QACzF,IAAI,CAAC;YACJ,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACtB,QAAQ,EAAE,CAAC;QACZ,CAAC;QAAC,MAAM,CAAC,CAAA,CAAC;IACX,CAAC;AACF,CAAC;AAED,SAAS,+BAA+B,CAAC,YAAoB,EAAE,OAAmC;IACjG,MAAM,IAAI,GAAG,SAAS,CAAC,YAAY,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC;IACxD,MAAM,SAAS,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAChC,SAAS,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC1C,MAAM,YAAY,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;IACvC,MAAM,aAAa,GAAG,GAAG,IAAI,IAAI,OAAO,CAAC,GAAG,MAAM,CAAC;IACnD,aAAa,CAAC,aAAa,EAAE,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,EAAE,OAAO,CAAC,CAAC;IAC/D,UAAU,CAAC,aAAa,EAAE,IAAI,CAAC,CAAC;IAChC,IAAI,YAAY;QAAE,eAAe,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;AACpD,CAAC;AAED,MAAM,UAAU,gCAAgC,CAAC,YAAoB,EAAE,MAAuC;IAC7G,IAAI,OAAO,GAAG,8BAA8B,CAAC,YAAY,EAAE,MAAM,CAAC,SAAS,CAAC,CAAC;IAC7E,IAAI,CAAC,OAAO,EAAE,CAAC;QACd,OAAO,GAAG;YACT,OAAO,EAAE,CAAC;YACV,SAAS,EAAE,MAAM,CAAC,SAAS;YAC3B,kBAAkB,EAAE,CAAC,CAAC;YACtB,KAAK,EAAE,sCAAsC,EAAE;SAC/C,CAAC;QACF,IAAI,CAAC;YACJ,KAAK,MAAM,IAAI,IAAI,YAAY,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;gBACpE,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC;oBAAE,SAAS;gBACvC,MAAM,MAAM,GAAY,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBACzC,IAAI,CAAC,iCAAiC,CAAC,MAAM,CAAC,IAAI,MAAM,CAAC,SAAS,KAAK,MAAM,CAAC,SAAS;oBAAE,SAAS;gBAClG,MAAM,QAAQ,GAAG,6BAA6B,CAAC,MAAM,CAAC,CAAC;gBACvD,IAAI,QAAQ,IAAI,OAAO,CAAC,kBAAkB;oBAAE,SAAS;gBACrD,mCAAmC,CAAC,OAAO,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;gBAC3D,OAAO,CAAC,kBAAkB,GAAG,QAAQ,CAAC;YACvC,CAAC;QACF,CAAC;QAAC,MAAM,CAAC;YACR,4DAA4D;QAC7D,CAAC;IACF,CAAC;IACD,MAAM,QAAQ,GAAG,6BAA6B,CAAC,MAAM,CAAC,CAAC;IACvD,IAAI,QAAQ,GAAG,OAAO,CAAC,kBAAkB,EAAE,CAAC;QAC3C,mCAAmC,CAAC,OAAO,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;QAC3D,OAAO,CAAC,kBAAkB,GAAG,QAAQ,CAAC;IACvC,CAAC;IACD,+BAA+B,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;AACxD,CAAC","sourcesContent":["import { createHash } from \"node:crypto\";\nimport {\n\texistsSync,\n\tmkdirSync,\n\treaddirSync,\n\treadFileSync,\n\trenameSync,\n\tstatSync,\n\tunlinkSync,\n\twriteFileSync,\n} from \"node:fs\";\nimport { dirname, join } from \"node:path\";\nimport type {\n\tToolArgumentExecutionOutcome,\n\tToolArgumentTeachState,\n\tToolArgumentValidationTelemetryEvent,\n} from \"@caupulican/pi-ai\";\nimport {\n\tisToolArgumentValidationLogRecord,\n\ttype ToolArgumentValidationLogRecord,\n} from \"./tool-recovery-log-records.ts\";\n\nconst MAX_DETAIL_KEYS = 1_000;\nconst MAX_STATS_FILES = 1_000;\nconst MAX_STATS_FILE_AGE_MS = 90 * 24 * 60 * 60 * 1000;\nconst OTHER_DETAIL_KEY = \"__other__\";\n\nexport interface ToolArgumentValidationTeachEfficacy {\n\trecurrenceBefore: number;\n\trecurrenceAfter: number;\n\trepairedThenSucceeded: number;\n\trepairedThenFailed: number;\n\trepairedThenNotRun: number;\n}\n\nexport interface ToolArgumentValidationStats {\n\tclean: number;\n\trepaired: number;\n\tbounced: number;\n\tfailureModes: Record<string, number>;\n\trepairsApplied: Record<string, number>;\n\ttaught: Record<ToolArgumentTeachState, number>;\n\texecutionOutcome: Record<ToolArgumentExecutionOutcome, number>;\n\tteachEfficacy: Record<string, ToolArgumentValidationTeachEfficacy>;\n}\n\nexport interface ToolArgumentValidationRecord extends ToolArgumentValidationTelemetryEvent {\n\tversion: 1;\n}\n\ninterface PersistedToolRecoveryStats {\n\tversion: 1;\n\tsessionId: string;\n\tlastRecordSequence: number;\n\tstats: ToolArgumentValidationStats;\n}\n\nexport function createEmptyToolArgumentValidationStats(): ToolArgumentValidationStats {\n\treturn {\n\t\tclean: 0,\n\t\trepaired: 0,\n\t\tbounced: 0,\n\t\tfailureModes: {},\n\t\trepairsApplied: {},\n\t\ttaught: { none: 0, note: 0, rule: 0 },\n\t\texecutionOutcome: { not_run: 0, succeeded: 0, failed: 0 },\n\t\tteachEfficacy: {},\n\t};\n}\n\nfunction boundedDetailKey(record: Record<string, unknown>, key: string): string {\n\tif (record[key] !== undefined || Object.keys(record).length < MAX_DETAIL_KEYS - 1) return key;\n\treturn OTHER_DETAIL_KEY;\n}\n\nfunction getTeachEfficacy(stats: ToolArgumentValidationStats, key: string): ToolArgumentValidationTeachEfficacy {\n\tconst existing = stats.teachEfficacy[key];\n\tif (existing) return existing;\n\tconst created: ToolArgumentValidationTeachEfficacy = {\n\t\trecurrenceBefore: 0,\n\t\trecurrenceAfter: 0,\n\t\trepairedThenSucceeded: 0,\n\t\trepairedThenFailed: 0,\n\t\trepairedThenNotRun: 0,\n\t};\n\tstats.teachEfficacy[key] = created;\n\treturn created;\n}\n\nexport function consumeToolArgumentValidationRecord(\n\tstats: ToolArgumentValidationStats,\n\trecord: ToolArgumentValidationRecord,\n): void {\n\tstats[record.outcome] += 1;\n\tconst taught = record.taught ?? \"none\";\n\tconst executionOutcome = record.executionOutcome ?? \"not_run\";\n\tstats.taught[taught] += 1;\n\tstats.executionOutcome[executionOutcome] += 1;\n\tconst modes = new Set([...record.failureModes, ...record.repairsApplied]);\n\tfor (const mode of record.failureModes) {\n\t\tconst key = boundedDetailKey(stats.failureModes, mode);\n\t\tstats.failureModes[key] = (stats.failureModes[key] ?? 0) + 1;\n\t}\n\tfor (const repair of record.repairsApplied) {\n\t\tconst key = boundedDetailKey(stats.repairsApplied, repair);\n\t\tstats.repairsApplied[key] = (stats.repairsApplied[key] ?? 0) + 1;\n\t}\n\tfor (const mode of modes) {\n\t\tconst key = boundedDetailKey(\n\t\t\tstats.teachEfficacy,\n\t\t\t`${record.provider ?? \"unknown\"}/${record.model ?? \"unknown\"}:${mode}`,\n\t\t);\n\t\tconst efficacy = getTeachEfficacy(stats, key);\n\t\tif (taught === \"none\") efficacy.recurrenceBefore++;\n\t\telse efficacy.recurrenceAfter++;\n\t}\n\tif (record.outcome === \"repaired\") {\n\t\tfor (const repair of record.repairsApplied) {\n\t\t\tconst key = boundedDetailKey(\n\t\t\t\tstats.teachEfficacy,\n\t\t\t\t`${record.provider ?? \"unknown\"}/${record.model ?? \"unknown\"}:${repair}`,\n\t\t\t);\n\t\t\tconst efficacy = getTeachEfficacy(stats, key);\n\t\t\tif (executionOutcome === \"succeeded\") efficacy.repairedThenSucceeded++;\n\t\t\tif (executionOutcome === \"failed\") efficacy.repairedThenFailed++;\n\t\t\tif (executionOutcome === \"not_run\") efficacy.repairedThenNotRun++;\n\t\t}\n\t}\n}\n\nexport function getToolRecoveryRecordSequence(\n\trecord: Pick<ToolArgumentValidationLogRecord, \"recordId\" | \"sessionId\">,\n): number {\n\tconst prefix = `${record.sessionId}:`;\n\tif (!record.recordId.startsWith(prefix)) return -1;\n\tconst sequence = Number(record.recordId.slice(prefix.length));\n\treturn Number.isSafeInteger(sequence) && sequence >= 0 ? sequence : -1;\n}\n\nfunction statsPath(eventLogPath: string, sessionId: string): string {\n\tconst id = createHash(\"sha256\").update(sessionId).digest(\"hex\").slice(0, 24);\n\treturn join(dirname(eventLogPath), \"tool-recovery-stats\", `${id}.json`);\n}\n\nfunction isFiniteCount(value: unknown): value is number {\n\treturn typeof value === \"number\" && Number.isFinite(value) && value >= 0;\n}\n\nfunction isCountRecord(value: unknown): value is Record<string, number> {\n\treturn !!value && typeof value === \"object\" && Object.values(value).every(isFiniteCount);\n}\n\nfunction isStatsShape(value: unknown): value is ToolArgumentValidationStats {\n\tif (!value || typeof value !== \"object\") return false;\n\tconst stats = value as Partial<ToolArgumentValidationStats>;\n\tif (!isFiniteCount(stats.clean) || !isFiniteCount(stats.repaired) || !isFiniteCount(stats.bounced)) return false;\n\tif (!isCountRecord(stats.failureModes) || !isCountRecord(stats.repairsApplied)) return false;\n\tif (!isCountRecord(stats.taught) || !isCountRecord(stats.executionOutcome)) return false;\n\tif (!stats.teachEfficacy || typeof stats.teachEfficacy !== \"object\") return false;\n\treturn Object.values(stats.teachEfficacy).every(\n\t\t(value) =>\n\t\t\t!!value &&\n\t\t\tisFiniteCount(value.recurrenceBefore) &&\n\t\t\tisFiniteCount(value.recurrenceAfter) &&\n\t\t\tisFiniteCount(value.repairedThenSucceeded) &&\n\t\t\tisFiniteCount(value.repairedThenFailed) &&\n\t\t\tisFiniteCount(value.repairedThenNotRun),\n\t);\n}\n\nfunction isPersistedStats(value: unknown, sessionId: string): value is PersistedToolRecoveryStats {\n\tif (!value || typeof value !== \"object\") return false;\n\tconst record = value as Partial<PersistedToolRecoveryStats>;\n\treturn (\n\t\trecord.version === 1 &&\n\t\trecord.sessionId === sessionId &&\n\t\tisFiniteCount(record.lastRecordSequence) &&\n\t\tisStatsShape(record.stats)\n\t);\n}\n\nexport function readPersistedToolRecoveryStats(\n\teventLogPath: string,\n\tsessionId: string,\n): PersistedToolRecoveryStats | undefined {\n\tconst path = statsPath(eventLogPath, sessionId);\n\tif (!existsSync(path)) return undefined;\n\ttry {\n\t\tconst parsed: unknown = JSON.parse(readFileSync(path, \"utf-8\"));\n\t\treturn isPersistedStats(parsed, sessionId) ? parsed : undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nfunction pruneStatsFiles(directory: string, currentPath: string): void {\n\tconst now = Date.now();\n\tconst files = readdirSync(directory)\n\t\t.filter((name) => name.endsWith(\".json\"))\n\t\t.map((name) => {\n\t\t\tconst path = join(directory, name);\n\t\t\treturn { path, mtimeMs: statSync(path).mtimeMs };\n\t\t})\n\t\t.sort((left, right) => left.mtimeMs - right.mtimeMs);\n\tlet retained = files.length;\n\tfor (const file of files) {\n\t\tif (file.path === currentPath) continue;\n\t\tif (retained <= MAX_STATS_FILES && now - file.mtimeMs <= MAX_STATS_FILE_AGE_MS) continue;\n\t\ttry {\n\t\t\tunlinkSync(file.path);\n\t\t\tretained--;\n\t\t} catch {}\n\t}\n}\n\nfunction writePersistedToolRecoveryStats(eventLogPath: string, summary: PersistedToolRecoveryStats): void {\n\tconst path = statsPath(eventLogPath, summary.sessionId);\n\tconst directory = dirname(path);\n\tmkdirSync(directory, { recursive: true });\n\tconst isNewSummary = !existsSync(path);\n\tconst temporaryPath = `${path}.${process.pid}.tmp`;\n\twriteFileSync(temporaryPath, JSON.stringify(summary), \"utf-8\");\n\trenameSync(temporaryPath, path);\n\tif (isNewSummary) pruneStatsFiles(directory, path);\n}\n\nexport function updatePersistedToolRecoveryStats(eventLogPath: string, record: ToolArgumentValidationLogRecord): void {\n\tlet summary = readPersistedToolRecoveryStats(eventLogPath, record.sessionId);\n\tif (!summary) {\n\t\tsummary = {\n\t\t\tversion: 1,\n\t\t\tsessionId: record.sessionId,\n\t\t\tlastRecordSequence: -1,\n\t\t\tstats: createEmptyToolArgumentValidationStats(),\n\t\t};\n\t\ttry {\n\t\t\tfor (const line of readFileSync(eventLogPath, \"utf-8\").split(\"\\n\")) {\n\t\t\t\tif (line.trim().length === 0) continue;\n\t\t\t\tconst parsed: unknown = JSON.parse(line);\n\t\t\t\tif (!isToolArgumentValidationLogRecord(parsed) || parsed.sessionId !== record.sessionId) continue;\n\t\t\t\tconst sequence = getToolRecoveryRecordSequence(parsed);\n\t\t\t\tif (sequence <= summary.lastRecordSequence) continue;\n\t\t\t\tconsumeToolArgumentValidationRecord(summary.stats, parsed);\n\t\t\t\tsummary.lastRecordSequence = sequence;\n\t\t\t}\n\t\t} catch {\n\t\t\t// Fall through and at least account for the current record.\n\t\t}\n\t}\n\tconst sequence = getToolRecoveryRecordSequence(record);\n\tif (sequence > summary.lastRecordSequence) {\n\t\tconsumeToolArgumentValidationRecord(summary.stats, record);\n\t\tsummary.lastRecordSequence = sequence;\n\t}\n\twritePersistedToolRecoveryStats(eventLogPath, summary);\n}\n"]}

package/dist/core/tool-repair-health.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tool-repair-health.d.ts","sourceRoot":"","sources":["../../src/core/tool-repair-health.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,oBAAoB,EAAyB,MAAM,8BAA8B,CAAC;AAChG,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAoBzE,wBAAgB,4BAA4B,CAC3C,KAAK,EAAE,oBAAoB,EAC3B,GAAG,GAAE,IAAiB,EACtB,WAAW,CAAC,EAAE,uBAAuB,GACnC,MAAM,CAwDR","sourcesContent":["import chalk from \"chalk\";\nimport type { ModelAdaptationStore, StoredModelAdaptation } from \"./models/adaptation-store.ts\";\nimport type { ToolRecoveryLoggerStats } from \"./tool-recovery-logger.ts\";\n\nfunction formatAgeDays(iso: string, now: Date): string {\n\tconst ms = Date.parse(iso);\n\tif (!Number.isFinite(ms)) return \"age unknown\";\n\tconst days = Math.max(0, Math.floor((now.getTime() - ms) / (24 * 60 * 60 * 1000)));\n\treturn `${days}d ago`;\n}\n\nfunction sortProfiles(profiles: StoredModelAdaptation[]): StoredModelAdaptation[] {\n\treturn [...profiles].sort((left, right) => left.model.localeCompare(right.model));\n}\n\nfunction formatLoggerStats(stats: ToolRecoveryLoggerStats | undefined): string[] {\n\tif (!stats) return [];\n\treturn [\n\t\t`recovery logging: ${stats.enabled ? \"enabled\" : \"disabled\"} queued=${stats.queued} inFlight=${stats.inFlight} dropped=${stats.dropped} failures=${stats.failures} workerStarts=${stats.workerStarts} crashes=${stats.workerCrashes} respawns=${stats.respawns}`,\n\t];\n}\n\nexport function formatToolRepairHealthReport(\n\tstore: ModelAdaptationStore,\n\tnow: Date = new Date(),\n\tloggerStats?: ToolRecoveryLoggerStats,\n): string {\n\tconst profiles = sortProfiles(store.getForHost());\n\tif (profiles.length === 0) {\n\t\tconst loggerLines = formatLoggerStats(loggerStats);\n\t\treturn [\"Tool repair health: no model adaptation records for this host.\", ...loggerLines].join(\"\\n\");\n\t}\n\n\tconst lines = [chalk.bold(\"Tool repair health\"), ...formatLoggerStats(loggerStats)];\n\tfor (const entry of profiles) {\n\t\tlines.push(`${entry.model}`);\n\t\tconst toolProbe = entry.profile.toolProbe;\n\t\tif (!toolProbe) {\n\t\t\tlines.push(\"  tool probe: none\");\n\t\t} else {\n\t\t\tconst variant = toolProbe.variant ? ` (${toolProbe.variant})` : \"\";\n\t\t\tconst nativeGrade = toolProbe.nativeGrade ? ` native=${toolProbe.nativeGrade}` : \"\";\n\t\t\tlines.push(\n\t\t\t\t`  tool probe: v${toolProbe.version} ${toolProbe.status}${variant}${nativeGrade} ${formatAgeDays(toolProbe.probedAt, now)}`,\n\t\t\t);\n\t\t\tif (toolProbe.diagnostic) lines.push(`  probe diagnostic: ${toolProbe.diagnostic}`);\n\t\t}\n\t\tconst protocol = entry.profile.protocol;\n\t\tif (!protocol) {\n\t\t\tlines.push(\"  protocol: none\");\n\t\t} else if (protocol.status === \"failed\") {\n\t\t\tlines.push(`  protocol: v${protocol.version} failed ${formatAgeDays(protocol.attemptedAt, now)}`);\n\t\t\tlines.push(`  variants tried: ${protocol.variantsTried.join(\", \")}`);\n\t\t\tlines.push(`  reset: /toolprotocol-reset ${entry.model}`);\n\t\t} else {\n\t\t\tlines.push(\n\t\t\t\t`  protocol: v${protocol.version} ${protocol.variant} calibrated ${formatAgeDays(protocol.calibratedAt, now)}`,\n\t\t\t);\n\t\t}\n\t\tif (entry.profile.rules.length === 0) {\n\t\t\tlines.push(\"  rules: none\");\n\t\t} else {\n\t\t\tlines.push(\"  rules:\");\n\t\t\tfor (const rule of [...entry.profile.rules].sort((left, right) => left.mode.localeCompare(right.mode))) {\n\t\t\t\tlines.push(`    - ${rule.mode} (${formatAgeDays(rule.lastFiredAt, now)}): ${rule.text}`);\n\t\t\t}\n\t\t}\n\t\tconst teachEntries = Object.entries(entry.profile.teachStats).sort(([left], [right]) =>\n\t\t\tleft.localeCompare(right),\n\t\t);\n\t\tif (teachEntries.length === 0) {\n\t\t\tlines.push(\"  teach stats: none\");\n\t\t} else {\n\t\t\tlines.push(\"  teach stats:\");\n\t\t\tfor (const [mode, stats] of teachEntries) {\n\t\t\t\tlines.push(\n\t\t\t\t\t`    - ${mode}: taught=${stats.taught} before=${stats.recurrenceBefore} after=${stats.recurrenceAfter}`,\n\t\t\t\t);\n\t\t\t}\n\t\t}\n\t}\n\treturn lines.join(\"\\n\");\n}\n"]}
+{"version":3,"file":"tool-repair-health.d.ts","sourceRoot":"","sources":["../../src/core/tool-repair-health.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,oBAAoB,EAAyB,MAAM,8BAA8B,CAAC;AAChG,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAoBzE,wBAAgB,4BAA4B,CAC3C,KAAK,EAAE,oBAAoB,EAC3B,GAAG,GAAE,IAAiB,EACtB,WAAW,CAAC,EAAE,uBAAuB,GACnC,MAAM,CAwDR"}

