engine: misc cleanup; fix validation and statistical errors

Signed-off-by: NotAShelf <raf@notashelf.dev>
Change-Id: I07a4b6005a1f8c5176e3a6f3c02d48186a6a6964

src/engine/context.ts

@@ -0,0 +1,288 @@
+import type { AnalysisResult, WebhookEvent } from "../types.js";
+import { readFileSync, writeFileSync, existsSync } from "fs";
+import { getLogger } from "../logger.js";
+
+// Author reputation tracking
+interface AuthorStats {
+  totalContributions: number;
+  positiveImpacts: number;
+  negativeImpacts: number;
+  neutralImpacts: number;
+  averageConfidence: number;
+  lastSeen: string;
+}
+
+// Repository pattern tracking
+interface RepoPatterns {
+  typicalPRSize: number;
+  typicalFileCount: number;
+  commonLabels: string[];
+  activeAuthors: string[];
+  lastUpdated: string;
+}
+
+// Context data structure
+interface ContextData {
+  authors: Record<string, AuthorStats>;
+  repositories: Record<string, RepoPatterns>;
+  globalStats: {
+    totalAnalyses: number;
+    version: number;
+  };
+}
+
+const CONTEXT_VERSION = 1;
+const MAX_ACTIVE_AUTHORS = 100;
+
+function createDefaultContext(): ContextData {
+  return {
+    authors: {},
+    repositories: {},
+    globalStats: {
+      totalAnalyses: 0,
+      version: CONTEXT_VERSION,
+    },
+  };
+}
+
+let contextData: ContextData = createDefaultContext();
+let contextFile: string | null = null;
+
+export function initContext(stateFile?: string): void {
+  contextFile = stateFile || ".troutbot-context.json";
+  loadContext();
+}
+
+function loadContext(): void {
+  if (!contextFile || !existsSync(contextFile)) {
+    contextData = createDefaultContext();
+    return;
+  }
+
+  try {
+    const data = readFileSync(contextFile, "utf-8");
+    const parsed = JSON.parse(data);
+
+    // Validate structure
+    if (
+      parsed &&
+      typeof parsed === "object" &&
+      parsed.globalStats?.version === CONTEXT_VERSION &&
+      typeof parsed.authors === "object" &&
+      parsed.authors !== null &&
+      !Array.isArray(parsed.authors) &&
+      typeof parsed.repositories === "object" &&
+      parsed.repositories !== null &&
+      !Array.isArray(parsed.repositories)
+    ) {
+      contextData = parsed;
+    } else {
+      getLogger().warn("Invalid context format, resetting");
+      contextData = createDefaultContext();
+    }
+  } catch {
+    contextData = createDefaultContext();
+  }
+}
+
+function saveContext(): void {
+  if (!contextFile) return;
+
+  try {
+    writeFileSync(contextFile, JSON.stringify(contextData, null, 2));
+  } catch (err) {
+    getLogger().warn("Failed to save context", err);
+  }
+}
+
+export function updateContext(
+  event: WebhookEvent,
+  result: AnalysisResult,
+): void {
+  const author = event.author;
+  const repo = `${event.owner}/${event.repo}`;
+
+  // Update author stats
+  if (!contextData.authors[author]) {
+    contextData.authors[author] = {
+      totalContributions: 0,
+      positiveImpacts: 0,
+      negativeImpacts: 0,
+      neutralImpacts: 0,
+      averageConfidence: 0,
+      lastSeen: new Date().toISOString(),
+    };
+  }
+
+  const authorStats = contextData.authors[author];
+  authorStats.totalContributions++;
+  authorStats.lastSeen = new Date().toISOString();
+
+  if (result.impact === "positive") authorStats.positiveImpacts++;
+  else if (result.impact === "negative") authorStats.negativeImpacts++;
+  else authorStats.neutralImpacts++;
+
+  // Update running average confidence
+  authorStats.averageConfidence =
+    (authorStats.averageConfidence * (authorStats.totalContributions - 1) +
+      result.confidence) /
+    authorStats.totalContributions;
+
+  // Update repo patterns (simplified)
+  if (!contextData.repositories[repo]) {
+    contextData.repositories[repo] = {
+      typicalPRSize: 0,
+      typicalFileCount: 0,
+      commonLabels: [],
+      activeAuthors: [],
+      lastUpdated: new Date().toISOString(),
+    };
+  }
+
+  const repoPatterns = contextData.repositories[repo];
