deepseek-coder-cli 1.0.0

package/dist/core/realityScore.d.ts

@@ -0,0 +1,159 @@
+/**
+ * Reality Score System - Hallucination Detection via Testability
+ *
+ * Core insight: The more testable/verifiable an edit is, the more "real" it is.
+ * High reality score = grounded in verifiable facts
+ * Low reality score = potentially hallucinated
+ *
+ * This scoring system works even without running full RL tournaments -
+ * it provides instant feedback on whether an edit is likely to be real.
+ */
+/**
+ * Reality score breakdown for an edit operation
+ */
+export interface RealityScore {
+    /** Overall reality score 0-100 (higher = more real/testable) */
+    total: number;
+    /** Individual component scores */
+    components: {
+        /** Does the target file exist? (0-20) */
+        fileExists: number;
+        /** Does the old_string content actually exist in the file? (0-25) */
+        contentMatch: number;
+        /** Will the resulting code be syntactically valid? (0-20) */
+        syntaxValid: number;
+        /** Are imports/dependencies real and resolvable? (0-15) */
+        importsValid: number;
+        /** Can we run a quick test to verify? (0-20) */
+        testable: number;
+    };
+    /** Confidence level based on what we could verify */
+    confidence: 'high' | 'medium' | 'low';
+    /** Potential hallucination indicators */
+    warnings: string[];
+    /** What tests/checks passed */
+    passed: string[];
+    /** What tests/checks failed */
+    failed: string[];
+}
+/**
+ * Edit operation to score
+ */
+export interface EditOperation {
+    filePath: string;
+    oldString: string;
+    newString: string;
+    replaceAll?: boolean;
+}
+/**
+ * Score an edit operation for reality/testability
+ */
+export declare function scoreEditReality(edit: EditOperation): RealityScore;
+/**
+ * Quick reality check - returns true if edit is likely real, false if potentially hallucinated
+ */
+export declare function quickRealityCheck(edit: EditOperation): {
+    isReal: boolean;
+    reason: string;
+};
+/**
+ * Get a human-readable summary of the reality score
+ */
+export declare function formatRealityScore(score: RealityScore): string;
+/**
+ * Score threshold recommendations
+ */
+export declare const REALITY_THRESHOLDS: {
+    /** Edits below this score should be rejected */
+    REJECT: number;
+    /** Edits below this score need human review */
+    REVIEW: number;
+    /** Edits at or above this score can proceed automatically */
+    AUTO_APPROVE: number;
+};
+/**
+ * Edit outcome for RL feedback
+ */
+export interface EditOutcome {
+    /** Timestamp of the edit */
+    timestamp: number;
+    /** File that was edited */
+    filePath: string;
+    /** Reality score at time of edit */
+    realityScore: number;
+    /** Confidence level */
+    confidence: 'high' | 'medium' | 'low';
+    /** Did the edit succeed without issues? */
+    success: boolean;
+    /** Was there a build error after? */
+    buildError?: boolean;
+    /** Was there a test failure after? */
+    testFailure?: boolean;
+    /** Was the edit reverted? */
+    reverted?: boolean;
+    /** Time until outcome was determined (ms) */
+    outcomeDelayMs?: number;
+}
+/**
+ * RL feedback statistics
+ */
+export interface RLFeedbackStats {
+    /** Total edits tracked */
+    totalEdits: number;
+    /** Success rate by confidence level */
+    successRates: {
+        high: {
+            total: number;
+            success: number;
+            rate: number;
+        };
+        medium: {
+            total: number;
+            success: number;
+            rate: number;
+        };
+        low: {
+            total: number;
+            success: number;
+            rate: number;
+        };
+    };
+    /** Average score for successful edits */
+    avgSuccessScore: number;
+    /** Average score for failed edits */
+    avgFailureScore: number;
+    /** Optimal threshold (where success rate crosses 80%) */
+    optimalThreshold: number;
+    /** Score distribution */
+    scoreDistribution: {
+        '0-25': number;
+        '26-50': number;
+        '51-75': number;
+        '76-100': number;
+    };
+}
+/**
+ * Record an edit outcome for RL learning
+ */
+export declare function recordEditOutcome(outcome: EditOutcome): void;
+/**
+ * Record a successful edit
+ */
+export declare function recordSuccessfulEdit(filePath: string, score: RealityScore, outcomeDelayMs?: number): void;
+/**
+ * Record a failed edit (build error, test failure, revert)
+ */
+export declare function recordFailedEdit(filePath: string, score: RealityScore, reason: 'build_error' | 'test_failure' | 'reverted' | 'other'): void;
+/**
+ * Calculate RL feedback statistics
+ */
+export declare function calculateRLStats(): RLFeedbackStats;
+/**
+ * Get RL feedback recommendation for current session
+ */
+export declare function getRLRecommendation(): string;
+/**
+ * Clear RL feedback buffer (for testing or reset)
+ */
+export declare function clearRLFeedback(): void;
+//# sourceMappingURL=realityScore.d.ts.map