package/dist/core/tool-repair-health.js.map

@@ -1 +1 @@
-{"version":3,"file":"tool-repair-health.js","sourceRoot":"","sources":["../../src/core/tool-repair-health.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,OAAO,CAAC;AAI1B,SAAS,aAAa,CAAC,GAAW,EAAE,GAAS,EAAU;IACtD,MAAM,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC3B,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;QAAE,OAAO,aAAa,CAAC;IAC/C,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,GAAG,CAAC,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;IACnF,OAAO,GAAG,IAAI,OAAO,CAAC;AAAA,CACtB;AAED,SAAS,YAAY,CAAC,QAAiC,EAA2B;IACjF,OAAO,CAAC,GAAG,QAAQ,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC;AAAA,CAClF;AAED,SAAS,iBAAiB,CAAC,KAA0C,EAAY;IAChF,IAAI,CAAC,KAAK;QAAE,OAAO,EAAE,CAAC;IACtB,OAAO;QACN,qBAAqB,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,WAAW,KAAK,CAAC,MAAM,aAAa,KAAK,CAAC,QAAQ,YAAY,KAAK,CAAC,OAAO,aAAa,KAAK,CAAC,QAAQ,iBAAiB,KAAK,CAAC,YAAY,YAAY,KAAK,CAAC,aAAa,aAAa,KAAK,CAAC,QAAQ,EAAE;KAChQ,CAAC;AAAA,CACF;AAED,MAAM,UAAU,4BAA4B,CAC3C,KAA2B,EAC3B,GAAG,GAAS,IAAI,IAAI,EAAE,EACtB,WAAqC,EAC5B;IACT,MAAM,QAAQ,GAAG,YAAY,CAAC,KAAK,CAAC,UAAU,EAAE,CAAC,CAAC;IAClD,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC3B,MAAM,WAAW,GAAG,iBAAiB,CAAC,WAAW,CAAC,CAAC;QACnD,OAAO,CAAC,gEAAgE,EAAE,GAAG,WAAW,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACtG,CAAC;IAED,MAAM,KAAK,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,EAAE,GAAG,iBAAiB,CAAC,WAAW,CAAC,CAAC,CAAC;IACpF,KAAK,MAAM,KAAK,IAAI,QAAQ,EAAE,CAAC;QAC9B,KAAK,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC;QAC7B,MAAM,SAAS,GAAG,KAAK,CAAC,OAAO,CAAC,SAAS,CAAC;QAC1C,IAAI,CAAC,SAAS,EAAE,CAAC;YAChB,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;QAClC,CAAC;aAAM,CAAC;YACP,MAAM,OAAO,GAAG,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,SAAS,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;YACnE,MAAM,WAAW,GAAG,SAAS,CAAC,WAAW,CAAC,CAAC,CAAC,WAAW,SAAS,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACpF,KAAK,CAAC,IAAI,CACT,kBAAkB,SAAS,CAAC,OAAO,IAAI,SAAS,CAAC,MAAM,GAAG,OAAO,GAAG,WAAW,IAAI,aAAa,CAAC,SAAS,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAC3H,CAAC;YACF,IAAI,SAAS,CAAC,UAAU;gBAAE,KAAK,CAAC,IAAI,CAAC,uBAAuB,SAAS,CAAC,UAAU,EAAE,CAAC,CAAC;QACrF,CAAC;QACD,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC;QACxC,IAAI,CAAC,QAAQ,EAAE,CAAC;YACf,KAAK,CAAC,IAAI,CAAC,kBAAkB,CAAC,CAAC;QAChC,CAAC;aAAM,IAAI,QAAQ,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YACzC,KAAK,CAAC,IAAI,CAAC,gBAAgB,QAAQ,CAAC,OAAO,WAAW,aAAa,CAAC,QAAQ,CAAC,WAAW,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;YAClG,KAAK,CAAC,IAAI,CAAC,qBAAqB,QAAQ,CAAC,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACrE,KAAK,CAAC,IAAI,CAAC,gCAAgC,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC;QAC3D,CAAC;aAAM,CAAC;YACP,KAAK,CAAC,IAAI,CACT,gBAAgB,QAAQ,CAAC,OAAO,IAAI,QAAQ,CAAC,OAAO,eAAe,aAAa,CAAC,QAAQ,CAAC,YAAY,EAAE,GAAG,CAAC,EAAE,CAC9G,CAAC;QACH,CAAC;QACD,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACtC,KAAK,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;QAC7B,CAAC;aAAM,CAAC;YACP,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;YACvB,KAAK,MAAM,IAAI,IAAI,CAAC,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;gBACxG,KAAK,CAAC,IAAI,CAAC,SAAS,IAAI,CAAC,IAAI,KAAK,aAAa,CAAC,IAAI,CAAC,WAAW,EAAE,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;YAC1F,CAAC;QACF,CAAC;QACD,MAAM,YAAY,GAAG,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,EAAE,CACtF,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,CACzB,CAAC;QACF,IAAI,YAAY,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC/B,KAAK,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;QACnC,CAAC;aAAM,CAAC;YACP,KAAK,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;YAC7B,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,YAAY,EAAE,CAAC;gBAC1C,KAAK,CAAC,IAAI,CACT,SAAS,IAAI,YAAY,KAAK,CAAC,MAAM,WAAW,KAAK,CAAC,gBAAgB,UAAU,KAAK,CAAC,eAAe,EAAE,CACvG,CAAC;YACH,CAAC;QACF,CAAC;IACF,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAAA,CACxB","sourcesContent":["import chalk from \"chalk\";\nimport type { ModelAdaptationStore, StoredModelAdaptation } from \"./models/adaptation-store.ts\";\nimport type { ToolRecoveryLoggerStats } from \"./tool-recovery-logger.ts\";\n\nfunction formatAgeDays(iso: string, now: Date): string {\n\tconst ms = Date.parse(iso);\n\tif (!Number.isFinite(ms)) return \"age unknown\";\n\tconst days = Math.max(0, Math.floor((now.getTime() - ms) / (24 * 60 * 60 * 1000)));\n\treturn `${days}d ago`;\n}\n\nfunction sortProfiles(profiles: StoredModelAdaptation[]): StoredModelAdaptation[] {\n\treturn [...profiles].sort((left, right) => left.model.localeCompare(right.model));\n}\n\nfunction formatLoggerStats(stats: ToolRecoveryLoggerStats | undefined): string[] {\n\tif (!stats) return [];\n\treturn [\n\t\t`recovery logging: ${stats.enabled ? \"enabled\" : \"disabled\"} queued=${stats.queued} inFlight=${stats.inFlight} dropped=${stats.dropped} failures=${stats.failures} workerStarts=${stats.workerStarts} crashes=${stats.workerCrashes} respawns=${stats.respawns}`,\n\t];\n}\n\nexport function formatToolRepairHealthReport(\n\tstore: ModelAdaptationStore,\n\tnow: Date = new Date(),\n\tloggerStats?: ToolRecoveryLoggerStats,\n): string {\n\tconst profiles = sortProfiles(store.getForHost());\n\tif (profiles.length === 0) {\n\t\tconst loggerLines = formatLoggerStats(loggerStats);\n\t\treturn [\"Tool repair health: no model adaptation records for this host.\", ...loggerLines].join(\"\\n\");\n\t}\n\n\tconst lines = [chalk.bold(\"Tool repair health\"), ...formatLoggerStats(loggerStats)];\n\tfor (const entry of profiles) {\n\t\tlines.push(`${entry.model}`);\n\t\tconst toolProbe = entry.profile.toolProbe;\n\t\tif (!toolProbe) {\n\t\t\tlines.push(\"  tool probe: none\");\n\t\t} else {\n\t\t\tconst variant = toolProbe.variant ? ` (${toolProbe.variant})` : \"\";\n\t\t\tconst nativeGrade = toolProbe.nativeGrade ? ` native=${toolProbe.nativeGrade}` : \"\";\n\t\t\tlines.push(\n\t\t\t\t`  tool probe: v${toolProbe.version} ${toolProbe.status}${variant}${nativeGrade} ${formatAgeDays(toolProbe.probedAt, now)}`,\n\t\t\t);\n\t\t\tif (toolProbe.diagnostic) lines.push(`  probe diagnostic: ${toolProbe.diagnostic}`);\n\t\t}\n\t\tconst protocol = entry.profile.protocol;\n\t\tif (!protocol) {\n\t\t\tlines.push(\"  protocol: none\");\n\t\t} else if (protocol.status === \"failed\") {\n\t\t\tlines.push(`  protocol: v${protocol.version} failed ${formatAgeDays(protocol.attemptedAt, now)}`);\n\t\t\tlines.push(`  variants tried: ${protocol.variantsTried.join(\", \")}`);\n\t\t\tlines.push(`  reset: /toolprotocol-reset ${entry.model}`);\n\t\t} else {\n\t\t\tlines.push(\n\t\t\t\t`  protocol: v${protocol.version} ${protocol.variant} calibrated ${formatAgeDays(protocol.calibratedAt, now)}`,\n\t\t\t);\n\t\t}\n\t\tif (entry.profile.rules.length === 0) {\n\t\t\tlines.push(\"  rules: none\");\n\t\t} else {\n\t\t\tlines.push(\"  rules:\");\n\t\t\tfor (const rule of [...entry.profile.rules].sort((left, right) => left.mode.localeCompare(right.mode))) {\n\t\t\t\tlines.push(`    - ${rule.mode} (${formatAgeDays(rule.lastFiredAt, now)}): ${rule.text}`);\n\t\t\t}\n\t\t}\n\t\tconst teachEntries = Object.entries(entry.profile.teachStats).sort(([left], [right]) =>\n\t\t\tleft.localeCompare(right),\n\t\t);\n\t\tif (teachEntries.length === 0) {\n\t\t\tlines.push(\"  teach stats: none\");\n\t\t} else {\n\t\t\tlines.push(\"  teach stats:\");\n\t\t\tfor (const [mode, stats] of teachEntries) {\n\t\t\t\tlines.push(\n\t\t\t\t\t`    - ${mode}: taught=${stats.taught} before=${stats.recurrenceBefore} after=${stats.recurrenceAfter}`,\n\t\t\t\t);\n\t\t\t}\n\t\t}\n\t}\n\treturn lines.join(\"\\n\");\n}\n"]}
+{"version":3,"file":"tool-repair-health.js","sourceRoot":"","sources":["../../src/core/tool-repair-health.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,OAAO,CAAC;AAI1B,SAAS,aAAa,CAAC,GAAW,EAAE,GAAS;IAC5C,MAAM,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC3B,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;QAAE,OAAO,aAAa,CAAC;IAC/C,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,GAAG,CAAC,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;IACnF,OAAO,GAAG,IAAI,OAAO,CAAC;AACvB,CAAC;AAED,SAAS,YAAY,CAAC,QAAiC;IACtD,OAAO,CAAC,GAAG,QAAQ,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC;AACnF,CAAC;AAED,SAAS,iBAAiB,CAAC,KAA0C;IACpE,IAAI,CAAC,KAAK;QAAE,OAAO,EAAE,CAAC;IACtB,OAAO;QACN,qBAAqB,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,WAAW,KAAK,CAAC,MAAM,aAAa,KAAK,CAAC,QAAQ,YAAY,KAAK,CAAC,OAAO,aAAa,KAAK,CAAC,QAAQ,iBAAiB,KAAK,CAAC,YAAY,YAAY,KAAK,CAAC,aAAa,aAAa,KAAK,CAAC,QAAQ,EAAE;KAChQ,CAAC;AACH,CAAC;AAED,MAAM,UAAU,4BAA4B,CAC3C,KAA2B,EAC3B,GAAG,GAAS,IAAI,IAAI,EAAE,EACtB,WAAqC;IAErC,MAAM,QAAQ,GAAG,YAAY,CAAC,KAAK,CAAC,UAAU,EAAE,CAAC,CAAC;IAClD,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC3B,MAAM,WAAW,GAAG,iBAAiB,CAAC,WAAW,CAAC,CAAC;QACnD,OAAO,CAAC,gEAAgE,EAAE,GAAG,WAAW,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACtG,CAAC;IAED,MAAM,KAAK,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,EAAE,GAAG,iBAAiB,CAAC,WAAW,CAAC,CAAC,CAAC;IACpF,KAAK,MAAM,KAAK,IAAI,QAAQ,EAAE,CAAC;QAC9B,KAAK,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC;QAC7B,MAAM,SAAS,GAAG,KAAK,CAAC,OAAO,CAAC,SAAS,CAAC;QAC1C,IAAI,CAAC,SAAS,EAAE,CAAC;YAChB,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;QAClC,CAAC;aAAM,CAAC;YACP,MAAM,OAAO,GAAG,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,SAAS,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;YACnE,MAAM,WAAW,GAAG,SAAS,CAAC,WAAW,CAAC,CAAC,CAAC,WAAW,SAAS,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACpF,KAAK,CAAC,IAAI,CACT,kBAAkB,SAAS,CAAC,OAAO,IAAI,SAAS,CAAC,MAAM,GAAG,OAAO,GAAG,WAAW,IAAI,aAAa,CAAC,SAAS,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAC3H,CAAC;YACF,IAAI,SAAS,CAAC,UAAU;gBAAE,KAAK,CAAC,IAAI,CAAC,uBAAuB,SAAS,CAAC,UAAU,EAAE,CAAC,CAAC;QACrF,CAAC;QACD,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC;QACxC,IAAI,CAAC,QAAQ,EAAE,CAAC;YACf,KAAK,CAAC,IAAI,CAAC,kBAAkB,CAAC,CAAC;QAChC,CAAC;aAAM,IAAI,QAAQ,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YACzC,KAAK,CAAC,IAAI,CAAC,gBAAgB,QAAQ,CAAC,OAAO,WAAW,aAAa,CAAC,QAAQ,CAAC,WAAW,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;YAClG,KAAK,CAAC,IAAI,CAAC,qBAAqB,QAAQ,CAAC,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACrE,KAAK,CAAC,IAAI,CAAC,gCAAgC,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC;QAC3D,CAAC;aAAM,CAAC;YACP,KAAK,CAAC,IAAI,CACT,gBAAgB,QAAQ,CAAC,OAAO,IAAI,QAAQ,CAAC,OAAO,eAAe,aAAa,CAAC,QAAQ,CAAC,YAAY,EAAE,GAAG,CAAC,EAAE,CAC9G,CAAC;QACH,CAAC;QACD,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACtC,KAAK,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;QAC7B,CAAC;aAAM,CAAC;YACP,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;YACvB,KAAK,MAAM,IAAI,IAAI,CAAC,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;gBACxG,KAAK,CAAC,IAAI,CAAC,SAAS,IAAI,CAAC,IAAI,KAAK,aAAa,CAAC,IAAI,CAAC,WAAW,EAAE,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;YAC1F,CAAC;QACF,CAAC;QACD,MAAM,YAAY,GAAG,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,EAAE,CACtF,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,CACzB,CAAC;QACF,IAAI,YAAY,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC/B,KAAK,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;QACnC,CAAC;aAAM,CAAC;YACP,KAAK,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;YAC7B,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,YAAY,EAAE,CAAC;gBAC1C,KAAK,CAAC,IAAI,CACT,SAAS,IAAI,YAAY,KAAK,CAAC,MAAM,WAAW,KAAK,CAAC,gBAAgB,UAAU,KAAK,CAAC,eAAe,EAAE,CACvG,CAAC;YACH,CAAC;QACF,CAAC;IACF,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACzB,CAAC","sourcesContent":["import chalk from \"chalk\";\nimport type { ModelAdaptationStore, StoredModelAdaptation } from \"./models/adaptation-store.ts\";\nimport type { ToolRecoveryLoggerStats } from \"./tool-recovery-logger.ts\";\n\nfunction formatAgeDays(iso: string, now: Date): string {\n\tconst ms = Date.parse(iso);\n\tif (!Number.isFinite(ms)) return \"age unknown\";\n\tconst days = Math.max(0, Math.floor((now.getTime() - ms) / (24 * 60 * 60 * 1000)));\n\treturn `${days}d ago`;\n}\n\nfunction sortProfiles(profiles: StoredModelAdaptation[]): StoredModelAdaptation[] {\n\treturn [...profiles].sort((left, right) => left.model.localeCompare(right.model));\n}\n\nfunction formatLoggerStats(stats: ToolRecoveryLoggerStats | undefined): string[] {\n\tif (!stats) return [];\n\treturn [\n\t\t`recovery logging: ${stats.enabled ? \"enabled\" : \"disabled\"} queued=${stats.queued} inFlight=${stats.inFlight} dropped=${stats.dropped} failures=${stats.failures} workerStarts=${stats.workerStarts} crashes=${stats.workerCrashes} respawns=${stats.respawns}`,\n\t];\n}\n\nexport function formatToolRepairHealthReport(\n\tstore: ModelAdaptationStore,\n\tnow: Date = new Date(),\n\tloggerStats?: ToolRecoveryLoggerStats,\n): string {\n\tconst profiles = sortProfiles(store.getForHost());\n\tif (profiles.length === 0) {\n\t\tconst loggerLines = formatLoggerStats(loggerStats);\n\t\treturn [\"Tool repair health: no model adaptation records for this host.\", ...loggerLines].join(\"\\n\");\n\t}\n\n\tconst lines = [chalk.bold(\"Tool repair health\"), ...formatLoggerStats(loggerStats)];\n\tfor (const entry of profiles) {\n\t\tlines.push(`${entry.model}`);\n\t\tconst toolProbe = entry.profile.toolProbe;\n\t\tif (!toolProbe) {\n\t\t\tlines.push(\"  tool probe: none\");\n\t\t} else {\n\t\t\tconst variant = toolProbe.variant ? ` (${toolProbe.variant})` : \"\";\n\t\t\tconst nativeGrade = toolProbe.nativeGrade ? ` native=${toolProbe.nativeGrade}` : \"\";\n\t\t\tlines.push(\n\t\t\t\t`  tool probe: v${toolProbe.version} ${toolProbe.status}${variant}${nativeGrade} ${formatAgeDays(toolProbe.probedAt, now)}`,\n\t\t\t);\n\t\t\tif (toolProbe.diagnostic) lines.push(`  probe diagnostic: ${toolProbe.diagnostic}`);\n\t\t}\n\t\tconst protocol = entry.profile.protocol;\n\t\tif (!protocol) {\n\t\t\tlines.push(\"  protocol: none\");\n\t\t} else if (protocol.status === \"failed\") {\n\t\t\tlines.push(`  protocol: v${protocol.version} failed ${formatAgeDays(protocol.attemptedAt, now)}`);\n\t\t\tlines.push(`  variants tried: ${protocol.variantsTried.join(\", \")}`);\n\t\t\tlines.push(`  reset: /toolprotocol-reset ${entry.model}`);\n\t\t} else {\n\t\t\tlines.push(\n\t\t\t\t`  protocol: v${protocol.version} ${protocol.variant} calibrated ${formatAgeDays(protocol.calibratedAt, now)}`,\n\t\t\t);\n\t\t}\n\t\tif (entry.profile.rules.length === 0) {\n\t\t\tlines.push(\"  rules: none\");\n\t\t} else {\n\t\t\tlines.push(\"  rules:\");\n\t\t\tfor (const rule of [...entry.profile.rules].sort((left, right) => left.mode.localeCompare(right.mode))) {\n\t\t\t\tlines.push(`    - ${rule.mode} (${formatAgeDays(rule.lastFiredAt, now)}): ${rule.text}`);\n\t\t\t}\n\t\t}\n\t\tconst teachEntries = Object.entries(entry.profile.teachStats).sort(([left], [right]) =>\n\t\t\tleft.localeCompare(right),\n\t\t);\n\t\tif (teachEntries.length === 0) {\n\t\t\tlines.push(\"  teach stats: none\");\n\t\t} else {\n\t\t\tlines.push(\"  teach stats:\");\n\t\t\tfor (const [mode, stats] of teachEntries) {\n\t\t\t\tlines.push(\n\t\t\t\t\t`    - ${mode}: taught=${stats.taught} before=${stats.recurrenceBefore} after=${stats.recurrenceAfter}`,\n\t\t\t\t);\n\t\t\t}\n\t\t}\n\t}\n\treturn lines.join(\"\\n\");\n}\n"]}