+  if (!repoPatterns.activeAuthors.includes(author)) {
+    repoPatterns.activeAuthors.push(author);
+    // Enforce max size to prevent unbounded growth
+    if (repoPatterns.activeAuthors.length > MAX_ACTIVE_AUTHORS) {
+      repoPatterns.activeAuthors.shift();
+    }
+  }
+  repoPatterns.lastUpdated = new Date().toISOString();
+
+  // Update global stats
+  contextData.globalStats.totalAnalyses++;
+
+  saveContext();
+}
+
+export function getAuthorReputation(author: string): {
+  isTrusted: boolean;
+  isNew: boolean;
+  reputation: number;
+  history: string;
+} {
+  const stats = contextData.authors[author];
+
+  if (!stats) {
+    return {
+      isTrusted: false,
+      isNew: true,
+      reputation: 0,
+      history: "First-time contributor",
+    };
+  }
+
+  const successRate =
+    stats.totalContributions > 0
+      ? stats.positiveImpacts / stats.totalContributions
+      : 0;
+
+  const reputation = Math.min(
+    1,
+    successRate * 0.6 +
+      (Math.min(stats.totalContributions, 20) / 20) * 0.3 +
+      stats.averageConfidence * 0.1,
+  );
+
+  let history: string;
+  if (stats.totalContributions === 1) {
+    history = "1 contribution";
+  } else if (stats.totalContributions < 5) {
+    history = `${stats.totalContributions} contributions, ${(successRate * 100).toFixed(0)}% positive`;
+  } else {
+    history = `${stats.totalContributions} contributions, ${stats.positiveImpacts}+/${stats.negativeImpacts}-/${stats.neutralImpacts}~`;
+  }
+
+  return {
+    isTrusted: reputation > 0.7 && stats.totalContributions >= 5,
+    isNew: stats.totalContributions < 3,
+    reputation,
+    history,
+  };
+}
+
+export function getRepoContext(
+  owner: string,
+  repo: string,
+): {
+  isActive: boolean;
+  communitySize: number;
+  maturity: string;
+} {
+  const repoKey = `${owner}/${repo}`;
+  const patterns = contextData.repositories[repoKey];
+
+  if (!patterns) {
+    return {
+      isActive: false,
+      communitySize: 0,
+      maturity: "unknown",
+    };
+  }
+
+  const communitySize = patterns.activeAuthors.length;
+  let maturity: string;
+
+  if (contextData.globalStats.totalAnalyses < 10) {
+    maturity = "new";
+  } else if (communitySize < 3) {
+    maturity = "small-team";
+  } else if (communitySize < 10) {
+    maturity = "growing";
+  } else {
+    maturity = "established";
+  }
+
+  return {
+    isActive: true,
+    communitySize,
+    maturity,
+  };
+}
+
+export function getContextualInsights(
+  event: WebhookEvent,
+  backendResults: Record<string, AnalysisResult>,
+): string[] {
+  const insights: string[] = [];
+  const authorRep = getAuthorReputation(event.author);
+  const repoCtx = getRepoContext(event.owner, event.repo);
+
+  // Author-based insights
+  if (authorRep.isNew) {
+    insights.push(
+      `Welcome ${event.author}! This appears to be your first contribution.`,
+    );
+  } else if (authorRep.isTrusted) {
+    insights.push(
+      `${event.author} is a trusted contributor with ${authorRep.history}.`,
+    );
+  } else if (authorRep.reputation < 0.3) {
+    insights.push(
+      `${event.author} has had mixed results recently (${authorRep.history}).`,
+    );
+  }
+
+  // Repo-based insights
+  if (repoCtx.maturity === "new") {
+    insights.push("This repository is still building up analysis history.");
+  }
+
+  // Cross-backend pattern detection
+  const impacts = Object.values(backendResults).map((r) => r.impact);
+  const allPositive = impacts.every((i) => i === "positive");
+  const allNegative = impacts.every((i) => i === "negative");
+  const mixed = new Set(impacts).size > 1;
+
+  if (allPositive && impacts.length >= 2) {
+    insights.push("All analysis backends agree: this looks solid.");
+  } else if (allNegative && impacts.length >= 2) {
+    insights.push("Multiple concerns detected across different dimensions.");
+  } else if (mixed) {
+    insights.push(
+      "Mixed signals - some aspects look good, others need attention.",
+    );
+  }
+
+  return insights;
+}

