@pennyfarthing/core 10.3.1 → 11.0.0-alpha.0

package/README.md

@@ -1,6 +1,6 @@
 # Pennyfarthing
 
-**v10.3.1** | *The outer loop goes once, the inner loop goes many times.*
+**v10.4.0** | *The outer loop goes once, the inner loop goes many times.*
 
 <img src="pennyfarthing.png" alt="Pennyfarthing Logo" width="75" style="float:left; margin:10px">
 
@@ -31,11 +31,11 @@ A scientific study of how strong personalities affect AI agent behavior:
 - **Benchmarking System** - `/solo`, `/benchmark-control`, `/benchmark` for statistical evaluation
 - **JobFair** - Discovering which characters excel at roles beyond their native specialization
 
-The 97 persona themes (Discworld, Star Trek, Breaking Bad, etc.) are instruments of inquiry, not decoration. Early findings show character expertise often trumps abstract personality scores.
+The 98 persona themes (Discworld, Star Trek, Breaking Bad, etc.) are instruments of inquiry, not decoration. Early findings show character expertise often trumps abstract personality scores.
 
 ### 3. Integration & Tooling
 
-- **Cyclist Visual Terminal** - Electron-based IDE with 15 draggable Dockview panels, agent portraits, tool visualization, and workflow controls
+- **Cyclist Visual Terminal** - Electron-based IDE with 16 draggable Dockview panels, agent portraits, tool visualization, and workflow controls
 - **Jira Integration** - Bidirectional sync, epic auto-creation, sprint velocity
 - **Sprint Management** - Story tracking with `current-sprint.yaml`
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pennyfarthing/core",
-  "version": "10.3.1",
+  "version": "11.0.0-alpha.0",
   "description": "Claude Code agent framework with TDD workflow and persona system",
   "type": "module",
   "bin": {
@@ -30,6 +30,19 @@
   "bugs": {
     "url": "https://github.com/1898andCo/pennyfarthing/issues"
   },
+  "scripts": {
+    "prepublishOnly": "find pennyfarthing-dist/scripts -name '*.sh' ! -name 'find-root.sh' -exec chmod +x {} +",
+    "postinstall": "node scripts/postinstall.cjs",
+    "build": "pnpm --filter @pennyfarthing/shared build && pnpm --filter @pennyfarthing/core build && pnpm --filter @pennyfarthing/cyclist build && ./scripts/generate-skill-docs.sh",
+    "docs": "./scripts/generate-skill-docs.sh",
+    "dev": "pnpm -r --parallel dev",
+    "test": "pnpm -r test",
+    "lint": "eslint 'packages/*/src/**/*.ts' --max-warnings 0",
+    "lint:md": "markdownlint-cli2 'pennyfarthing-dist/**/*.md'",
+    "validate:refs": "node scripts/validate-refs.js",
+    "test:validate-refs": "node --test scripts/validate-refs.test.js",
+    "clean": "pnpm -r clean"
+  },
   "dependencies": {
     "@pennyfarthing/shared": "^10.0.3",
     "chalk": "^5.3.0",
@@ -50,16 +63,5 @@
   "engines": {
     "node": ">=18.0.0"
   },
-  "scripts": {
-    "postinstall": "node scripts/postinstall.cjs",
-    "build": "pnpm --filter @pennyfarthing/shared build && pnpm --filter @pennyfarthing/core build && pnpm --filter @pennyfarthing/cyclist build && ./scripts/generate-skill-docs.sh",
-    "docs": "./scripts/generate-skill-docs.sh",
-    "dev": "pnpm -r --parallel dev",
-    "test": "pnpm -r test",
-    "lint": "eslint 'packages/*/src/**/*.ts' --max-warnings 0",
-    "lint:md": "markdownlint-cli2 'pennyfarthing-dist/**/*.md'",
-    "validate:refs": "node scripts/validate-refs.js",
-    "test:validate-refs": "node --test scripts/validate-refs.test.js",
-    "clean": "pnpm -r clean"
-  }
-}
+  "packageManager": "pnpm@9.0.0"
+}