package/dist/core/tool-repair-settings.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tool-repair-settings.d.ts","sourceRoot":"","sources":["../../src/core/tool-repair-settings.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,uBAAuB,CAAC;AAEtD,MAAM,WAAW,0BAA0B;IAC1C,MAAM,EAAE,OAAO,CAAC;IAChB,KAAK,EAAE,OAAO,CAAC;IACf,YAAY,EAAE,OAAO,GAAG,SAAS,CAAC;IAClC,OAAO,EAAE,OAAO,CAAC;CACjB;AAED,KAAK,OAAO,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC,CAAC;AAOlD,wBAAgB,yBAAyB,CACxC,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,YAAY,CAAC,EACtC,GAAG,EAAE,OAAO,GACV,0BAA0B,CAS5B;AAED,wBAAgB,gCAAgC,CAAC,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,YAAY,CAAC,GAAG,0BAA0B,CAEnH","sourcesContent":["import type { Settings } from \"./settings-manager.ts\";\n\nexport interface ResolvedToolRepairSettings {\n\trepair: boolean;\n\tteach: boolean;\n\ttextProtocol: boolean | undefined;\n\tlogging: boolean;\n}\n\ntype EnvLike = Record<string, string | undefined>;\n\nfunction isTruthyEnvFlag(value: string | undefined): boolean {\n\tif (!value) return false;\n\treturn [\"1\", \"true\", \"yes\", \"on\"].includes(value.trim().toLowerCase());\n}\n\nexport function resolveToolRepairSettings(\n\tsettings: Pick<Settings, \"toolRepair\">,\n\tenv: EnvLike,\n): ResolvedToolRepairSettings {\n\treturn {\n\t\trepair: !isTruthyEnvFlag(env.PI_TOOL_REPAIR_DISABLED),\n\t\tteach: !isTruthyEnvFlag(env.PI_TOOL_REPAIR_TEACH_DISABLED) && settings.toolRepair?.teach !== false,\n\t\ttextProtocol: isTruthyEnvFlag(env.PI_TEXT_TOOL_CALL_PROTOCOL_DISABLED)\n\t\t\t? false\n\t\t\t: settings.toolRepair?.textProtocol,\n\t\tlogging: settings.toolRepair?.logging !== false,\n\t};\n}\n\nexport function resolveCurrentToolRepairSettings(settings: Pick<Settings, \"toolRepair\">): ResolvedToolRepairSettings {\n\treturn resolveToolRepairSettings(settings, process.env);\n}\n"]}
+{"version":3,"file":"tool-repair-settings.d.ts","sourceRoot":"","sources":["../../src/core/tool-repair-settings.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,uBAAuB,CAAC;AAEtD,MAAM,WAAW,0BAA0B;IAC1C,MAAM,EAAE,OAAO,CAAC;IAChB,KAAK,EAAE,OAAO,CAAC;IACf,YAAY,EAAE,OAAO,GAAG,SAAS,CAAC;IAClC,OAAO,EAAE,OAAO,CAAC;CACjB;AAED,KAAK,OAAO,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC,CAAC;AAOlD,wBAAgB,yBAAyB,CACxC,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,YAAY,CAAC,EACtC,GAAG,EAAE,OAAO,GACV,0BAA0B,CAS5B;AAED,wBAAgB,gCAAgC,CAAC,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,YAAY,CAAC,GAAG,0BAA0B,CAEnH"}

package/dist/core/tool-repair-settings.js.map

@@ -1 +1 @@
-{"version":3,"file":"tool-repair-settings.js","sourceRoot":"","sources":["../../src/core/tool-repair-settings.ts"],"names":[],"mappings":"AAWA,SAAS,eAAe,CAAC,KAAyB,EAAW;IAC5D,IAAI,CAAC,KAAK;QAAE,OAAO,KAAK,CAAC;IACzB,OAAO,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,IAAI,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,CAAC;AAAA,CACvE;AAED,MAAM,UAAU,yBAAyB,CACxC,QAAsC,EACtC,GAAY,EACiB;IAC7B,OAAO;QACN,MAAM,EAAE,CAAC,eAAe,CAAC,GAAG,CAAC,uBAAuB,CAAC;QACrD,KAAK,EAAE,CAAC,eAAe,CAAC,GAAG,CAAC,6BAA6B,CAAC,IAAI,QAAQ,CAAC,UAAU,EAAE,KAAK,KAAK,KAAK;QAClG,YAAY,EAAE,eAAe,CAAC,GAAG,CAAC,mCAAmC,CAAC;YACrE,CAAC,CAAC,KAAK;YACP,CAAC,CAAC,QAAQ,CAAC,UAAU,EAAE,YAAY;QACpC,OAAO,EAAE,QAAQ,CAAC,UAAU,EAAE,OAAO,KAAK,KAAK;KAC/C,CAAC;AAAA,CACF;AAED,MAAM,UAAU,gCAAgC,CAAC,QAAsC,EAA8B;IACpH,OAAO,yBAAyB,CAAC,QAAQ,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC;AAAA,CACxD","sourcesContent":["import type { Settings } from \"./settings-manager.ts\";\n\nexport interface ResolvedToolRepairSettings {\n\trepair: boolean;\n\tteach: boolean;\n\ttextProtocol: boolean | undefined;\n\tlogging: boolean;\n}\n\ntype EnvLike = Record<string, string | undefined>;\n\nfunction isTruthyEnvFlag(value: string | undefined): boolean {\n\tif (!value) return false;\n\treturn [\"1\", \"true\", \"yes\", \"on\"].includes(value.trim().toLowerCase());\n}\n\nexport function resolveToolRepairSettings(\n\tsettings: Pick<Settings, \"toolRepair\">,\n\tenv: EnvLike,\n): ResolvedToolRepairSettings {\n\treturn {\n\t\trepair: !isTruthyEnvFlag(env.PI_TOOL_REPAIR_DISABLED),\n\t\tteach: !isTruthyEnvFlag(env.PI_TOOL_REPAIR_TEACH_DISABLED) && settings.toolRepair?.teach !== false,\n\t\ttextProtocol: isTruthyEnvFlag(env.PI_TEXT_TOOL_CALL_PROTOCOL_DISABLED)\n\t\t\t? false\n\t\t\t: settings.toolRepair?.textProtocol,\n\t\tlogging: settings.toolRepair?.logging !== false,\n\t};\n}\n\nexport function resolveCurrentToolRepairSettings(settings: Pick<Settings, \"toolRepair\">): ResolvedToolRepairSettings {\n\treturn resolveToolRepairSettings(settings, process.env);\n}\n"]}
+{"version":3,"file":"tool-repair-settings.js","sourceRoot":"","sources":["../../src/core/tool-repair-settings.ts"],"names":[],"mappings":"AAWA,SAAS,eAAe,CAAC,KAAyB;IACjD,IAAI,CAAC,KAAK;QAAE,OAAO,KAAK,CAAC;IACzB,OAAO,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,IAAI,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,CAAC;AACxE,CAAC;AAED,MAAM,UAAU,yBAAyB,CACxC,QAAsC,EACtC,GAAY;IAEZ,OAAO;QACN,MAAM,EAAE,CAAC,eAAe,CAAC,GAAG,CAAC,uBAAuB,CAAC;QACrD,KAAK,EAAE,CAAC,eAAe,CAAC,GAAG,CAAC,6BAA6B,CAAC,IAAI,QAAQ,CAAC,UAAU,EAAE,KAAK,KAAK,KAAK;QAClG,YAAY,EAAE,eAAe,CAAC,GAAG,CAAC,mCAAmC,CAAC;YACrE,CAAC,CAAC,KAAK;YACP,CAAC,CAAC,QAAQ,CAAC,UAAU,EAAE,YAAY;QACpC,OAAO,EAAE,QAAQ,CAAC,UAAU,EAAE,OAAO,KAAK,KAAK;KAC/C,CAAC;AACH,CAAC;AAED,MAAM,UAAU,gCAAgC,CAAC,QAAsC;IACtF,OAAO,yBAAyB,CAAC,QAAQ,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC;AACzD,CAAC","sourcesContent":["import type { Settings } from \"./settings-manager.ts\";\n\nexport interface ResolvedToolRepairSettings {\n\trepair: boolean;\n\tteach: boolean;\n\ttextProtocol: boolean | undefined;\n\tlogging: boolean;\n}\n\ntype EnvLike = Record<string, string | undefined>;\n\nfunction isTruthyEnvFlag(value: string | undefined): boolean {\n\tif (!value) return false;\n\treturn [\"1\", \"true\", \"yes\", \"on\"].includes(value.trim().toLowerCase());\n}\n\nexport function resolveToolRepairSettings(\n\tsettings: Pick<Settings, \"toolRepair\">,\n\tenv: EnvLike,\n): ResolvedToolRepairSettings {\n\treturn {\n\t\trepair: !isTruthyEnvFlag(env.PI_TOOL_REPAIR_DISABLED),\n\t\tteach: !isTruthyEnvFlag(env.PI_TOOL_REPAIR_TEACH_DISABLED) && settings.toolRepair?.teach !== false,\n\t\ttextProtocol: isTruthyEnvFlag(env.PI_TEXT_TOOL_CALL_PROTOCOL_DISABLED)\n\t\t\t? false\n\t\t\t: settings.toolRepair?.textProtocol,\n\t\tlogging: settings.toolRepair?.logging !== false,\n\t};\n}\n\nexport function resolveCurrentToolRepairSettings(settings: Pick<Settings, \"toolRepair\">): ResolvedToolRepairSettings {\n\treturn resolveToolRepairSettings(settings, process.env);\n}\n"]}

package/dist/core/tool-selection/expected-utility.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"expected-utility.d.ts","sourceRoot":"","sources":["../../../src/core/tool-selection/expected-utility.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,sBAAsB;IACtC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,cAAc,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,MAAM,CAAC;IACnB,eAAe,EAAE,MAAM,CAAC;IACxB,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,MAAM,CAAC;CACpB;AAED,eAAO,MAAM,gCAAgC,EAAE,sBAU9C,CAAC;AAEF,MAAM,WAAW,wBAAwB;IACxC,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAC7B,0FAA0F;IAC1F,iBAAiB,CAAC,EAAE,OAAO,CAAC;IAC5B,0EAA0E;IAC1E,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC5B;AAED,MAAM,WAAW,mBAAoB,SAAQ,wBAAwB;IACpE,kBAAkB,EAAE,MAAM,CAAC;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,MAAM,wBAAwB,GAAG,WAAW,GAAG,WAAW,GAAG,SAAS,CAAC;AAE7E,MAAM,WAAW,qBAAqB;IACrC,WAAW,EAAE,wBAAwB,CAAC;IACtC,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,mBAAmB,EAAE,CAAC;CAC9B;AAUD,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,MAAM,CAK1E;AAED,wBAAgB,iBAAiB,CAAC,aAAa,EAAE,SAAS,MAAM,EAAE,GAAG,MAAM,CAM1E;AAUD,wBAAgB,6BAA6B,CAC5C,UAAU,EAAE,SAAS,wBAAwB,EAAE,EAC/C,OAAO,GAAE,sBAAyD,GAChE,mBAAmB,EAAE,CAqBvB;AAED,wBAAgB,qBAAqB,CACpC,UAAU,EAAE,SAAS,wBAAwB,EAAE,EAC/C,OAAO,GAAE,sBAAyD,GAChE,qBAAqB,CAwBvB","sourcesContent":["export interface ExpectedUtilityWeights {\n\tlatency: number;\n\ttokens: number;\n\trisk: number;\n\tcontext: number;\n\tlatencyScaleMs: number;\n\ttokenScale: number;\n\tminimumEvidence: number;\n\tminimumMargin: number;\n\thighEntropy: number;\n}\n\nexport const DEFAULT_EXPECTED_UTILITY_WEIGHTS: ExpectedUtilityWeights = {\n\tlatency: 0.15,\n\ttokens: 0.1,\n\trisk: 0.2,\n\tcontext: 0.1,\n\tlatencyScaleMs: 5_000,\n\ttokenScale: 1_000,\n\tminimumEvidence: 3,\n\tminimumMargin: 0.1,\n\thighEntropy: 0.85,\n};\n\nexport interface ExpectedUtilityCandidate {\n\ttool: string;\n\tvalue: number;\n\talpha: number;\n\tbeta: number;\n\tsampleCount: number;\n\tlatencyMs?: number;\n\ttokenEstimate?: number;\n\triskCost?: number;\n\tcontextCost?: number;\n\tdeterministicMatch?: boolean;\n\t/** Candidates with unresolved paths may be observed/shortlisted but never recommended. */\n\tautomaticEligible?: boolean;\n\t/** `no_tool` uses a known probability rather than a learned posterior. */\n\tsuccessProbability?: number;\n}\n\nexport interface RankedToolCandidate extends ExpectedUtilityCandidate {\n\tsuccessProbability: number;\n\tlatencyCost: number;\n\ttokenCost: number;\n\triskCost: number;\n\tcontextCost: number;\n\tutility: number;\n\tprobability: number;\n}\n\nexport type ToolSelectionDisposition = \"recommend\" | \"shortlist\" | \"abstain\";\n\nexport interface ToolSelectionDecision {\n\tdisposition: ToolSelectionDisposition;\n\trecommendation?: string;\n\tshortlist: string[];\n\tentropy: number;\n\tmargin: number;\n\tranked: RankedToolCandidate[];\n}\n\nfunction clampUnit(value: number): number {\n\treturn Math.max(0, Math.min(1, value));\n}\n\nfunction finiteOr(value: number | undefined, fallback: number): number {\n\treturn typeof value === \"number\" && Number.isFinite(value) ? value : fallback;\n}\n\nexport function betaSuccessProbability(alpha: number, beta: number): number {\n\tconst positiveAlpha = Math.max(0, finiteOr(alpha, 0));\n\tconst positiveBeta = Math.max(0, finiteOr(beta, 0));\n\tconst total = positiveAlpha + positiveBeta;\n\treturn total > 0 ? positiveAlpha / total : 0.5;\n}\n\nexport function normalizedEntropy(probabilities: readonly number[]): number {\n\tif (probabilities.length <= 1) return 0;\n\tconst positive = probabilities.filter((probability) => probability > 0 && Number.isFinite(probability));\n\tif (positive.length === 0) return 0;\n\tconst entropy = -positive.reduce((total, probability) => total + probability * Math.log(probability), 0);\n\treturn entropy <= 0 ? 0 : entropy / Math.log(probabilities.length);\n}\n\nfunction softmax(utilities: readonly number[]): number[] {\n\tif (utilities.length === 0) return [];\n\tconst maximum = Math.max(...utilities);\n\tconst exponentials = utilities.map((utility) => Math.exp(Math.max(-60, utility - maximum)));\n\tconst total = exponentials.reduce((sum, value) => sum + value, 0);\n\treturn total > 0 ? exponentials.map((value) => value / total) : utilities.map(() => 1 / utilities.length);\n}\n\nexport function rankExpectedUtilityCandidates(\n\tcandidates: readonly ExpectedUtilityCandidate[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): RankedToolCandidate[] {\n\tconst unranked = candidates.map((candidate) => {\n\t\tconst successProbability = clampUnit(\n\t\t\tcandidate.successProbability ?? betaSuccessProbability(candidate.alpha, candidate.beta),\n\t\t);\n\t\tconst latencyCost = clampUnit(finiteOr(candidate.latencyMs, 0) / Math.max(1, weights.latencyScaleMs));\n\t\tconst tokenCost = clampUnit(finiteOr(candidate.tokenEstimate, 0) / Math.max(1, weights.tokenScale));\n\t\tconst riskCost = clampUnit(finiteOr(candidate.riskCost, 0));\n\t\tconst contextCost = clampUnit(finiteOr(candidate.contextCost, 0));\n\t\tconst utility =\n\t\t\tsuccessProbability * clampUnit(candidate.value) -\n\t\t\tweights.latency * latencyCost -\n\t\t\tweights.tokens * tokenCost -\n\t\t\tweights.risk * riskCost -\n\t\t\tweights.context * contextCost;\n\t\treturn { ...candidate, successProbability, latencyCost, tokenCost, riskCost, contextCost, utility };\n\t});\n\tconst probabilities = softmax(unranked.map((candidate) => candidate.utility));\n\treturn unranked\n\t\t.map((candidate, index) => ({ ...candidate, probability: probabilities[index] ?? 0 }))\n\t\t.sort((left, right) => right.utility - left.utility || left.tool.localeCompare(right.tool));\n}\n\nexport function decideExpectedUtility(\n\tcandidates: readonly ExpectedUtilityCandidate[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): ToolSelectionDecision {\n\tconst ranked = rankExpectedUtilityCandidates(candidates, weights);\n\tconst best = ranked[0];\n\tconst runnerUp = ranked[1];\n\tconst entropy = normalizedEntropy(ranked.map((candidate) => candidate.probability));\n\tconst margin = best && runnerUp ? best.utility - runnerUp.utility : (best?.utility ?? 0);\n\tconst shortlist = ranked.slice(0, 3).map((candidate) => candidate.tool);\n\tif (!best || best.utility <= 0) {\n\t\treturn { disposition: \"abstain\", shortlist: [], entropy, margin, ranked };\n\t}\n\tif (entropy >= weights.highEntropy || margin < weights.minimumMargin) {\n\t\treturn { disposition: \"shortlist\", shortlist, entropy, margin, ranked };\n\t}\n\tif (best.automaticEligible !== false && (best.deterministicMatch || best.sampleCount >= weights.minimumEvidence)) {\n\t\treturn {\n\t\t\tdisposition: \"recommend\",\n\t\t\trecommendation: best.tool,\n\t\t\tshortlist: [],\n\t\t\tentropy,\n\t\t\tmargin,\n\t\t\tranked,\n\t\t};\n\t}\n\treturn { disposition: \"abstain\", shortlist: [], entropy, margin, ranked };\n}\n"]}
+{"version":3,"file":"expected-utility.d.ts","sourceRoot":"","sources":["../../../src/core/tool-selection/expected-utility.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,sBAAsB;IACtC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,cAAc,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,MAAM,CAAC;IACnB,eAAe,EAAE,MAAM,CAAC;IACxB,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,MAAM,CAAC;CACpB;AAED,eAAO,MAAM,gCAAgC,EAAE,sBAU9C,CAAC;AAEF,MAAM,WAAW,wBAAwB;IACxC,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAC7B,0FAA0F;IAC1F,iBAAiB,CAAC,EAAE,OAAO,CAAC;IAC5B,0EAA0E;IAC1E,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC5B;AAED,MAAM,WAAW,mBAAoB,SAAQ,wBAAwB;IACpE,kBAAkB,EAAE,MAAM,CAAC;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,MAAM,wBAAwB,GAAG,WAAW,GAAG,WAAW,GAAG,SAAS,CAAC;AAE7E,MAAM,WAAW,qBAAqB;IACrC,WAAW,EAAE,wBAAwB,CAAC;IACtC,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,mBAAmB,EAAE,CAAC;CAC9B;AAUD,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,MAAM,CAK1E;AAED,wBAAgB,iBAAiB,CAAC,aAAa,EAAE,SAAS,MAAM,EAAE,GAAG,MAAM,CAM1E;AAUD,wBAAgB,6BAA6B,CAC5C,UAAU,EAAE,SAAS,wBAAwB,EAAE,EAC/C,OAAO,GAAE,sBAAyD,GAChE,mBAAmB,EAAE,CAqBvB;AAED,wBAAgB,qBAAqB,CACpC,UAAU,EAAE,SAAS,wBAAwB,EAAE,EAC/C,OAAO,GAAE,sBAAyD,GAChE,qBAAqB,CAwBvB"}