package/dist/core/realityScore.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"realityScore.d.ts","sourceRoot":"","sources":["../../src/core/realityScore.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAOH;;GAEG;AACH,MAAM,WAAW,YAAY;IAC3B,gEAAgE;IAChE,KAAK,EAAE,MAAM,CAAC;IAEd,kCAAkC;IAClC,UAAU,EAAE;QACV,yCAAyC;QACzC,UAAU,EAAE,MAAM,CAAC;QACnB,qEAAqE;QACrE,YAAY,EAAE,MAAM,CAAC;QACrB,6DAA6D;QAC7D,WAAW,EAAE,MAAM,CAAC;QACpB,2DAA2D;QAC3D,YAAY,EAAE,MAAM,CAAC;QACrB,gDAAgD;QAChD,QAAQ,EAAE,MAAM,CAAC;KAClB,CAAC;IAEF,qDAAqD;IACrD,UAAU,EAAE,MAAM,GAAG,QAAQ,GAAG,KAAK,CAAC;IAEtC,yCAAyC;IACzC,QAAQ,EAAE,MAAM,EAAE,CAAC;IAEnB,+BAA+B;IAC/B,MAAM,EAAE,MAAM,EAAE,CAAC;IAEjB,+BAA+B;IAC/B,MAAM,EAAE,MAAM,EAAE,CAAC;CAClB;AAED;;GAEG;AACH,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,OAAO,CAAC;CACtB;AAED;;GAEG;AACH,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,aAAa,GAAG,YAAY,CAqDlE;AAmeD;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,aAAa,GAAG;IAAE,MAAM,EAAE,OAAO,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,CA0B1F;AAED;;GAEG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,YAAY,GAAG,MAAM,CAmB9D;AAED;;GAEG;AACH,eAAO,MAAM,kBAAkB;IAC7B,gDAAgD;;IAEhD,+CAA+C;;IAE/C,6DAA6D;;CAE9D,CAAC;AAMF;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B,4BAA4B;IAC5B,SAAS,EAAE,MAAM,CAAC;IAClB,2BAA2B;IAC3B,QAAQ,EAAE,MAAM,CAAC;IACjB,oCAAoC;IACpC,YAAY,EAAE,MAAM,CAAC;IACrB,uBAAuB;IACvB,UAAU,EAAE,MAAM,GAAG,QAAQ,GAAG,KAAK,CAAC;IACtC,2CAA2C;IAC3C,OAAO,EAAE,OAAO,CAAC;IACjB,qCAAqC;IACrC,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,sCAAsC;IACtC,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,6BAA6B;IAC7B,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,6CAA6C;IAC7C,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB;AAED;;GAEG;AACH,MAAM,WAAW,eAAe;IAC9B,0BAA0B;IAC1B,UAAU,EAAE,MAAM,CAAC;IACnB,uCAAuC;IACvC,YAAY,EAAE;QACZ,IAAI,EAAE;YAAE,KAAK,EAAE,MAAM,CAAC;YAAC,OAAO,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAA;SAAE,CAAC;QACvD,MAAM,EAAE;YAAE,KAAK,EAAE,MAAM,CAAC;YAAC,OAAO,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAA;SAAE,CAAC;QACzD,GAAG,EAAE;YAAE,KAAK,EAAE,MAAM,CAAC;YAAC,OAAO,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAA;SAAE,CAAC;KACvD,CAAC;IACF,yCAAyC;IACzC,eAAe,EAAE,MAAM,CAAC;IACxB,qCAAqC;IACrC,eAAe,EAAE,MAAM,CAAC;IACxB,yDAAyD;IACzD,gBAAgB,EAAE,MAAM,CAAC;IACzB,yBAAyB;IACzB,iBAAiB,EAAE;QACjB,MAAM,EAAE,MAAM,CAAC;QACf,OAAO,EAAE,MAAM,CAAC;QAChB,OAAO,EAAE,MAAM,CAAC;QAChB,QAAQ,EAAE,MAAM,CAAC;KAClB,CAAC;CACH;AAMD;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,WAAW,GAAG,IAAI,CAW5D;AAED;;GAEG;AACH,wBAAgB,oBAAoB,CAClC,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,YAAY,EACnB,cAAc,CAAC,EAAE,MAAM,GACtB,IAAI,CASN;AAED;;GAEG;AACH,wBAAgB,gBAAgB,CAC9B,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,YAAY,EACnB,MAAM,EAAE,aAAa,GAAG,cAAc,GAAG,UAAU,GAAG,OAAO,GAC5D,IAAI,CAWN;AAED;;GAEG;AACH,wBAAgB,gBAAgB,IAAI,eAAe,CAsFlD;AAED;;GAEG;AACH,wBAAgB,mBAAmB,IAAI,MAAM,CAsC5C;AAED;;GAEG;AACH,wBAAgB,eAAe,IAAI,IAAI,CAEtC"}