package/packages/core/dist/benchmark/benchmark-integration.d.ts

@@ -0,0 +1,182 @@
+/**
+ * Benchmark Integration Module
+ *
+ * Story 11-8: Integrate with Benchmark Output
+ * Story 12-6: Update for local results (Epic 12 migration)
+ *
+ * Correlates Chernoff faces and OCEAN profiles with benchmark performance data.
+ * Reads benchmark results from internal/results/ directory (or BENCHMARK_PATH env var).
+ */
+export interface OceanScores {
+    O: number;
+    C: number;
+    E: number;
+    A: number;
+    N: number;
+}
+export interface BenchmarkResult {
+    theme: string;
+    role: string;
+    character: string;
+    scenario: string;
+    mean: number;
+    stdDev: number;
+    delta: number;
+    cohensD?: number;
+    n: number;
+    scores: number[];
+    ocean: OceanScores;
+    face: string;
+    benchmarkMissing?: boolean;
+}
+export interface CorrelationResult {
+    O: {
+        effect: number;
+        direction: 'positive' | 'negative' | 'none';
+    };
+    C: {
+        effect: number;
+        direction: 'positive' | 'negative' | 'none';
+    };
+    E: {
+        effect: number;
+        direction: 'positive' | 'negative' | 'none';
+    };
+    A: {
+        effect: number;
+        direction: 'positive' | 'negative' | 'none';
+    };
+    N: {
+        effect: number;
+        direction: 'positive' | 'negative' | 'none';
+    };
+    strongest: {
+        dimension: keyof OceanScores;
+        effect: number;
+    };
+}
+export interface OptimalProfile {
+    ocean: OceanScores;
+    reasoning: string;
+}
+export interface RoleRecommendations {
+    role: string;
+    topThemes: Array<{
+        theme: string;
+        character: string;
+        score: number;
+        ocean: OceanScores;
+    }>;
+    avoidThemes: Array<{
+        theme: string;
+        character: string;
+        score: number;
+    }>;
+    insight: string;
+}
+export interface PerformerResult {
+    theme: string;
+    character: string;
+    score: number;
+    delta: number;
+    ocean: OceanScores;
+    face: string;
+}
+export interface QueryOptions {
+    scenario?: string;
+    role?: string;
+    filter?: string;
+    ocean?: string;
+    limit?: number;
+    minScore?: number;
+    sortBy?: 'score' | 'delta' | 'name';
+}
+export interface BenchmarkReportResult {
+    markdown: string;
+    data: {
+        performers: PerformerResult[];
+        correlation: CorrelationResult;
+        recommendations: RoleRecommendations;
+        errorCorrelation?: OceanErrorCorrelation;
+    };
+}
+export interface ErrorTypeCell {
+    correlation: number;
+    arrow: string;
+}
+export interface OceanErrorCorrelation {
+    matrix: {
+        [dimension: string]: {
+            reasoning: ErrorTypeCell;
+            planning: ErrorTypeCell;
+            execution: ErrorTypeCell;
+        };
+    };
+    strongest: {
+        dimension: string;
+        errorType: string;
+        correlation: number;
+    };
+}
+export interface JudgeScore {
+    detection_by_type?: {
+        reasoning: number;
+        planning: number;
+        execution: number;
+    };
+}
+export interface BenchmarkResultWithOcean {
+    ocean: OceanScores;
+    mean: number;
+}
+/**
+ * Load benchmark data from thunderdome results
+ */
+export declare function loadBenchmarkData(scenario: string, role: string): BenchmarkResult[];
+/**
+ * Get benchmark result with face visualization attached
+ */
+export declare function getBenchmarkWithFace(theme: string, role: string, scenario: string): BenchmarkResult | null;
+/**
+ * Calculate OCEAN correlation with benchmark performance
+ */
+export declare function calculateOceanCorrelation(scenario: string, role: string): CorrelationResult;
+/**
+ * Generate markdown correlation report
+ */
+export declare function generateCorrelationReport(scenario: string, role: string): string;
+/**
+ * Get optimal OCEAN profile for a role based on benchmark data
+ */
+export declare function getOptimalProfile(role: string): OptimalProfile;
+/**
+ * Get role recommendations (top themes, themes to avoid)
+ */
+export declare function getRoleRecommendations(role: string): RoleRecommendations;
+/**
+ * Find top performers for a scenario/role with optional filters
+ */
+export declare function findTopPerformers(options: QueryOptions): PerformerResult[];
+/**
+ * General query interface for benchmark data
+ */
+export declare function queryBenchmarks(options: QueryOptions): PerformerResult[];
+/**
+ * Calculate OCEAN × error-type correlation matrix
+ * Story 14-5: Correlates OCEAN dimensions with error detection rates
+ */
+export declare function calculateErrorTypeCorrelation(results: BenchmarkResultWithOcean[], judgeScores: JudgeScore[]): OceanErrorCorrelation;
+/**
+ * Generate markdown heat map for OCEAN × error-type correlations
+ * Story 14-5: Produces 5×3 matrix with directional arrows and effect sizes
+ */
+export declare function generateOceanErrorHeatMap(correlation: OceanErrorCorrelation): string;
+/**
+ * Generate complete benchmark report with faces and correlations
+ */
+export declare function generateBenchmarkReport(options: {
+    scenario: string;
+    role: string;
+    includeErrorTypeCorrelation?: boolean;
+}): BenchmarkReportResult;
+//# sourceMappingURL=benchmark-integration.d.ts.map