package/dist/core/tool-selection/expected-utility.js.map

@@ -1 +1 @@
-{"version":3,"file":"expected-utility.js","sourceRoot":"","sources":["../../../src/core/tool-selection/expected-utility.ts"],"names":[],"mappings":"AAYA,MAAM,CAAC,MAAM,gCAAgC,GAA2B;IACvE,OAAO,EAAE,IAAI;IACb,MAAM,EAAE,GAAG;IACX,IAAI,EAAE,GAAG;IACT,OAAO,EAAE,GAAG;IACZ,cAAc,EAAE,KAAK;IACrB,UAAU,EAAE,KAAK;IACjB,eAAe,EAAE,CAAC;IAClB,aAAa,EAAE,GAAG;IAClB,WAAW,EAAE,IAAI;CACjB,CAAC;AAwCF,SAAS,SAAS,CAAC,KAAa,EAAU;IACzC,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC;AAAA,CACvC;AAED,SAAS,QAAQ,CAAC,KAAyB,EAAE,QAAgB,EAAU;IACtE,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CAC9E;AAED,MAAM,UAAU,sBAAsB,CAAC,KAAa,EAAE,IAAY,EAAU;IAC3E,MAAM,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,CAAC;IACtD,MAAM,YAAY,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACpD,MAAM,KAAK,GAAG,aAAa,GAAG,YAAY,CAAC;IAC3C,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,GAAG,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC;AAAA,CAC/C;AAED,MAAM,UAAU,iBAAiB,CAAC,aAAgC,EAAU;IAC3E,IAAI,aAAa,CAAC,MAAM,IAAI,CAAC;QAAE,OAAO,CAAC,CAAC;IACxC,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,WAAW,EAAE,EAAE,CAAC,WAAW,GAAG,CAAC,IAAI,MAAM,CAAC,QAAQ,CAAC,WAAW,CAAC,CAAC,CAAC;IACxG,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IACpC,MAAM,OAAO,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,WAAW,EAAE,EAAE,CAAC,KAAK,GAAG,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,WAAW,CAAC,EAAE,CAAC,CAAC,CAAC;IACzG,OAAO,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC;AAAA,CACnE;AAED,SAAS,OAAO,CAAC,SAA4B,EAAY;IACxD,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACtC,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,SAAS,CAAC,CAAC;IACvC,MAAM,YAAY,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,OAAO,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;IAC5F,MAAM,KAAK,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,EAAE,CAAC,CAAC,CAAC;IAClE,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC;AAAA,CAC1G;AAED,MAAM,UAAU,6BAA6B,CAC5C,UAA+C,EAC/C,OAAO,GAA2B,gCAAgC,EAC1C;IACxB,MAAM,QAAQ,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC;QAC9C,MAAM,kBAAkB,GAAG,SAAS,CACnC,SAAS,CAAC,kBAAkB,IAAI,sBAAsB,CAAC,SAAS,CAAC,KAAK,EAAE,SAAS,CAAC,IAAI,CAAC,CACvF,CAAC;QACF,MAAM,WAAW,GAAG,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,SAAS,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,cAAc,CAAC,CAAC,CAAC;QACtG,MAAM,SAAS,GAAG,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,aAAa,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,UAAU,CAAC,CAAC,CAAC;QACpG,MAAM,QAAQ,GAAG,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,CAAC;QAC5D,MAAM,WAAW,GAAG,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,CAAC;QAClE,MAAM,OAAO,GACZ,kBAAkB,GAAG,SAAS,CAAC,SAAS,CAAC,KAAK,CAAC;YAC/C,OAAO,CAAC,OAAO,GAAG,WAAW;YAC7B,OAAO,CAAC,MAAM,GAAG,SAAS;YAC1B,OAAO,CAAC,IAAI,GAAG,QAAQ;YACvB,OAAO,CAAC,OAAO,GAAG,WAAW,CAAC;QAC/B,OAAO,EAAE,GAAG,SAAS,EAAE,kBAAkB,EAAE,WAAW,EAAE,SAAS,EAAE,QAAQ,EAAE,WAAW,EAAE,OAAO,EAAE,CAAC;IAAA,CACpG,CAAC,CAAC;IACH,MAAM,aAAa,GAAG,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC;IAC9E,OAAO,QAAQ;SACb,GAAG,CAAC,CAAC,SAAS,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,SAAS,EAAE,WAAW,EAAE,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;SACrF,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;AAAA,CAC7F;AAED,MAAM,UAAU,qBAAqB,CACpC,UAA+C,EAC/C,OAAO,GAA2B,gCAAgC,EAC1C;IACxB,MAAM,MAAM,GAAG,6BAA6B,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;IAClE,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;IACvB,MAAM,QAAQ,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;IAC3B,MAAM,OAAO,GAAG,iBAAiB,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,WAAW,CAAC,CAAC,CAAC;IACpF,MAAM,MAAM,GAAG,IAAI,IAAI,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,GAAG,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,OAAO,IAAI,CAAC,CAAC,CAAC;IACzF,MAAM,SAAS,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;IACxE,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,OAAO,IAAI,CAAC,EAAE,CAAC;QAChC,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;IAC3E,CAAC;IACD,IAAI,OAAO,IAAI,OAAO,CAAC,WAAW,IAAI,MAAM,GAAG,OAAO,CAAC,aAAa,EAAE,CAAC;QACtE,OAAO,EAAE,WAAW,EAAE,WAAW,EAAE,SAAS,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;IACzE,CAAC;IACD,IAAI,IAAI,CAAC,iBAAiB,KAAK,KAAK,IAAI,CAAC,IAAI,CAAC,kBAAkB,IAAI,IAAI,CAAC,WAAW,IAAI,OAAO,CAAC,eAAe,CAAC,EAAE,CAAC;QAClH,OAAO;YACN,WAAW,EAAE,WAAW;YACxB,cAAc,EAAE,IAAI,CAAC,IAAI;YACzB,SAAS,EAAE,EAAE;YACb,OAAO;YACP,MAAM;YACN,MAAM;SACN,CAAC;IACH,CAAC;IACD,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;AAAA,CAC1E","sourcesContent":["export interface ExpectedUtilityWeights {\n\tlatency: number;\n\ttokens: number;\n\trisk: number;\n\tcontext: number;\n\tlatencyScaleMs: number;\n\ttokenScale: number;\n\tminimumEvidence: number;\n\tminimumMargin: number;\n\thighEntropy: number;\n}\n\nexport const DEFAULT_EXPECTED_UTILITY_WEIGHTS: ExpectedUtilityWeights = {\n\tlatency: 0.15,\n\ttokens: 0.1,\n\trisk: 0.2,\n\tcontext: 0.1,\n\tlatencyScaleMs: 5_000,\n\ttokenScale: 1_000,\n\tminimumEvidence: 3,\n\tminimumMargin: 0.1,\n\thighEntropy: 0.85,\n};\n\nexport interface ExpectedUtilityCandidate {\n\ttool: string;\n\tvalue: number;\n\talpha: number;\n\tbeta: number;\n\tsampleCount: number;\n\tlatencyMs?: number;\n\ttokenEstimate?: number;\n\triskCost?: number;\n\tcontextCost?: number;\n\tdeterministicMatch?: boolean;\n\t/** Candidates with unresolved paths may be observed/shortlisted but never recommended. */\n\tautomaticEligible?: boolean;\n\t/** `no_tool` uses a known probability rather than a learned posterior. */\n\tsuccessProbability?: number;\n}\n\nexport interface RankedToolCandidate extends ExpectedUtilityCandidate {\n\tsuccessProbability: number;\n\tlatencyCost: number;\n\ttokenCost: number;\n\triskCost: number;\n\tcontextCost: number;\n\tutility: number;\n\tprobability: number;\n}\n\nexport type ToolSelectionDisposition = \"recommend\" | \"shortlist\" | \"abstain\";\n\nexport interface ToolSelectionDecision {\n\tdisposition: ToolSelectionDisposition;\n\trecommendation?: string;\n\tshortlist: string[];\n\tentropy: number;\n\tmargin: number;\n\tranked: RankedToolCandidate[];\n}\n\nfunction clampUnit(value: number): number {\n\treturn Math.max(0, Math.min(1, value));\n}\n\nfunction finiteOr(value: number | undefined, fallback: number): number {\n\treturn typeof value === \"number\" && Number.isFinite(value) ? value : fallback;\n}\n\nexport function betaSuccessProbability(alpha: number, beta: number): number {\n\tconst positiveAlpha = Math.max(0, finiteOr(alpha, 0));\n\tconst positiveBeta = Math.max(0, finiteOr(beta, 0));\n\tconst total = positiveAlpha + positiveBeta;\n\treturn total > 0 ? positiveAlpha / total : 0.5;\n}\n\nexport function normalizedEntropy(probabilities: readonly number[]): number {\n\tif (probabilities.length <= 1) return 0;\n\tconst positive = probabilities.filter((probability) => probability > 0 && Number.isFinite(probability));\n\tif (positive.length === 0) return 0;\n\tconst entropy = -positive.reduce((total, probability) => total + probability * Math.log(probability), 0);\n\treturn entropy <= 0 ? 0 : entropy / Math.log(probabilities.length);\n}\n\nfunction softmax(utilities: readonly number[]): number[] {\n\tif (utilities.length === 0) return [];\n\tconst maximum = Math.max(...utilities);\n\tconst exponentials = utilities.map((utility) => Math.exp(Math.max(-60, utility - maximum)));\n\tconst total = exponentials.reduce((sum, value) => sum + value, 0);\n\treturn total > 0 ? exponentials.map((value) => value / total) : utilities.map(() => 1 / utilities.length);\n}\n\nexport function rankExpectedUtilityCandidates(\n\tcandidates: readonly ExpectedUtilityCandidate[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): RankedToolCandidate[] {\n\tconst unranked = candidates.map((candidate) => {\n\t\tconst successProbability = clampUnit(\n\t\t\tcandidate.successProbability ?? betaSuccessProbability(candidate.alpha, candidate.beta),\n\t\t);\n\t\tconst latencyCost = clampUnit(finiteOr(candidate.latencyMs, 0) / Math.max(1, weights.latencyScaleMs));\n\t\tconst tokenCost = clampUnit(finiteOr(candidate.tokenEstimate, 0) / Math.max(1, weights.tokenScale));\n\t\tconst riskCost = clampUnit(finiteOr(candidate.riskCost, 0));\n\t\tconst contextCost = clampUnit(finiteOr(candidate.contextCost, 0));\n\t\tconst utility =\n\t\t\tsuccessProbability * clampUnit(candidate.value) -\n\t\t\tweights.latency * latencyCost -\n\t\t\tweights.tokens * tokenCost -\n\t\t\tweights.risk * riskCost -\n\t\t\tweights.context * contextCost;\n\t\treturn { ...candidate, successProbability, latencyCost, tokenCost, riskCost, contextCost, utility };\n\t});\n\tconst probabilities = softmax(unranked.map((candidate) => candidate.utility));\n\treturn unranked\n\t\t.map((candidate, index) => ({ ...candidate, probability: probabilities[index] ?? 0 }))\n\t\t.sort((left, right) => right.utility - left.utility || left.tool.localeCompare(right.tool));\n}\n\nexport function decideExpectedUtility(\n\tcandidates: readonly ExpectedUtilityCandidate[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): ToolSelectionDecision {\n\tconst ranked = rankExpectedUtilityCandidates(candidates, weights);\n\tconst best = ranked[0];\n\tconst runnerUp = ranked[1];\n\tconst entropy = normalizedEntropy(ranked.map((candidate) => candidate.probability));\n\tconst margin = best && runnerUp ? best.utility - runnerUp.utility : (best?.utility ?? 0);\n\tconst shortlist = ranked.slice(0, 3).map((candidate) => candidate.tool);\n\tif (!best || best.utility <= 0) {\n\t\treturn { disposition: \"abstain\", shortlist: [], entropy, margin, ranked };\n\t}\n\tif (entropy >= weights.highEntropy || margin < weights.minimumMargin) {\n\t\treturn { disposition: \"shortlist\", shortlist, entropy, margin, ranked };\n\t}\n\tif (best.automaticEligible !== false && (best.deterministicMatch || best.sampleCount >= weights.minimumEvidence)) {\n\t\treturn {\n\t\t\tdisposition: \"recommend\",\n\t\t\trecommendation: best.tool,\n\t\t\tshortlist: [],\n\t\t\tentropy,\n\t\t\tmargin,\n\t\t\tranked,\n\t\t};\n\t}\n\treturn { disposition: \"abstain\", shortlist: [], entropy, margin, ranked };\n}\n"]}
+{"version":3,"file":"expected-utility.js","sourceRoot":"","sources":["../../../src/core/tool-selection/expected-utility.ts"],"names":[],"mappings":"AAYA,MAAM,CAAC,MAAM,gCAAgC,GAA2B;IACvE,OAAO,EAAE,IAAI;IACb,MAAM,EAAE,GAAG;IACX,IAAI,EAAE,GAAG;IACT,OAAO,EAAE,GAAG;IACZ,cAAc,EAAE,KAAK;IACrB,UAAU,EAAE,KAAK;IACjB,eAAe,EAAE,CAAC;IAClB,aAAa,EAAE,GAAG;IAClB,WAAW,EAAE,IAAI;CACjB,CAAC;AAwCF,SAAS,SAAS,CAAC,KAAa;IAC/B,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC;AACxC,CAAC;AAED,SAAS,QAAQ,CAAC,KAAyB,EAAE,QAAgB;IAC5D,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC;AAC/E,CAAC;AAED,MAAM,UAAU,sBAAsB,CAAC,KAAa,EAAE,IAAY;IACjE,MAAM,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,CAAC;IACtD,MAAM,YAAY,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACpD,MAAM,KAAK,GAAG,aAAa,GAAG,YAAY,CAAC;IAC3C,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,GAAG,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC;AAChD,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,aAAgC;IACjE,IAAI,aAAa,CAAC,MAAM,IAAI,CAAC;QAAE,OAAO,CAAC,CAAC;IACxC,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,WAAW,EAAE,EAAE,CAAC,WAAW,GAAG,CAAC,IAAI,MAAM,CAAC,QAAQ,CAAC,WAAW,CAAC,CAAC,CAAC;IACxG,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IACpC,MAAM,OAAO,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,WAAW,EAAE,EAAE,CAAC,KAAK,GAAG,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,WAAW,CAAC,EAAE,CAAC,CAAC,CAAC;IACzG,OAAO,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC;AACpE,CAAC;AAED,SAAS,OAAO,CAAC,SAA4B;IAC5C,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACtC,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,SAAS,CAAC,CAAC;IACvC,MAAM,YAAY,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,OAAO,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;IAC5F,MAAM,KAAK,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,EAAE,CAAC,CAAC,CAAC;IAClE,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC;AAC3G,CAAC;AAED,MAAM,UAAU,6BAA6B,CAC5C,UAA+C,EAC/C,OAAO,GAA2B,gCAAgC;IAElE,MAAM,QAAQ,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE;QAC7C,MAAM,kBAAkB,GAAG,SAAS,CACnC,SAAS,CAAC,kBAAkB,IAAI,sBAAsB,CAAC,SAAS,CAAC,KAAK,EAAE,SAAS,CAAC,IAAI,CAAC,CACvF,CAAC;QACF,MAAM,WAAW,GAAG,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,SAAS,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,cAAc,CAAC,CAAC,CAAC;QACtG,MAAM,SAAS,GAAG,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,aAAa,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,UAAU,CAAC,CAAC,CAAC;QACpG,MAAM,QAAQ,GAAG,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,CAAC;QAC5D,MAAM,WAAW,GAAG,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,CAAC;QAClE,MAAM,OAAO,GACZ,kBAAkB,GAAG,SAAS,CAAC,SAAS,CAAC,KAAK,CAAC;YAC/C,OAAO,CAAC,OAAO,GAAG,WAAW;YAC7B,OAAO,CAAC,MAAM,GAAG,SAAS;YAC1B,OAAO,CAAC,IAAI,GAAG,QAAQ;YACvB,OAAO,CAAC,OAAO,GAAG,WAAW,CAAC;QAC/B,OAAO,EAAE,GAAG,SAAS,EAAE,kBAAkB,EAAE,WAAW,EAAE,SAAS,EAAE,QAAQ,EAAE,WAAW,EAAE,OAAO,EAAE,CAAC;IACrG,CAAC,CAAC,CAAC;IACH,MAAM,aAAa,GAAG,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC;IAC9E,OAAO,QAAQ;SACb,GAAG,CAAC,CAAC,SAAS,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,SAAS,EAAE,WAAW,EAAE,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;SACrF,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;AAC9F,CAAC;AAED,MAAM,UAAU,qBAAqB,CACpC,UAA+C,EAC/C,OAAO,GAA2B,gCAAgC;IAElE,MAAM,MAAM,GAAG,6BAA6B,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;IAClE,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;IACvB,MAAM,QAAQ,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;IAC3B,MAAM,OAAO,GAAG,iBAAiB,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,WAAW,CAAC,CAAC,CAAC;IACpF,MAAM,MAAM,GAAG,IAAI,IAAI,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,GAAG,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,OAAO,IAAI,CAAC,CAAC,CAAC;IACzF,MAAM,SAAS,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;IACxE,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,OAAO,IAAI,CAAC,EAAE,CAAC;QAChC,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;IAC3E,CAAC;IACD,IAAI,OAAO,IAAI,OAAO,CAAC,WAAW,IAAI,MAAM,GAAG,OAAO,CAAC,aAAa,EAAE,CAAC;QACtE,OAAO,EAAE,WAAW,EAAE,WAAW,EAAE,SAAS,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;IACzE,CAAC;IACD,IAAI,IAAI,CAAC,iBAAiB,KAAK,KAAK,IAAI,CAAC,IAAI,CAAC,kBAAkB,IAAI,IAAI,CAAC,WAAW,IAAI,OAAO,CAAC,eAAe,CAAC,EAAE,CAAC;QAClH,OAAO;YACN,WAAW,EAAE,WAAW;YACxB,cAAc,EAAE,IAAI,CAAC,IAAI;YACzB,SAAS,EAAE,EAAE;YACb,OAAO;YACP,MAAM;YACN,MAAM;SACN,CAAC;IACH,CAAC;IACD,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;AAC3E,CAAC","sourcesContent":["export interface ExpectedUtilityWeights {\n\tlatency: number;\n\ttokens: number;\n\trisk: number;\n\tcontext: number;\n\tlatencyScaleMs: number;\n\ttokenScale: number;\n\tminimumEvidence: number;\n\tminimumMargin: number;\n\thighEntropy: number;\n}\n\nexport const DEFAULT_EXPECTED_UTILITY_WEIGHTS: ExpectedUtilityWeights = {\n\tlatency: 0.15,\n\ttokens: 0.1,\n\trisk: 0.2,\n\tcontext: 0.1,\n\tlatencyScaleMs: 5_000,\n\ttokenScale: 1_000,\n\tminimumEvidence: 3,\n\tminimumMargin: 0.1,\n\thighEntropy: 0.85,\n};\n\nexport interface ExpectedUtilityCandidate {\n\ttool: string;\n\tvalue: number;\n\talpha: number;\n\tbeta: number;\n\tsampleCount: number;\n\tlatencyMs?: number;\n\ttokenEstimate?: number;\n\triskCost?: number;\n\tcontextCost?: number;\n\tdeterministicMatch?: boolean;\n\t/** Candidates with unresolved paths may be observed/shortlisted but never recommended. */\n\tautomaticEligible?: boolean;\n\t/** `no_tool` uses a known probability rather than a learned posterior. */\n\tsuccessProbability?: number;\n}\n\nexport interface RankedToolCandidate extends ExpectedUtilityCandidate {\n\tsuccessProbability: number;\n\tlatencyCost: number;\n\ttokenCost: number;\n\triskCost: number;\n\tcontextCost: number;\n\tutility: number;\n\tprobability: number;\n}\n\nexport type ToolSelectionDisposition = \"recommend\" | \"shortlist\" | \"abstain\";\n\nexport interface ToolSelectionDecision {\n\tdisposition: ToolSelectionDisposition;\n\trecommendation?: string;\n\tshortlist: string[];\n\tentropy: number;\n\tmargin: number;\n\tranked: RankedToolCandidate[];\n}\n\nfunction clampUnit(value: number): number {\n\treturn Math.max(0, Math.min(1, value));\n}\n\nfunction finiteOr(value: number | undefined, fallback: number): number {\n\treturn typeof value === \"number\" && Number.isFinite(value) ? value : fallback;\n}\n\nexport function betaSuccessProbability(alpha: number, beta: number): number {\n\tconst positiveAlpha = Math.max(0, finiteOr(alpha, 0));\n\tconst positiveBeta = Math.max(0, finiteOr(beta, 0));\n\tconst total = positiveAlpha + positiveBeta;\n\treturn total > 0 ? positiveAlpha / total : 0.5;\n}\n\nexport function normalizedEntropy(probabilities: readonly number[]): number {\n\tif (probabilities.length <= 1) return 0;\n\tconst positive = probabilities.filter((probability) => probability > 0 && Number.isFinite(probability));\n\tif (positive.length === 0) return 0;\n\tconst entropy = -positive.reduce((total, probability) => total + probability * Math.log(probability), 0);\n\treturn entropy <= 0 ? 0 : entropy / Math.log(probabilities.length);\n}\n\nfunction softmax(utilities: readonly number[]): number[] {\n\tif (utilities.length === 0) return [];\n\tconst maximum = Math.max(...utilities);\n\tconst exponentials = utilities.map((utility) => Math.exp(Math.max(-60, utility - maximum)));\n\tconst total = exponentials.reduce((sum, value) => sum + value, 0);\n\treturn total > 0 ? exponentials.map((value) => value / total) : utilities.map(() => 1 / utilities.length);\n}\n\nexport function rankExpectedUtilityCandidates(\n\tcandidates: readonly ExpectedUtilityCandidate[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): RankedToolCandidate[] {\n\tconst unranked = candidates.map((candidate) => {\n\t\tconst successProbability = clampUnit(\n\t\t\tcandidate.successProbability ?? betaSuccessProbability(candidate.alpha, candidate.beta),\n\t\t);\n\t\tconst latencyCost = clampUnit(finiteOr(candidate.latencyMs, 0) / Math.max(1, weights.latencyScaleMs));\n\t\tconst tokenCost = clampUnit(finiteOr(candidate.tokenEstimate, 0) / Math.max(1, weights.tokenScale));\n\t\tconst riskCost = clampUnit(finiteOr(candidate.riskCost, 0));\n\t\tconst contextCost = clampUnit(finiteOr(candidate.contextCost, 0));\n\t\tconst utility =\n\t\t\tsuccessProbability * clampUnit(candidate.value) -\n\t\t\tweights.latency * latencyCost -\n\t\t\tweights.tokens * tokenCost -\n\t\t\tweights.risk * riskCost -\n\t\t\tweights.context * contextCost;\n\t\treturn { ...candidate, successProbability, latencyCost, tokenCost, riskCost, contextCost, utility };\n\t});\n\tconst probabilities = softmax(unranked.map((candidate) => candidate.utility));\n\treturn unranked\n\t\t.map((candidate, index) => ({ ...candidate, probability: probabilities[index] ?? 0 }))\n\t\t.sort((left, right) => right.utility - left.utility || left.tool.localeCompare(right.tool));\n}\n\nexport function decideExpectedUtility(\n\tcandidates: readonly ExpectedUtilityCandidate[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): ToolSelectionDecision {\n\tconst ranked = rankExpectedUtilityCandidates(candidates, weights);\n\tconst best = ranked[0];\n\tconst runnerUp = ranked[1];\n\tconst entropy = normalizedEntropy(ranked.map((candidate) => candidate.probability));\n\tconst margin = best && runnerUp ? best.utility - runnerUp.utility : (best?.utility ?? 0);\n\tconst shortlist = ranked.slice(0, 3).map((candidate) => candidate.tool);\n\tif (!best || best.utility <= 0) {\n\t\treturn { disposition: \"abstain\", shortlist: [], entropy, margin, ranked };\n\t}\n\tif (entropy >= weights.highEntropy || margin < weights.minimumMargin) {\n\t\treturn { disposition: \"shortlist\", shortlist, entropy, margin, ranked };\n\t}\n\tif (best.automaticEligible !== false && (best.deterministicMatch || best.sampleCount >= weights.minimumEvidence)) {\n\t\treturn {\n\t\t\tdisposition: \"recommend\",\n\t\t\trecommendation: best.tool,\n\t\t\tshortlist: [],\n\t\t\tentropy,\n\t\t\tmargin,\n\t\t\tranked,\n\t\t};\n\t}\n\treturn { disposition: \"abstain\", shortlist: [], entropy, margin, ranked };\n}\n"]}

package/dist/core/tool-selection/promotion.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"promotion.d.ts","sourceRoot":"","sources":["../../../src/core/tool-selection/promotion.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AAEH,OAAO,EAIN,KAAK,sBAAsB,EAC3B,MAAM,uBAAuB,CAAC;AAC/B,OAAO,KAAK,EAAE,oBAAoB,EAAE,wBAAwB,EAAE,MAAM,6BAA6B,CAAC;AAElG,MAAM,WAAW,iBAAiB;IACjC,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,wBAAwB,CAAC;IACtC,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,qBAAqB;IACrC,mFAAmF;IACnF,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,CAAC;IACpB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;CAChB;AAqBD;;;;GAIG;AACH,wBAAgB,qBAAqB,CACpC,cAAc,EAAE,SAAS,oBAAoB,EAAE,EAC/C,OAAO,GAAE,sBAAyD,GAChE,qBAAqB,CAevB;AAED;;;;;GAKG;AACH,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,SAAS,iBAAiB,EAAE,GAAG,MAAM,GAAG,SAAS,CAUhG","sourcesContent":["/**\n * Evidence-gated tool-selection promotion (closing the observe/promote loop).\n *\n * `tool-selection-controller.ts` records, per tool call, a full expected-utility decision\n * (`ToolSelectionController.begin()`), but that per-call decision is scoped to whichever\n * tool the model already chose — only the chosen tool is ever eligible to be \"recommended\"\n * (see `candidateFor` in tool-selection-controller.ts), so it can validate a choice but never\n * suggest a different one. This module is the SEPARATE, aggregate read-side: given the\n * per-(model,intent,tool) track record accumulated in `ToolPerformanceStore` over many calls,\n * decide whether one tool has become the clear, evidence-backed default for a (model, intent)\n * pair — and if so, render a compact, cache-stable prompt hint.\n *\n * Reuses the exact same expected-utility gate (`decideExpectedUtility`) as call-time decisions,\n * so promotion and the call-time recommend/shortlist/abstain gate share one threshold\n * vocabulary: positive best-utility, a minimum margin over the runner-up, bounded entropy, and a\n * minimum sample count. A tool is promoted only when the gate returns \"recommend\"; ambiguous\n * (high-entropy/low-margin) or thin-evidence buckets never surface a hint — they stay silent,\n * which is the \"shortlist on high entropy, never a false-confident pick\" behavior.\n *\n * Purely functional: no store/env access, so it is trivial to unit test and never mutates state.\n * The stateful pieces (kill switches, reading the store, formatting into the live system prompt)\n * live in `tool-selection-controller.ts` and `system-prompt-builder.ts`.\n */\n\nimport {\n\tDEFAULT_EXPECTED_UTILITY_WEIGHTS,\n\tdecideExpectedUtility,\n\ttype ExpectedUtilityCandidate,\n\ttype ExpectedUtilityWeights,\n} from \"./expected-utility.ts\";\nimport type { ToolPerformanceStats, ToolSelectionIntentClass } from \"./tool-performance-store.ts\";\n\nexport interface ToolSelectionHint {\n\tmodelRef: string;\n\tintentClass: ToolSelectionIntentClass;\n\ttool: string;\n\tsampleCount: number;\n\tmargin: number;\n\tentropy: number;\n}\n\nexport interface ToolPromotionDecision {\n\t/** The promoted tool, present only when the evidence gate returned \"recommend\". */\n\ttool?: string;\n\tsampleCount: number;\n\tmargin: number;\n\tentropy: number;\n}\n\nfunction candidateFromStats(stats: ToolPerformanceStats): ExpectedUtilityCandidate {\n\treturn {\n\t\ttool: stats.tool,\n\t\t// All candidates here were recorded under the SAME (modelRef, intentClass) bucket — the\n\t\t// write path only ever stores a tool's stats under its own classified intent (see\n\t\t// classifyToolIntent + candidateFor in tool-selection-controller.ts) — so there is no\n\t\t// per-candidate affinity prior left to apply; every candidate is equally \"in scope\".\n\t\tvalue: 1,\n\t\talpha: stats.alpha,\n\t\tbeta: stats.beta,\n\t\tsampleCount: stats.sampleCount,\n\t\tlatencyMs: stats.latencyEwmaMs,\n\t\ttokenEstimate:\n\t\t\tstats.inputTokenEstimateEwma === undefined && stats.outputTokenEstimateEwma === undefined\n\t\t\t\t? undefined\n\t\t\t\t: (stats.inputTokenEstimateEwma ?? 0) + (stats.outputTokenEstimateEwma ?? 0),\n\t};\n}\n\n/**\n * Evidence-gated promotion for one (model, intentClass) bucket: which tool, if any, has become\n * the clear default given its accumulated track record. `statsForIntent` should already be\n * scoped to a single (modelRef, intentClass) — see `ToolPerformanceStore.getStatsForIntent`.\n */\nexport function evaluateToolPromotion(\n\tstatsForIntent: readonly ToolPerformanceStats[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): ToolPromotionDecision {\n\tif (statsForIntent.length === 0) {\n\t\treturn { sampleCount: 0, margin: 0, entropy: 0 };\n\t}\n\tconst decision = decideExpectedUtility(statsForIntent.map(candidateFromStats), weights);\n\tif (decision.disposition !== \"recommend\" || !decision.recommendation) {\n\t\treturn { sampleCount: 0, margin: decision.margin, entropy: decision.entropy };\n\t}\n\tconst winner = decision.ranked.find((candidate) => candidate.tool === decision.recommendation);\n\treturn {\n\t\ttool: decision.recommendation,\n\t\tsampleCount: winner?.sampleCount ?? 0,\n\t\tmargin: decision.margin,\n\t\tentropy: decision.entropy,\n\t};\n}\n\n/**\n * Renders the active hints into a small, stable prompt block. Deliberately omits live numbers\n * (sample counts, rates) that would change every time evidence accumulates — only the SET of\n * (intent -> tool) pairs is included, so the block's text changes only when the promoted tool for\n * an intent actually flips, not on every tool call (system-prompt cache stability).\n */\nexport function formatToolSelectionHints(hints: readonly ToolSelectionHint[]): string | undefined {\n\tif (hints.length === 0) return undefined;\n\tconst lines = [...hints]\n\t\t.sort((left, right) => left.intentClass.localeCompare(right.intentClass))\n\t\t.map((hint) => `- ${hint.intentClass}: \\`${hint.tool}\\` has the established track record for this model.`);\n\treturn [\n\t\t\"Learned tool preferences (evidence-gated observation, not a directive):\",\n\t\t...lines,\n\t\t\"Treat as a shortlist signal only — use judgment for the actual task at hand.\",\n\t].join(\"\\n\");\n}\n"]}
+{"version":3,"file":"promotion.d.ts","sourceRoot":"","sources":["../../../src/core/tool-selection/promotion.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AAEH,OAAO,EAIN,KAAK,sBAAsB,EAC3B,MAAM,uBAAuB,CAAC;AAC/B,OAAO,KAAK,EAAE,oBAAoB,EAAE,wBAAwB,EAAE,MAAM,6BAA6B,CAAC;AAElG,MAAM,WAAW,iBAAiB;IACjC,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,wBAAwB,CAAC;IACtC,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,qBAAqB;IACrC,mFAAmF;IACnF,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,CAAC;IACpB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;CAChB;AAqBD;;;;GAIG;AACH,wBAAgB,qBAAqB,CACpC,cAAc,EAAE,SAAS,oBAAoB,EAAE,EAC/C,OAAO,GAAE,sBAAyD,GAChE,qBAAqB,CAevB;AAED;;;;;GAKG;AACH,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,SAAS,iBAAiB,EAAE,GAAG,MAAM,GAAG,SAAS,CAUhG"}