package/packages/core/dist/benchmark/benchmark-integration.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"benchmark-integration.d.ts","sourceRoot":"","sources":["../../src/benchmark/benchmark-integration.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AA6CH,MAAM,WAAW,WAAW;IAC1B,CAAC,EAAE,MAAM,CAAC;IACV,CAAC,EAAE,MAAM,CAAC;IACV,CAAC,EAAE,MAAM,CAAC;IACV,CAAC,EAAE,MAAM,CAAC;IACV,CAAC,EAAE,MAAM,CAAC;CACX;AAED,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,CAAC,EAAE,MAAM,CAAC;IACV,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,KAAK,EAAE,WAAW,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,gBAAgB,CAAC,EAAE,OAAO,CAAC;CAC5B;AAED,MAAM,WAAW,iBAAiB;IAChC,CAAC,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,UAAU,GAAG,UAAU,GAAG,MAAM,CAAA;KAAE,CAAC;IACnE,CAAC,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,UAAU,GAAG,UAAU,GAAG,MAAM,CAAA;KAAE,CAAC;IACnE,CAAC,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,UAAU,GAAG,UAAU,GAAG,MAAM,CAAA;KAAE,CAAC;IACnE,CAAC,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,UAAU,GAAG,UAAU,GAAG,MAAM,CAAA;KAAE,CAAC;IACnE,CAAC,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,UAAU,GAAG,UAAU,GAAG,MAAM,CAAA;KAAE,CAAC;IACnE,SAAS,EAAE;QAAE,SAAS,EAAE,MAAM,WAAW,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC;CAC7D;AAED,MAAM,WAAW,cAAc;IAC7B,KAAK,EAAE,WAAW,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,KAAK,CAAC;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,WAAW,CAAA;KAAE,CAAC,CAAC;IAC1F,WAAW,EAAE,KAAK,CAAC;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACxE,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,WAAW,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,MAAM,CAAC,EAAE,OAAO,GAAG,OAAO,GAAG,MAAM,CAAC;CACrC;AAED,MAAM,WAAW,qBAAqB;IACpC,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE;QACJ,UAAU,EAAE,eAAe,EAAE,CAAC;QAC9B,WAAW,EAAE,iBAAiB,CAAC;QAC/B,eAAe,EAAE,mBAAmB,CAAC;QACrC,gBAAgB,CAAC,EAAE,qBAAqB,CAAC;KAC1C,CAAC;CACH;AAGD,MAAM,WAAW,aAAa;IAC5B,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,qBAAqB;IACpC,MAAM,EAAE;QACN,CAAC,SAAS,EAAE,MAAM,GAAG;YACnB,SAAS,EAAE,aAAa,CAAC;YACzB,QAAQ,EAAE,aAAa,CAAC;YACxB,SAAS,EAAE,aAAa,CAAC;SAC1B,CAAC;KACH,CAAC;IACF,SAAS,EAAE;QACT,SAAS,EAAE,MAAM,CAAC;QAClB,SAAS,EAAE,MAAM,CAAC;QAClB,WAAW,EAAE,MAAM,CAAC;KACrB,CAAC;CACH;AAED,MAAM,WAAW,UAAU;IACzB,iBAAiB,CAAC,EAAE;QAClB,SAAS,EAAE,MAAM,CAAC;QAClB,QAAQ,EAAE,MAAM,CAAC;QACjB,SAAS,EAAE,MAAM,CAAC;KACnB,CAAC;CACH;AAED,MAAM,WAAW,wBAAwB;IACvC,KAAK,EAAE,WAAW,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;CACd;AA6ND;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,eAAe,EAAE,CA0BnF;AAED;;GAEG;AACH,wBAAgB,oBAAoB,CAClC,KAAK,EAAE,MAAM,EACb,IAAI,EAAE,MAAM,EACZ,QAAQ,EAAE,MAAM,GACf,eAAe,GAAG,IAAI,CA0CxB;AAED;;GAEG;AACH,wBAAgB,yBAAyB,CACvC,QAAQ,EAAE,MAAM,EAChB,IAAI,EAAE,MAAM,GACX,iBAAiB,CAwBnB;AAED;;GAEG;AACH,wBAAgB,yBAAyB,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,MAAM,CA6BhF;AAED;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,GAAG,cAAc,CAkC9D;AAED;;GAEG;AACH,wBAAgB,sBAAsB,CAAC,IAAI,EAAE,MAAM,GAAG,mBAAmB,CA2DxE;AAED;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,YAAY,GAAG,eAAe,EAAE,CAuC1E;AAED;;GAEG;AACH,wBAAgB,eAAe,CAAC,OAAO,EAAE,YAAY,GAAG,eAAe,EAAE,CA6CxE;AAyED;;;GAGG;AACH,wBAAgB,6BAA6B,CAC3C,OAAO,EAAE,wBAAwB,EAAE,EACnC,WAAW,EAAE,UAAU,EAAE,GACxB,qBAAqB,CAiCvB;AAED;;;GAGG;AACH,wBAAgB,yBAAyB,CAAC,WAAW,EAAE,qBAAqB,GAAG,MAAM,CAqCpF;AAED;;GAEG;AACH,wBAAgB,uBAAuB,CAAC,OAAO,EAAE;IAC/C,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,2BAA2B,CAAC,EAAE,OAAO,CAAC;CACvC,GAAG,qBAAqB,CAwExB"}