package/dist/core/tool-selection/promotion.js.map

@@ -1 +1 @@
-{"version":3,"file":"promotion.js","sourceRoot":"","sources":["../../../src/core/tool-selection/promotion.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AAEH,OAAO,EACN,gCAAgC,EAChC,qBAAqB,GAGrB,MAAM,uBAAuB,CAAC;AAoB/B,SAAS,kBAAkB,CAAC,KAA2B,EAA4B;IAClF,OAAO;QACN,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,0FAAwF;QACxF,kFAAkF;QAClF,wFAAsF;QACtF,qFAAqF;QACrF,KAAK,EAAE,CAAC;QACR,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,WAAW,EAAE,KAAK,CAAC,WAAW;QAC9B,SAAS,EAAE,KAAK,CAAC,aAAa;QAC9B,aAAa,EACZ,KAAK,CAAC,sBAAsB,KAAK,SAAS,IAAI,KAAK,CAAC,uBAAuB,KAAK,SAAS;YACxF,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,CAAC,KAAK,CAAC,sBAAsB,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,uBAAuB,IAAI,CAAC,CAAC;KAC9E,CAAC;AAAA,CACF;AAED;;;;GAIG;AACH,MAAM,UAAU,qBAAqB,CACpC,cAA+C,EAC/C,OAAO,GAA2B,gCAAgC,EAC1C;IACxB,IAAI,cAAc,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACjC,OAAO,EAAE,WAAW,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IAClD,CAAC;IACD,MAAM,QAAQ,GAAG,qBAAqB,CAAC,cAAc,CAAC,GAAG,CAAC,kBAAkB,CAAC,EAAE,OAAO,CAAC,CAAC;IACxF,IAAI,QAAQ,CAAC,WAAW,KAAK,WAAW,IAAI,CAAC,QAAQ,CAAC,cAAc,EAAE,CAAC;QACtE,OAAO,EAAE,WAAW,EAAE,CAAC,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,OAAO,EAAE,QAAQ,CAAC,OAAO,EAAE,CAAC;IAC/E,CAAC;IACD,MAAM,MAAM,GAAG,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,IAAI,KAAK,QAAQ,CAAC,cAAc,CAAC,CAAC;IAC/F,OAAO;QACN,IAAI,EAAE,QAAQ,CAAC,cAAc;QAC7B,WAAW,EAAE,MAAM,EAAE,WAAW,IAAI,CAAC;QACrC,MAAM,EAAE,QAAQ,CAAC,MAAM;QACvB,OAAO,EAAE,QAAQ,CAAC,OAAO;KACzB,CAAC;AAAA,CACF;AAED;;;;;GAKG;AACH,MAAM,UAAU,wBAAwB,CAAC,KAAmC,EAAsB;IACjG,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IACzC,MAAM,KAAK,GAAG,CAAC,GAAG,KAAK,CAAC;SACtB,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,WAAW,CAAC,aAAa,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;SACxE,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,KAAK,IAAI,CAAC,WAAW,OAAO,IAAI,CAAC,IAAI,qDAAqD,CAAC,CAAC;IAC5G,OAAO;QACN,yEAAyE;QACzE,GAAG,KAAK;QACR,gFAA8E;KAC9E,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAAA,CACb","sourcesContent":["/**\n * Evidence-gated tool-selection promotion (closing the observe/promote loop).\n *\n * `tool-selection-controller.ts` records, per tool call, a full expected-utility decision\n * (`ToolSelectionController.begin()`), but that per-call decision is scoped to whichever\n * tool the model already chose — only the chosen tool is ever eligible to be \"recommended\"\n * (see `candidateFor` in tool-selection-controller.ts), so it can validate a choice but never\n * suggest a different one. This module is the SEPARATE, aggregate read-side: given the\n * per-(model,intent,tool) track record accumulated in `ToolPerformanceStore` over many calls,\n * decide whether one tool has become the clear, evidence-backed default for a (model, intent)\n * pair — and if so, render a compact, cache-stable prompt hint.\n *\n * Reuses the exact same expected-utility gate (`decideExpectedUtility`) as call-time decisions,\n * so promotion and the call-time recommend/shortlist/abstain gate share one threshold\n * vocabulary: positive best-utility, a minimum margin over the runner-up, bounded entropy, and a\n * minimum sample count. A tool is promoted only when the gate returns \"recommend\"; ambiguous\n * (high-entropy/low-margin) or thin-evidence buckets never surface a hint — they stay silent,\n * which is the \"shortlist on high entropy, never a false-confident pick\" behavior.\n *\n * Purely functional: no store/env access, so it is trivial to unit test and never mutates state.\n * The stateful pieces (kill switches, reading the store, formatting into the live system prompt)\n * live in `tool-selection-controller.ts` and `system-prompt-builder.ts`.\n */\n\nimport {\n\tDEFAULT_EXPECTED_UTILITY_WEIGHTS,\n\tdecideExpectedUtility,\n\ttype ExpectedUtilityCandidate,\n\ttype ExpectedUtilityWeights,\n} from \"./expected-utility.ts\";\nimport type { ToolPerformanceStats, ToolSelectionIntentClass } from \"./tool-performance-store.ts\";\n\nexport interface ToolSelectionHint {\n\tmodelRef: string;\n\tintentClass: ToolSelectionIntentClass;\n\ttool: string;\n\tsampleCount: number;\n\tmargin: number;\n\tentropy: number;\n}\n\nexport interface ToolPromotionDecision {\n\t/** The promoted tool, present only when the evidence gate returned \"recommend\". */\n\ttool?: string;\n\tsampleCount: number;\n\tmargin: number;\n\tentropy: number;\n}\n\nfunction candidateFromStats(stats: ToolPerformanceStats): ExpectedUtilityCandidate {\n\treturn {\n\t\ttool: stats.tool,\n\t\t// All candidates here were recorded under the SAME (modelRef, intentClass) bucket — the\n\t\t// write path only ever stores a tool's stats under its own classified intent (see\n\t\t// classifyToolIntent + candidateFor in tool-selection-controller.ts) — so there is no\n\t\t// per-candidate affinity prior left to apply; every candidate is equally \"in scope\".\n\t\tvalue: 1,\n\t\talpha: stats.alpha,\n\t\tbeta: stats.beta,\n\t\tsampleCount: stats.sampleCount,\n\t\tlatencyMs: stats.latencyEwmaMs,\n\t\ttokenEstimate:\n\t\t\tstats.inputTokenEstimateEwma === undefined && stats.outputTokenEstimateEwma === undefined\n\t\t\t\t? undefined\n\t\t\t\t: (stats.inputTokenEstimateEwma ?? 0) + (stats.outputTokenEstimateEwma ?? 0),\n\t};\n}\n\n/**\n * Evidence-gated promotion for one (model, intentClass) bucket: which tool, if any, has become\n * the clear default given its accumulated track record. `statsForIntent` should already be\n * scoped to a single (modelRef, intentClass) — see `ToolPerformanceStore.getStatsForIntent`.\n */\nexport function evaluateToolPromotion(\n\tstatsForIntent: readonly ToolPerformanceStats[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): ToolPromotionDecision {\n\tif (statsForIntent.length === 0) {\n\t\treturn { sampleCount: 0, margin: 0, entropy: 0 };\n\t}\n\tconst decision = decideExpectedUtility(statsForIntent.map(candidateFromStats), weights);\n\tif (decision.disposition !== \"recommend\" || !decision.recommendation) {\n\t\treturn { sampleCount: 0, margin: decision.margin, entropy: decision.entropy };\n\t}\n\tconst winner = decision.ranked.find((candidate) => candidate.tool === decision.recommendation);\n\treturn {\n\t\ttool: decision.recommendation,\n\t\tsampleCount: winner?.sampleCount ?? 0,\n\t\tmargin: decision.margin,\n\t\tentropy: decision.entropy,\n\t};\n}\n\n/**\n * Renders the active hints into a small, stable prompt block. Deliberately omits live numbers\n * (sample counts, rates) that would change every time evidence accumulates — only the SET of\n * (intent -> tool) pairs is included, so the block's text changes only when the promoted tool for\n * an intent actually flips, not on every tool call (system-prompt cache stability).\n */\nexport function formatToolSelectionHints(hints: readonly ToolSelectionHint[]): string | undefined {\n\tif (hints.length === 0) return undefined;\n\tconst lines = [...hints]\n\t\t.sort((left, right) => left.intentClass.localeCompare(right.intentClass))\n\t\t.map((hint) => `- ${hint.intentClass}: \\`${hint.tool}\\` has the established track record for this model.`);\n\treturn [\n\t\t\"Learned tool preferences (evidence-gated observation, not a directive):\",\n\t\t...lines,\n\t\t\"Treat as a shortlist signal only — use judgment for the actual task at hand.\",\n\t].join(\"\\n\");\n}\n"]}
+{"version":3,"file":"promotion.js","sourceRoot":"","sources":["../../../src/core/tool-selection/promotion.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AAEH,OAAO,EACN,gCAAgC,EAChC,qBAAqB,GAGrB,MAAM,uBAAuB,CAAC;AAoB/B,SAAS,kBAAkB,CAAC,KAA2B;IACtD,OAAO;QACN,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,wFAAwF;QACxF,kFAAkF;QAClF,sFAAsF;QACtF,qFAAqF;QACrF,KAAK,EAAE,CAAC;QACR,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,WAAW,EAAE,KAAK,CAAC,WAAW;QAC9B,SAAS,EAAE,KAAK,CAAC,aAAa;QAC9B,aAAa,EACZ,KAAK,CAAC,sBAAsB,KAAK,SAAS,IAAI,KAAK,CAAC,uBAAuB,KAAK,SAAS;YACxF,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,CAAC,KAAK,CAAC,sBAAsB,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,uBAAuB,IAAI,CAAC,CAAC;KAC9E,CAAC;AACH,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,qBAAqB,CACpC,cAA+C,EAC/C,OAAO,GAA2B,gCAAgC;IAElE,IAAI,cAAc,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACjC,OAAO,EAAE,WAAW,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IAClD,CAAC;IACD,MAAM,QAAQ,GAAG,qBAAqB,CAAC,cAAc,CAAC,GAAG,CAAC,kBAAkB,CAAC,EAAE,OAAO,CAAC,CAAC;IACxF,IAAI,QAAQ,CAAC,WAAW,KAAK,WAAW,IAAI,CAAC,QAAQ,CAAC,cAAc,EAAE,CAAC;QACtE,OAAO,EAAE,WAAW,EAAE,CAAC,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,OAAO,EAAE,QAAQ,CAAC,OAAO,EAAE,CAAC;IAC/E,CAAC;IACD,MAAM,MAAM,GAAG,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,IAAI,KAAK,QAAQ,CAAC,cAAc,CAAC,CAAC;IAC/F,OAAO;QACN,IAAI,EAAE,QAAQ,CAAC,cAAc;QAC7B,WAAW,EAAE,MAAM,EAAE,WAAW,IAAI,CAAC;QACrC,MAAM,EAAE,QAAQ,CAAC,MAAM;QACvB,OAAO,EAAE,QAAQ,CAAC,OAAO;KACzB,CAAC;AACH,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,wBAAwB,CAAC,KAAmC;IAC3E,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IACzC,MAAM,KAAK,GAAG,CAAC,GAAG,KAAK,CAAC;SACtB,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,WAAW,CAAC,aAAa,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;SACxE,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,KAAK,IAAI,CAAC,WAAW,OAAO,IAAI,CAAC,IAAI,qDAAqD,CAAC,CAAC;IAC5G,OAAO;QACN,yEAAyE;QACzE,GAAG,KAAK;QACR,8EAA8E;KAC9E,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACd,CAAC","sourcesContent":["/**\n * Evidence-gated tool-selection promotion (closing the observe/promote loop).\n *\n * `tool-selection-controller.ts` records, per tool call, a full expected-utility decision\n * (`ToolSelectionController.begin()`), but that per-call decision is scoped to whichever\n * tool the model already chose — only the chosen tool is ever eligible to be \"recommended\"\n * (see `candidateFor` in tool-selection-controller.ts), so it can validate a choice but never\n * suggest a different one. This module is the SEPARATE, aggregate read-side: given the\n * per-(model,intent,tool) track record accumulated in `ToolPerformanceStore` over many calls,\n * decide whether one tool has become the clear, evidence-backed default for a (model, intent)\n * pair — and if so, render a compact, cache-stable prompt hint.\n *\n * Reuses the exact same expected-utility gate (`decideExpectedUtility`) as call-time decisions,\n * so promotion and the call-time recommend/shortlist/abstain gate share one threshold\n * vocabulary: positive best-utility, a minimum margin over the runner-up, bounded entropy, and a\n * minimum sample count. A tool is promoted only when the gate returns \"recommend\"; ambiguous\n * (high-entropy/low-margin) or thin-evidence buckets never surface a hint — they stay silent,\n * which is the \"shortlist on high entropy, never a false-confident pick\" behavior.\n *\n * Purely functional: no store/env access, so it is trivial to unit test and never mutates state.\n * The stateful pieces (kill switches, reading the store, formatting into the live system prompt)\n * live in `tool-selection-controller.ts` and `system-prompt-builder.ts`.\n */\n\nimport {\n\tDEFAULT_EXPECTED_UTILITY_WEIGHTS,\n\tdecideExpectedUtility,\n\ttype ExpectedUtilityCandidate,\n\ttype ExpectedUtilityWeights,\n} from \"./expected-utility.ts\";\nimport type { ToolPerformanceStats, ToolSelectionIntentClass } from \"./tool-performance-store.ts\";\n\nexport interface ToolSelectionHint {\n\tmodelRef: string;\n\tintentClass: ToolSelectionIntentClass;\n\ttool: string;\n\tsampleCount: number;\n\tmargin: number;\n\tentropy: number;\n}\n\nexport interface ToolPromotionDecision {\n\t/** The promoted tool, present only when the evidence gate returned \"recommend\". */\n\ttool?: string;\n\tsampleCount: number;\n\tmargin: number;\n\tentropy: number;\n}\n\nfunction candidateFromStats(stats: ToolPerformanceStats): ExpectedUtilityCandidate {\n\treturn {\n\t\ttool: stats.tool,\n\t\t// All candidates here were recorded under the SAME (modelRef, intentClass) bucket — the\n\t\t// write path only ever stores a tool's stats under its own classified intent (see\n\t\t// classifyToolIntent + candidateFor in tool-selection-controller.ts) — so there is no\n\t\t// per-candidate affinity prior left to apply; every candidate is equally \"in scope\".\n\t\tvalue: 1,\n\t\talpha: stats.alpha,\n\t\tbeta: stats.beta,\n\t\tsampleCount: stats.sampleCount,\n\t\tlatencyMs: stats.latencyEwmaMs,\n\t\ttokenEstimate:\n\t\t\tstats.inputTokenEstimateEwma === undefined && stats.outputTokenEstimateEwma === undefined\n\t\t\t\t? undefined\n\t\t\t\t: (stats.inputTokenEstimateEwma ?? 0) + (stats.outputTokenEstimateEwma ?? 0),\n\t};\n}\n\n/**\n * Evidence-gated promotion for one (model, intentClass) bucket: which tool, if any, has become\n * the clear default given its accumulated track record. `statsForIntent` should already be\n * scoped to a single (modelRef, intentClass) — see `ToolPerformanceStore.getStatsForIntent`.\n */\nexport function evaluateToolPromotion(\n\tstatsForIntent: readonly ToolPerformanceStats[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): ToolPromotionDecision {\n\tif (statsForIntent.length === 0) {\n\t\treturn { sampleCount: 0, margin: 0, entropy: 0 };\n\t}\n\tconst decision = decideExpectedUtility(statsForIntent.map(candidateFromStats), weights);\n\tif (decision.disposition !== \"recommend\" || !decision.recommendation) {\n\t\treturn { sampleCount: 0, margin: decision.margin, entropy: decision.entropy };\n\t}\n\tconst winner = decision.ranked.find((candidate) => candidate.tool === decision.recommendation);\n\treturn {\n\t\ttool: decision.recommendation,\n\t\tsampleCount: winner?.sampleCount ?? 0,\n\t\tmargin: decision.margin,\n\t\tentropy: decision.entropy,\n\t};\n}\n\n/**\n * Renders the active hints into a small, stable prompt block. Deliberately omits live numbers\n * (sample counts, rates) that would change every time evidence accumulates — only the SET of\n * (intent -> tool) pairs is included, so the block's text changes only when the promoted tool for\n * an intent actually flips, not on every tool call (system-prompt cache stability).\n */\nexport function formatToolSelectionHints(hints: readonly ToolSelectionHint[]): string | undefined {\n\tif (hints.length === 0) return undefined;\n\tconst lines = [...hints]\n\t\t.sort((left, right) => left.intentClass.localeCompare(right.intentClass))\n\t\t.map((hint) => `- ${hint.intentClass}: \\`${hint.tool}\\` has the established track record for this model.`);\n\treturn [\n\t\t\"Learned tool preferences (evidence-gated observation, not a directive):\",\n\t\t...lines,\n\t\t\"Treat as a shortlist signal only — use judgment for the actual task at hand.\",\n\t].join(\"\\n\");\n}\n"]}

package/dist/core/tool-selection/tool-performance-store.d.ts

@@ -101,6 +101,8 @@ export declare class ToolPerformanceStore {
     }): ToolPerformanceStore;
     private createHostData;
     get(key: ToolPerformanceKey): ToolPerformanceStats;
+    /** One fresh durable snapshot of every per-tool track record for a model. */
+    getStatsForModel(modelRef: string): ToolPerformanceStats[];
     /** Every per-tool track record recorded for a (model,intent) bucket — the promotion.ts input. */
     getStatsForIntent(modelRef: string, intentClass: ToolSelectionIntentClass): ToolPerformanceStats[];
     /** Durable observe-mode agreement for one (model,intent) bucket (see {@link ToolSelectionIntentAgreement}). */

package/dist/core/tool-selection/tool-performance-store.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tool-performance-store.d.ts","sourceRoot":"","sources":["../../../src/core/tool-selection/tool-performance-store.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,KAAK,eAAe,EAAqC,MAAM,+BAA+B,CAAC;AAUxG,MAAM,MAAM,wBAAwB,GAAG,MAAM,GAAG,QAAQ,GAAG,SAAS,GAAG,OAAO,GAAG,UAAU,GAAG,SAAS,GAAG,OAAO,CAAC;AAElH,MAAM,WAAW,kBAAkB;IAClC,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,wBAAwB,CAAC;IACtC,IAAI,EAAE,MAAM,CAAC;CACb;AAED,MAAM,WAAW,oBAAqB,SAAQ,kBAAkB;IAC/D,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,sBAAsB,CAAC,EAAE,MAAM,CAAC;IAChC,sBAAsB,CAAC,EAAE,MAAM,CAAC;IAChC,uBAAuB,CAAC,EAAE,MAAM,CAAC;IACjC,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,wBAAwB;IACxC,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,wBAAwB,CAAC;IACtC,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,OAAO,CAAC;IACnB,SAAS,EAAE,OAAO,CAAC;IACnB,WAAW,EAAE,WAAW,GAAG,WAAW,GAAG,SAAS,CAAC;IACnD,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,mGAAmG;IACnG,MAAM,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACtE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,mBAAmB,CAAC,EAAE,MAAM,CAAC;CAC7B;AAED,MAAM,WAAW,oBAAoB;IACpC,iBAAiB,EAAE,MAAM,CAAC;IAC1B,kBAAkB,EAAE,MAAM,CAAC;IAC3B,uBAAuB,EAAE,MAAM,CAAC;IAChC,mBAAmB,EAAE,MAAM,CAAC;IAC5B,0BAA0B,EAAE,MAAM,CAAC;IACnC,cAAc,EAAE,MAAM,CAAC;IACvB,eAAe,EAAE,MAAM,CAAC;IACxB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,yBAAyB,CAAC,EAAE,MAAM,CAAC;IACnC,0BAA0B,CAAC,EAAE,MAAM,CAAC;CACpC;AAED;;;;;;;;;;GAUG;AACH,MAAM,WAAW,4BAA4B;IAC5C,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,wBAAwB,CAAC;IACtC,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,EAAE,MAAM,CAAC;IACvB,qBAAqB,EAAE,MAAM,CAAC;IAC9B,wBAAwB,EAAE,MAAM,CAAC;IACjC,aAAa,EAAE,MAAM,CAAC;CACtB;AASD,MAAM,WAAW,wBAAwB;IACxC,GAAG,EAAE,kBAAkB,CAAC;IACxB,OAAO,EAAE,OAAO,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,SAAS,EAAE,IAAI,CAAC,wBAAwB,EAAE,IAAI,GAAG,UAAU,GAAG,aAAa,GAAG,YAAY,GAAG,WAAW,CAAC,CAAC;IAC1G,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ;;;;;;OAMG;IACH,oBAAoB,CAAC,EAAE,OAAO,CAAC;CAC/B;AAgLD,qBAAa,oBAAoB;IAChC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAA0C;IAElE,YAAY,QAAQ,EAAE,MAAM,EAAE,OAAO,GAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAC;QAAC,QAAQ,CAAC,EAAE,OAAO,CAAA;KAAO,EAQtG;IAED,MAAM,CAAC,WAAW,CACjB,QAAQ,EAAE,MAAM,EAChB,OAAO,GAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAC;QAAC,QAAQ,CAAC,EAAE,OAAO,CAAA;KAAO,GACvE,oBAAoB,CAEtB;IAED,OAAO,CAAC,cAAc;IAItB,GAAG,CAAC,GAAG,EAAE,kBAAkB,GAAG,oBAAoB,CAIjD;IAED,mGAAiG;IACjG,iBAAiB,CAAC,QAAQ,EAAE,MAAM,EAAE,WAAW,EAAE,wBAAwB,GAAG,oBAAoB,EAAE,CAMjG;IAED,+GAA+G;IAC/G,kBAAkB,CAAC,QAAQ,EAAE,MAAM,EAAE,WAAW,EAAE,wBAAwB,GAAG,4BAA4B,CAIxG;IAED,sGAAoG;IACpG,sBAAsB,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,4BAA4B,EAAE,CAKxE;IAED,gBAAgB,CACf,GAAG,EAAE,kBAAkB,EACvB,OAAO,EAAE,UAAU,GAAG,SAAS,EAC/B,EAAE,SAA2B,GAC3B,oBAAoB,CAiBtB;IAED,eAAe,CAAC,WAAW,EAAE,wBAAwB,GAAG,oBAAoB,CAqE3E;IAED,UAAU,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,oBAAoB,CA6BlD;IAED,eAAe,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,wBAAwB,EAAE,CAS7D;CACD","sourcesContent":["import { stateFile } from \"../agent-paths.ts\";\nimport { type HostFingerprint, HostStateStore, isHostFingerprint } from \"../models/host-state-store.ts\";\nimport { isWorkerSession } from \"../session-role.ts\";\nimport { isRecordObject } from \"../util/value-guards.ts\";\n\nconst STORE_VERSION = 1;\nconst MAX_STATS_PER_HOST = 500;\nconst MAX_OBSERVATIONS_PER_HOST = 1_000;\nconst MAX_INTENT_AGREEMENT_PER_HOST = 500;\nconst EWMA_ALPHA = 0.25;\n\nexport type ToolSelectionIntentClass = \"read\" | \"search\" | \"execute\" | \"write\" | \"retrieve\" | \"explain\" | \"other\";\n\nexport interface ToolPerformanceKey {\n\tmodelRef: string;\n\tintentClass: ToolSelectionIntentClass;\n\ttool: string;\n}\n\nexport interface ToolPerformanceStats extends ToolPerformanceKey {\n\talpha: number;\n\tbeta: number;\n\tsampleCount: number;\n\tlatencyEwmaMs?: number;\n\tlatencyDeviationEwmaMs?: number;\n\tinputTokenEstimateEwma?: number;\n\toutputTokenEstimateEwma?: number;\n\trepairCount: number;\n\tbounceCount: number;\n\tfailureCount: number;\n\tlastUsedAt: string;\n}\n\nexport interface ToolSelectionObservation {\n\tat: string;\n\tmodelRef: string;\n\tintentClass: ToolSelectionIntentClass;\n\tactualTool: string;\n\tfirstTool: boolean;\n\tsucceeded: boolean;\n\tdisposition: \"recommend\" | \"shortlist\" | \"abstain\";\n\trecommendation?: string;\n\tshortlist: string[];\n\tentropy: number;\n\tmargin: number;\n\t/** Redacted ranking: names and numeric scores only, never prompts, arguments, paths, or output. */\n\tranked: Array<{ tool: string; utility: number; probability: number }>;\n\tlatencyMs?: number;\n\tinputTokenEstimate?: number;\n\toutputTokenEstimate?: number;\n}\n\nexport interface ToolSelectionMetrics {\n\tfirstToolAttempts: number;\n\tfirstToolSuccesses: number;\n\twrongToolOrFailureCount: number;\n\trecommendationCount: number;\n\trecommendationMatchedCount: number;\n\tshortlistCount: number;\n\tabstentionCount: number;\n\taverageLatencyMs?: number;\n\taverageInputTokenEstimate?: number;\n\taverageOutputTokenEstimate?: number;\n}\n\n/**\n * Durable, per-(model,intent) aggregate of the observe-mode loop: does the raw expected-utility\n * ranking's top pick (`ToolSelectionObservation.ranked[0]`) match what the model actually called\n * (`ToolSelectionObservation.actualTool`)? Tracked separately from the capped, rolling\n * `observations` log so evidence for a given (model,intent) pair survives that log's trimming.\n * `hintActive*` fields are the SAME agreement measure, but restricted to calls made while an\n * evidence-gated promotion hint (see promotion.ts) was active for this bucket — the hint's own\n * efficacy trace: it never gates activation directly (that is always\n * recomputed live from `ToolPerformanceStats`), but it is the durable evidence a report can show\n * for \"is the hint still earning its keep\".\n */\nexport interface ToolSelectionIntentAgreement {\n\tmodelRef: string;\n\tintentClass: ToolSelectionIntentClass;\n\tsampleCount: number;\n\tagreementCount: number;\n\thintActiveSampleCount: number;\n\thintActiveAgreementCount: number;\n\tlastUpdatedAt: string;\n}\n\ninterface HostToolPerformanceData {\n\thost: HostFingerprint;\n\tstats: Record<string, ToolPerformanceStats>;\n\tobservations: ToolSelectionObservation[];\n\tintentAgreement: Record<string, ToolSelectionIntentAgreement>;\n}\n\nexport interface ToolExecutionObservation {\n\tkey: ToolPerformanceKey;\n\tsuccess: boolean;\n\tlatencyMs: number;\n\tinputTokenEstimate?: number;\n\toutputTokenEstimate?: number;\n\tselection: Omit<ToolSelectionObservation, \"at\" | \"modelRef\" | \"intentClass\" | \"actualTool\" | \"succeeded\">;\n\tat?: string;\n\t/**\n\t * Whether an evidence-gated promotion hint (promotion.ts) was already active for this\n\t * (model,intent) bucket BEFORE this call was recorded — captured by the controller at\n\t * `begin()` time, so it reflects evidence up to but not including this observation. Used only\n\t * to bucket the durable agreement stats (`ToolSelectionIntentAgreement.hintActive*`); never\n\t * changes what gets recorded, only how it is split for the efficacy report.\n\t */\n\thintActiveAtCallTime?: boolean;\n}\n\nfunction statKey(key: ToolPerformanceKey): string {\n\treturn `${key.modelRef}\\0${key.intentClass}\\0${key.tool}`;\n}\n\nfunction intentAgreementKey(modelRef: string, intentClass: ToolSelectionIntentClass): string {\n\treturn `${modelRef}\\0${intentClass}`;\n}\n\nfunction isIntentClass(value: unknown): value is ToolSelectionIntentClass {\n\treturn (\n\t\tvalue === \"read\" ||\n\t\tvalue === \"search\" ||\n\t\tvalue === \"execute\" ||\n\t\tvalue === \"write\" ||\n\t\tvalue === \"retrieve\" ||\n\t\tvalue === \"explain\" ||\n\t\tvalue === \"other\"\n\t);\n}\n\nfunction isStats(value: unknown): value is ToolPerformanceStats {\n\treturn (\n\t\tisRecordObject(value) &&\n\t\ttypeof value.modelRef === \"string\" &&\n\t\tisIntentClass(value.intentClass) &&\n\t\ttypeof value.tool === \"string\" &&\n\t\ttypeof value.alpha === \"number\" &&\n\t\ttypeof value.beta === \"number\" &&\n\t\ttypeof value.sampleCount === \"number\" &&\n\t\ttypeof value.repairCount === \"number\" &&\n\t\ttypeof value.bounceCount === \"number\" &&\n\t\ttypeof value.failureCount === \"number\" &&\n\t\ttypeof value.lastUsedAt === \"string\"\n\t);\n}\n\nfunction isObservation(value: unknown): value is ToolSelectionObservation {\n\treturn (\n\t\tisRecordObject(value) &&\n\t\ttypeof value.at === \"string\" &&\n\t\ttypeof value.modelRef === \"string\" &&\n\t\tisIntentClass(value.intentClass) &&\n\t\ttypeof value.actualTool === \"string\" &&\n\t\ttypeof value.firstTool === \"boolean\" &&\n\t\ttypeof value.succeeded === \"boolean\" &&\n\t\t(value.disposition === \"recommend\" || value.disposition === \"shortlist\" || value.disposition === \"abstain\") &&\n\t\tArray.isArray(value.shortlist) &&\n\t\tvalue.shortlist.every((tool) => typeof tool === \"string\") &&\n\t\ttypeof value.entropy === \"number\" &&\n\t\ttypeof value.margin === \"number\" &&\n\t\tArray.isArray(value.ranked) &&\n\t\tvalue.ranked.every(\n\t\t\t(entry) =>\n\t\t\t\tisRecordObject(entry) &&\n\t\t\t\ttypeof entry.tool === \"string\" &&\n\t\t\t\ttypeof entry.utility === \"number\" &&\n\t\t\t\ttypeof entry.probability === \"number\",\n\t\t) &&\n\t\t(value.latencyMs === undefined || typeof value.latencyMs === \"number\") &&\n\t\t(value.inputTokenEstimate === undefined || typeof value.inputTokenEstimate === \"number\") &&\n\t\t(value.outputTokenEstimate === undefined || typeof value.outputTokenEstimate === \"number\")\n\t);\n}\n\nfunction isIntentAgreement(value: unknown): value is ToolSelectionIntentAgreement {\n\treturn (\n\t\tisRecordObject(value) &&\n\t\ttypeof value.modelRef === \"string\" &&\n\t\tisIntentClass(value.intentClass) &&\n\t\ttypeof value.sampleCount === \"number\" &&\n\t\ttypeof value.agreementCount === \"number\" &&\n\t\ttypeof value.hintActiveSampleCount === \"number\" &&\n\t\ttypeof value.hintActiveAgreementCount === \"number\" &&\n\t\ttypeof value.lastUpdatedAt === \"string\"\n\t);\n}\n\nfunction emptyStats(key: ToolPerformanceKey, at: string): ToolPerformanceStats {\n\treturn {\n\t\t...key,\n\t\talpha: 1,\n\t\tbeta: 1,\n\t\tsampleCount: 0,\n\t\trepairCount: 0,\n\t\tbounceCount: 0,\n\t\tfailureCount: 0,\n\t\tlastUsedAt: at,\n\t};\n}\n\nfunction emptyIntentAgreement(\n\tmodelRef: string,\n\tintentClass: ToolSelectionIntentClass,\n\tat: string,\n): ToolSelectionIntentAgreement {\n\treturn {\n\t\tmodelRef,\n\t\tintentClass,\n\t\tsampleCount: 0,\n\t\tagreementCount: 0,\n\t\thintActiveSampleCount: 0,\n\t\thintActiveAgreementCount: 0,\n\t\tlastUpdatedAt: at,\n\t};\n}\n\nfunction finiteNonNegative(value: number | undefined): number | undefined {\n\treturn typeof value === \"number\" && Number.isFinite(value) && value >= 0 ? value : undefined;\n}\n\nfunction updateEwma(previous: number | undefined, next: number | undefined): number | undefined {\n\tif (next === undefined) return previous;\n\treturn previous === undefined ? next : previous + EWMA_ALPHA * (next - previous);\n}\n\nfunction updateDeviation(\n\tpreviousValue: number | undefined,\n\tpreviousDeviation: number | undefined,\n\tnext: number | undefined,\n): number | undefined {\n\tif (next === undefined) return previousDeviation;\n\tconst deviation = previousValue === undefined ? 0 : Math.abs(next - previousValue);\n\treturn updateEwma(previousDeviation, deviation);\n}\n\nfunction parseHost(value: unknown, hostId: string): HostToolPerformanceData | undefined {\n\tif (\n\t\t!isRecordObject(value) ||\n\t\t!isHostFingerprint(value.host) ||\n\t\tvalue.host.id !== hostId ||\n\t\t!isRecordObject(value.stats) ||\n\t\t!Array.isArray(value.observations)\n\t)\n\t\treturn undefined;\n\tconst host = value.host;\n\t// intentAgreement is a purely additive field (older store files predate it) — tolerate absence\n\t// rather than bumping STORE_VERSION, same as any other backward-compatible default-empty field.\n\tconst intentAgreementRaw = isRecordObject(value.intentAgreement) ? value.intentAgreement : {};\n\treturn {\n\t\thost,\n\t\tstats: Object.fromEntries(\n\t\t\tObject.entries(value.stats).filter((entry): entry is [string, ToolPerformanceStats] => isStats(entry[1])),\n\t\t),\n\t\tobservations: value.observations.filter(isObservation),\n\t\tintentAgreement: Object.fromEntries(\n\t\t\tObject.entries(intentAgreementRaw).filter((entry): entry is [string, ToolSelectionIntentAgreement] =>\n\t\t\t\tisIntentAgreement(entry[1]),\n\t\t\t),\n\t\t),\n\t};\n}\n\nfunction trimStats(stats: Record<string, ToolPerformanceStats>): Record<string, ToolPerformanceStats> {\n\tconst entries = Object.entries(stats);\n\tif (entries.length <= MAX_STATS_PER_HOST) return stats;\n\treturn Object.fromEntries(\n\t\tentries\n\t\t\t.sort(([, left], [, right]) => Date.parse(right.lastUsedAt) - Date.parse(left.lastUsedAt))\n\t\t\t.slice(0, MAX_STATS_PER_HOST),\n\t);\n}\n\nfunction trimIntentAgreement(\n\trecords: Record<string, ToolSelectionIntentAgreement>,\n): Record<string, ToolSelectionIntentAgreement> {\n\tconst entries = Object.entries(records);\n\tif (entries.length <= MAX_INTENT_AGREEMENT_PER_HOST) return records;\n\treturn Object.fromEntries(\n\t\tentries\n\t\t\t.sort(([, left], [, right]) => Date.parse(right.lastUpdatedAt) - Date.parse(left.lastUpdatedAt))\n\t\t\t.slice(0, MAX_INTENT_AGREEMENT_PER_HOST),\n\t);\n}\n\nexport class ToolPerformanceStore {\n\tprivate readonly storage: HostStateStore<HostToolPerformanceData>;\n\n\tconstructor(filePath: string, options: { fingerprint?: () => HostFingerprint; readOnly?: boolean } = {}) {\n\t\tthis.storage = new HostStateStore({\n\t\t\tfilePath,\n\t\t\tversion: STORE_VERSION,\n\t\t\tfingerprint: options.fingerprint,\n\t\t\treadOnly: options.readOnly ?? isWorkerSession(),\n\t\t\tparseHost,\n\t\t});\n\t}\n\n\tstatic forAgentDir(\n\t\tagentDir: string,\n\t\toptions: { fingerprint?: () => HostFingerprint; readOnly?: boolean } = {},\n\t): ToolPerformanceStore {\n\t\treturn new ToolPerformanceStore(stateFile(agentDir, \"tool-performance.json\"), options);\n\t}\n\n\tprivate createHostData(host: HostFingerprint): HostToolPerformanceData {\n\t\treturn { host, stats: {}, observations: [], intentAgreement: {} };\n\t}\n\n\tget(key: ToolPerformanceKey): ToolPerformanceStats {\n\t\tconst host = this.storage.getHost();\n\t\tconst stats = host?.stats[statKey(key)];\n\t\treturn stats ? { ...stats } : emptyStats(key, new Date(0).toISOString());\n\t}\n\n\t/** Every per-tool track record recorded for a (model,intent) bucket — the promotion.ts input. */\n\tgetStatsForIntent(modelRef: string, intentClass: ToolSelectionIntentClass): ToolPerformanceStats[] {\n\t\tconst host = this.storage.getHost();\n\t\tif (!host) return [];\n\t\treturn Object.values(host.stats)\n\t\t\t.filter((stats) => stats.modelRef === modelRef && stats.intentClass === intentClass)\n\t\t\t.map((stats) => ({ ...stats }));\n\t}\n\n\t/** Durable observe-mode agreement for one (model,intent) bucket (see {@link ToolSelectionIntentAgreement}). */\n\tgetIntentAgreement(modelRef: string, intentClass: ToolSelectionIntentClass): ToolSelectionIntentAgreement {\n\t\tconst host = this.storage.getHost();\n\t\tconst record = host?.intentAgreement[intentAgreementKey(modelRef, intentClass)];\n\t\treturn record ? { ...record } : emptyIntentAgreement(modelRef, intentClass, new Date(0).toISOString());\n\t}\n\n\t/** All recorded (model,intent) agreement buckets, optionally scoped to one model — report input. */\n\tgetAllIntentAgreements(modelRef?: string): ToolSelectionIntentAgreement[] {\n\t\tconst host = this.storage.getHost();\n\t\treturn Object.values(host?.intentAgreement ?? {})\n\t\t\t.filter((record) => modelRef === undefined || record.modelRef === modelRef)\n\t\t\t.map((record) => ({ ...record }));\n\t}\n\n\trecordValidation(\n\t\tkey: ToolPerformanceKey,\n\t\toutcome: \"repaired\" | \"bounced\",\n\t\tat = new Date().toISOString(),\n\t): ToolPerformanceStats {\n\t\treturn this.storage.mutateCurrentHost(\n\t\t\t(host) => this.createHostData(host),\n\t\t\t(host) => {\n\t\t\t\tconst storageKey = statKey(key);\n\t\t\t\tconst current = host.stats[storageKey] ?? emptyStats(key, at);\n\t\t\t\tconst next: ToolPerformanceStats = {\n\t\t\t\t\t...current,\n\t\t\t\t\trepairCount: current.repairCount + (outcome === \"repaired\" ? 1 : 0),\n\t\t\t\t\tbounceCount: current.bounceCount + (outcome === \"bounced\" ? 1 : 0),\n\t\t\t\t\tlastUsedAt: at,\n\t\t\t\t};\n\t\t\t\thost.stats[storageKey] = next;\n\t\t\t\thost.stats = trimStats(host.stats);\n\t\t\t\treturn { result: { ...next }, changed: true };\n\t\t\t},\n\t\t);\n\t}\n\n\trecordExecution(observation: ToolExecutionObservation): ToolPerformanceStats {\n\t\tconst at = observation.at ?? new Date().toISOString();\n\t\treturn this.storage.mutateCurrentHost(\n\t\t\t(host) => this.createHostData(host),\n\t\t\t(host) => {\n\t\t\t\tconst storageKey = statKey(observation.key);\n\t\t\t\tconst current = host.stats[storageKey] ?? emptyStats(observation.key, at);\n\t\t\t\tconst latencyMs = finiteNonNegative(observation.latencyMs);\n\t\t\t\tconst inputTokenEstimate = finiteNonNegative(observation.inputTokenEstimate);\n\t\t\t\tconst outputTokenEstimate = finiteNonNegative(observation.outputTokenEstimate);\n\t\t\t\tconst next: ToolPerformanceStats = {\n\t\t\t\t\t...current,\n\t\t\t\t\talpha: current.alpha + (observation.success ? 1 : 0),\n\t\t\t\t\tbeta: current.beta + (observation.success ? 0 : 1),\n\t\t\t\t\tsampleCount: current.sampleCount + 1,\n\t\t\t\t\tlatencyEwmaMs: updateEwma(current.latencyEwmaMs, latencyMs),\n\t\t\t\t\tlatencyDeviationEwmaMs: updateDeviation(\n\t\t\t\t\t\tcurrent.latencyEwmaMs,\n\t\t\t\t\t\tcurrent.latencyDeviationEwmaMs,\n\t\t\t\t\t\tlatencyMs,\n\t\t\t\t\t),\n\t\t\t\t\tinputTokenEstimateEwma: updateEwma(current.inputTokenEstimateEwma, inputTokenEstimate),\n\t\t\t\t\toutputTokenEstimateEwma: updateEwma(current.outputTokenEstimateEwma, outputTokenEstimate),\n\t\t\t\t\tfailureCount: current.failureCount + (observation.success ? 0 : 1),\n\t\t\t\t\tlastUsedAt: at,\n\t\t\t\t};\n\t\t\t\thost.stats[storageKey] = next;\n\t\t\t\thost.stats = trimStats(host.stats);\n\t\t\t\thost.observations.push({\n\t\t\t\t\t...observation.selection,\n\t\t\t\t\tat,\n\t\t\t\t\tmodelRef: observation.key.modelRef,\n\t\t\t\t\tintentClass: observation.key.intentClass,\n\t\t\t\t\tactualTool: observation.key.tool,\n\t\t\t\t\tsucceeded: observation.success,\n\t\t\t\t\tranked: observation.selection.ranked.slice(0, 6),\n\t\t\t\t\tshortlist: observation.selection.shortlist.slice(0, 3),\n\t\t\t\t\tlatencyMs,\n\t\t\t\t\tinputTokenEstimate,\n\t\t\t\t\toutputTokenEstimate,\n\t\t\t\t});\n\t\t\t\tif (host.observations.length > MAX_OBSERVATIONS_PER_HOST) {\n\t\t\t\t\thost.observations = host.observations.slice(-MAX_OBSERVATIONS_PER_HOST);\n\t\t\t\t}\n\n\t\t\t\t// Observe-mode agreement: did the RAW ranking's top pick (before any actual-tool-only\n\t\t\t\t// eligibility gating) match what the model actually called? Recorded durably per\n\t\t\t\t// (model,intent), separate from the capped `observations` log above, so it survives trimming.\n\t\t\t\tconst predictedBest = observation.selection.ranked[0]?.tool;\n\t\t\t\tconst agreed = predictedBest !== undefined && predictedBest === observation.key.tool;\n\t\t\t\tconst agreementKey = intentAgreementKey(observation.key.modelRef, observation.key.intentClass);\n\t\t\t\tconst currentAgreement =\n\t\t\t\t\thost.intentAgreement[agreementKey] ??\n\t\t\t\t\temptyIntentAgreement(observation.key.modelRef, observation.key.intentClass, at);\n\t\t\t\thost.intentAgreement[agreementKey] = {\n\t\t\t\t\t...currentAgreement,\n\t\t\t\t\tsampleCount: currentAgreement.sampleCount + 1,\n\t\t\t\t\tagreementCount: currentAgreement.agreementCount + (agreed ? 1 : 0),\n\t\t\t\t\thintActiveSampleCount:\n\t\t\t\t\t\tcurrentAgreement.hintActiveSampleCount + (observation.hintActiveAtCallTime ? 1 : 0),\n\t\t\t\t\thintActiveAgreementCount:\n\t\t\t\t\t\tcurrentAgreement.hintActiveAgreementCount + (observation.hintActiveAtCallTime && agreed ? 1 : 0),\n\t\t\t\t\tlastUpdatedAt: at,\n\t\t\t\t};\n\t\t\t\thost.intentAgreement = trimIntentAgreement(host.intentAgreement);\n\n\t\t\t\treturn { result: { ...next }, changed: true };\n\t\t\t},\n\t\t);\n\t}\n\n\tgetMetrics(modelRef?: string): ToolSelectionMetrics {\n\t\tconst host = this.storage.getHost();\n\t\tconst observations = (host?.observations ?? []).filter(\n\t\t\t(observation) => modelRef === undefined || observation.modelRef === modelRef,\n\t\t);\n\t\tconst firstTools = observations.filter((observation) => observation.firstTool);\n\t\tconst recommended = observations.filter((observation) => observation.disposition === \"recommend\");\n\t\tconst average = (values: Array<number | undefined>): number | undefined => {\n\t\t\tconst present = values.filter((value): value is number => value !== undefined);\n\t\t\treturn present.length > 0 ? present.reduce((sum, value) => sum + value, 0) / present.length : undefined;\n\t\t};\n\t\treturn {\n\t\t\tfirstToolAttempts: firstTools.length,\n\t\t\tfirstToolSuccesses: firstTools.filter((observation) => observation.succeeded).length,\n\t\t\twrongToolOrFailureCount: observations.filter(\n\t\t\t\t(observation) =>\n\t\t\t\t\t!observation.succeeded ||\n\t\t\t\t\t(observation.disposition === \"recommend\" && observation.recommendation !== observation.actualTool),\n\t\t\t).length,\n\t\t\trecommendationCount: recommended.length,\n\t\t\trecommendationMatchedCount: recommended.filter(\n\t\t\t\t(observation) => observation.recommendation === observation.actualTool,\n\t\t\t).length,\n\t\t\tshortlistCount: observations.filter((observation) => observation.disposition === \"shortlist\").length,\n\t\t\tabstentionCount: observations.filter((observation) => observation.disposition === \"abstain\").length,\n\t\t\taverageLatencyMs: average(observations.map((observation) => observation.latencyMs)),\n\t\t\taverageInputTokenEstimate: average(observations.map((observation) => observation.inputTokenEstimate)),\n\t\t\taverageOutputTokenEstimate: average(observations.map((observation) => observation.outputTokenEstimate)),\n\t\t};\n\t}\n\n\tgetObservations(modelRef?: string): ToolSelectionObservation[] {\n\t\tconst host = this.storage.getHost();\n\t\treturn (host?.observations ?? [])\n\t\t\t.filter((observation) => modelRef === undefined || observation.modelRef === modelRef)\n\t\t\t.map((observation) => ({\n\t\t\t\t...observation,\n\t\t\t\tshortlist: [...observation.shortlist],\n\t\t\t\tranked: observation.ranked.map((candidate) => ({ ...candidate })),\n\t\t\t}));\n\t}\n}\n"]}
+{"version":3,"file":"tool-performance-store.d.ts","sourceRoot":"","sources":["../../../src/core/tool-selection/tool-performance-store.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,KAAK,eAAe,EAAqC,MAAM,+BAA+B,CAAC;AAYxG,MAAM,MAAM,wBAAwB,GAAG,MAAM,GAAG,QAAQ,GAAG,SAAS,GAAG,OAAO,GAAG,UAAU,GAAG,SAAS,GAAG,OAAO,CAAC;AAElH,MAAM,WAAW,kBAAkB;IAClC,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,wBAAwB,CAAC;IACtC,IAAI,EAAE,MAAM,CAAC;CACb;AAED,MAAM,WAAW,oBAAqB,SAAQ,kBAAkB;IAC/D,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,sBAAsB,CAAC,EAAE,MAAM,CAAC;IAChC,sBAAsB,CAAC,EAAE,MAAM,CAAC;IAChC,uBAAuB,CAAC,EAAE,MAAM,CAAC;IACjC,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,wBAAwB;IACxC,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,wBAAwB,CAAC;IACtC,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,OAAO,CAAC;IACnB,SAAS,EAAE,OAAO,CAAC;IACnB,WAAW,EAAE,WAAW,GAAG,WAAW,GAAG,SAAS,CAAC;IACnD,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,mGAAmG;IACnG,MAAM,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACtE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,mBAAmB,CAAC,EAAE,MAAM,CAAC;CAC7B;AAED,MAAM,WAAW,oBAAoB;IACpC,iBAAiB,EAAE,MAAM,CAAC;IAC1B,kBAAkB,EAAE,MAAM,CAAC;IAC3B,uBAAuB,EAAE,MAAM,CAAC;IAChC,mBAAmB,EAAE,MAAM,CAAC;IAC5B,0BAA0B,EAAE,MAAM,CAAC;IACnC,cAAc,EAAE,MAAM,CAAC;IACvB,eAAe,EAAE,MAAM,CAAC;IACxB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,yBAAyB,CAAC,EAAE,MAAM,CAAC;IACnC,0BAA0B,CAAC,EAAE,MAAM,CAAC;CACpC;AAED;;;;;;;;;;GAUG;AACH,MAAM,WAAW,4BAA4B;IAC5C,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,wBAAwB,CAAC;IACtC,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,EAAE,MAAM,CAAC;IACvB,qBAAqB,EAAE,MAAM,CAAC;IAC9B,wBAAwB,EAAE,MAAM,CAAC;IACjC,aAAa,EAAE,MAAM,CAAC;CACtB;AAUD,MAAM,WAAW,wBAAwB;IACxC,GAAG,EAAE,kBAAkB,CAAC;IACxB,OAAO,EAAE,OAAO,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,SAAS,EAAE,IAAI,CAAC,wBAAwB,EAAE,IAAI,GAAG,UAAU,GAAG,aAAa,GAAG,YAAY,GAAG,WAAW,CAAC,CAAC;IAC1G,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ;;;;;;OAMG;IACH,oBAAoB,CAAC,EAAE,OAAO,CAAC;CAC/B;AA+MD,qBAAa,oBAAoB;IAChC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAA0C;IAElE,YAAY,QAAQ,EAAE,MAAM,EAAE,OAAO,GAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAC;QAAC,QAAQ,CAAC,EAAE,OAAO,CAAA;KAAO,EAQtG;IAED,MAAM,CAAC,WAAW,CACjB,QAAQ,EAAE,MAAM,EAChB,OAAO,GAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAC;QAAC,QAAQ,CAAC,EAAE,OAAO,CAAA;KAAO,GACvE,oBAAoB,CAEtB;IAED,OAAO,CAAC,cAAc;IAItB,GAAG,CAAC,GAAG,EAAE,kBAAkB,GAAG,oBAAoB,CAIjD;IAED,6EAA6E;IAC7E,gBAAgB,CAAC,QAAQ,EAAE,MAAM,GAAG,oBAAoB,EAAE,CAMzD;IAED,iGAAiG;IACjG,iBAAiB,CAAC,QAAQ,EAAE,MAAM,EAAE,WAAW,EAAE,wBAAwB,GAAG,oBAAoB,EAAE,CAMjG;IAED,+GAA+G;IAC/G,kBAAkB,CAAC,QAAQ,EAAE,MAAM,EAAE,WAAW,EAAE,wBAAwB,GAAG,4BAA4B,CAIxG;IAED,oGAAoG;IACpG,sBAAsB,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,4BAA4B,EAAE,CAKxE;IAED,gBAAgB,CACf,GAAG,EAAE,kBAAkB,EACvB,OAAO,EAAE,UAAU,GAAG,SAAS,EAC/B,EAAE,SAA2B,GAC3B,oBAAoB,CAiBtB;IAED,eAAe,CAAC,WAAW,EAAE,wBAAwB,GAAG,oBAAoB,CA6E3E;IAED,UAAU,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,oBAAoB,CA6BlD;IAED,eAAe,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,wBAAwB,EAAE,CAS7D;CACD"}

package/dist/core/tool-selection/tool-performance-store.js

@@ -5,6 +5,8 @@ import { isRecordObject } from "../util/value-guards.js";
 const STORE_VERSION = 1;
 const MAX_STATS_PER_HOST = 500;
 const MAX_OBSERVATIONS_PER_HOST = 1_000;
+const MAX_OBSERVATION_BYTES_PER_HOST = 256 * 1024;
+const TARGET_OBSERVATION_BYTES_PER_HOST = 192 * 1024;
 const MAX_INTENT_AGREEMENT_PER_HOST = 500;
 const EWMA_ALPHA = 0.25;
 function statKey(key) {
@@ -104,6 +106,30 @@ function updateDeviation(previousValue, previousDeviation, next) {
     const deviation = previousValue === undefined ? 0 : Math.abs(next - previousValue);
     return updateEwma(previousDeviation, deviation);
 }
+function encodedObservationBytes(observation) {
+    return Buffer.byteLength(JSON.stringify(observation), "utf8");
+}
+function trimObservations(observations) {
+    let bytes = 2;
+    for (const observation of observations) {
+        bytes += encodedObservationBytes(observation) + (bytes > 2 ? 1 : 0);
+    }
+    if (observations.length <= MAX_OBSERVATIONS_PER_HOST && bytes <= MAX_OBSERVATION_BYTES_PER_HOST) {
+        return { observations, bytes };
+    }
+    const retained = [];
+    bytes = 2;
+    for (let index = observations.length - 1; index >= 0 && retained.length < MAX_OBSERVATIONS_PER_HOST; index -= 1) {
+        const observation = observations[index];
+        const addition = encodedObservationBytes(observation) + (retained.length > 0 ? 1 : 0);
+        if (bytes + addition > TARGET_OBSERVATION_BYTES_PER_HOST)
+            break;
+        retained.push(observation);
+        bytes += addition;
+    }
+    retained.reverse();
+    return { observations: retained, bytes };
+}
 function parseHost(value, hostId) {
     if (!isRecordObject(value) ||
         !isHostFingerprint(value.host) ||
@@ -112,13 +138,15 @@ function parseHost(value, hostId) {
         !Array.isArray(value.observations))
         return undefined;
     const host = value.host;
+    const observations = trimObservations(value.observations.filter(isObservation));
     // intentAgreement is a purely additive field (older store files predate it) — tolerate absence
     // rather than bumping STORE_VERSION, same as any other backward-compatible default-empty field.
     const intentAgreementRaw = isRecordObject(value.intentAgreement) ? value.intentAgreement : {};
     return {
         host,
         stats: Object.fromEntries(Object.entries(value.stats).filter((entry) => isStats(entry[1]))),
-        observations: value.observations.filter(isObservation),
+        observations: observations.observations,
+        observationBytes: observations.bytes,
         intentAgreement: Object.fromEntries(Object.entries(intentAgreementRaw).filter((entry) => isIntentAgreement(entry[1]))),
     };
 }
@@ -139,7 +167,6 @@ function trimIntentAgreement(records) {
         .slice(0, MAX_INTENT_AGREEMENT_PER_HOST));
 }
 export class ToolPerformanceStore {
-    storage;
     constructor(filePath, options = {}) {
         this.storage = new HostStateStore({
             filePath,
@@ -153,13 +180,22 @@ export class ToolPerformanceStore {
         return new ToolPerformanceStore(stateFile(agentDir, "tool-performance.json"), options);
     }
     createHostData(host) {
-        return { host, stats: {}, observations: [], intentAgreement: {} };
+        return { host, stats: {}, observations: [], observationBytes: 2, intentAgreement: {} };
     }
     get(key) {
         const host = this.storage.getHost();
         const stats = host?.stats[statKey(key)];
         return stats ? { ...stats } : emptyStats(key, new Date(0).toISOString());
     }
+    /** One fresh durable snapshot of every per-tool track record for a model. */
+    getStatsForModel(modelRef) {
+        const host = this.storage.getHost();
+        if (!host)
+            return [];
+        return Object.values(host.stats)
+            .filter((stats) => stats.modelRef === modelRef)
+            .map((stats) => ({ ...stats }));
+    }
     /** Every per-tool track record recorded for a (model,intent) bucket — the promotion.ts input. */
     getStatsForIntent(modelRef, intentClass) {
         const host = this.storage.getHost();
@@ -219,7 +255,7 @@ export class ToolPerformanceStore {
             };
             host.stats[storageKey] = next;
             host.stats = trimStats(host.stats);
-            host.observations.push({
+            const selectionObservation = {
                 ...observation.selection,
                 at,
                 modelRef: observation.key.modelRef,
@@ -231,9 +267,15 @@ export class ToolPerformanceStore {
                 latencyMs,
                 inputTokenEstimate,
                 outputTokenEstimate,
-            });
-            if (host.observations.length > MAX_OBSERVATIONS_PER_HOST) {
-                host.observations = host.observations.slice(-MAX_OBSERVATIONS_PER_HOST);
+            };
+            host.observations.push(selectionObservation);
+            host.observationBytes +=
+                encodedObservationBytes(selectionObservation) + (host.observations.length > 1 ? 1 : 0);
+            if (host.observations.length > MAX_OBSERVATIONS_PER_HOST ||
+                host.observationBytes > MAX_OBSERVATION_BYTES_PER_HOST) {
+                const trimmed = trimObservations(host.observations);
+                host.observations = trimmed.observations;
+                host.observationBytes = trimmed.bytes;
             }
             // Observe-mode agreement: did the RAW ranking's top pick (before any actual-tool-only
             // eligibility gating) match what the model actually called? Recorded durably per