src/engine/index.ts

@@ -4,26 +4,35 @@ import type {
   EngineConfig,
   Impact,
   WebhookEvent,
-} from '../types.js';
+} from "../types.js";
-import { ChecksBackend } from './checks.js';
+import { ChecksBackend } from "./checks.js";
-import { DiffBackend } from './diff.js';
+import { DiffBackend } from "./diff.js";
-import { QualityBackend } from './quality.js';
+import { QualityBackend } from "./quality.js";
-import { getLogger } from '../logger.js';
+import { getLogger } from "../logger.js";
-
+import {
-const impactToNumeric: Record<Impact, number> = {
+  initContext,
-  positive: 1,
+  updateContext,
-  neutral: 0,
+  getAuthorReputation,
-  negative: -1,
+  getRepoContext,
-};
+} from "./context.js";
 
 interface WeightedBackend {
   backend: EngineBackend;
   weight: number;
 }
 
+interface BackendResult {
+  backend: string;
+  impact: Impact;
+  confidence: number;
+  reasoning: string;
+  weight: number;
+}
+
 export class Engine {
   private backends: WeightedBackend[] = [];
   private confidenceThreshold: number;
+  private contextInitialized = false;
 
   constructor(config: EngineConfig) {
     this.confidenceThreshold = config.confidenceThreshold;
@@ -47,68 +56,453 @@ export class Engine {
     }
 
     if (this.backends.length === 0) {
-      throw new Error('No engine backends enabled');
+      throw new Error("No engine backends enabled");
+    }
+  }
+
+  initializeContext(stateFile?: string): void {
+    if (!this.contextInitialized) {
+      initContext(stateFile);
+      this.contextInitialized = true;
     }
   }
 
   async analyze(event: WebhookEvent): Promise<AnalysisResult> {
     const logger = getLogger();
-    const results = await Promise.all(
+
+    if (!this.contextInitialized) {
+      this.initializeContext();
+    }
+
+    const authorRep = getAuthorReputation(event.author);
+    const repoCtx = getRepoContext(event.owner, event.repo);
+
+    logger.debug(
+      `Analyzing ${event.type} #${event.number} by ${event.author} ` +
+        `(reputation: ${(authorRep.reputation * 100).toFixed(0)}%, repo: ${repoCtx.maturity})`,
+    );
+
+    // Run all backends
+    const backendResults: BackendResult[] = [];
+    const rawResults = await Promise.all(
       this.backends.map(async ({ backend, weight }) => {
         try {
           const result = await backend.analyze(event);
           logger.debug(
-            `Backend "${backend.name}": impact=${result.impact}, confidence=${result.confidence.toFixed(2)}`
+            `Backend "${backend.name}": impact=${result.impact}, confidence=${result.confidence.toFixed(2)}`,
           );
-          return { result, weight };
+          return { backend: backend.name, result, weight };
         } catch (err) {
           logger.error(`Backend "${backend.name}" threw unexpectedly`, err);
           return {
+            backend: backend.name,
             result: {
-              impact: 'neutral' as Impact,
+              impact: "neutral" as Impact,
               confidence: 0,
               reasoning: `${backend.name}: error`,
             },
             weight,
           };
         }
-      })
+      }),
     );
 
-    // Filter to backends that actually produced a signal (confidence > 0)
+    for (const r of rawResults) {
-    const active = results.filter((r) => r.result.confidence > 0);
+      backendResults.push({
+        backend: r.backend,
+        impact: r.result.impact,
+        confidence: r.result.confidence,
+        reasoning: r.result.reasoning,
+        weight: r.weight,
+      });
+    }
+
+    const active = backendResults.filter((r) => r.confidence > 0);
     if (active.length === 0) {
-      return { impact: 'neutral', confidence: 0, reasoning: 'No backends produced a signal.' };
+      return {
+        impact: "neutral",
+        confidence: 0,
+        reasoning: `Insufficient data: no analysis backends produced signals for ${event.type} #${event.number}.`,
+      };
     }
 
-    const totalWeight = active.reduce((s, r) => s + r.weight, 0);
+    // Calculate multi-dimensional scores
-    const combinedScore =
+    const dimensions = this.calculateDimensions(active);
-      active.reduce(
-        (s, r) => s + impactToNumeric[r.result.impact] * r.result.confidence * r.weight,
-        0
-      ) / totalWeight;
-    const combinedConfidence =
-      active.reduce((s, r) => s + r.result.confidence * r.weight, 0) / totalWeight;
 
+    // Detect correlations and patterns
+    const correlations = this.detectCorrelations(backendResults, dimensions);
+
+    // Calculate overall score and confidence
+    const { score, confidence, uncertainty } = this.calculateOverall(
+      active,
+      dimensions,
+      correlations,
+      authorRep,
+    );
+
+    // Determine impact
     let impact: Impact;
-    if (combinedScore > 0.1) {
+    if (score > 0.2) {
-      impact = 'positive';
+      impact = "positive";
-    } else if (combinedScore < -0.1) {
+    } else if (score < -0.2) {
-      impact = 'negative';
+      impact = "negative";
     } else {
-      impact = 'neutral';
+      impact = "neutral";
     }
 
-    if (combinedConfidence < this.confidenceThreshold) {
+    if (confidence < this.confidenceThreshold) {
-      impact = 'neutral';
+      impact = "neutral";
     }
 
-    const reasoning = results
+    // Generate analytical reasoning
-      .filter((r) => r.result.confidence > 0)
+    const reasoning = this.generateAnalyticalReasoning(
-      .map((r) => r.result.reasoning)
+      event,
-      .join(' ');
+      backendResults,
+      active,
+      dimensions,
+      correlations,
+      score,
+      confidence,
+      uncertainty,
+      authorRep,
+      repoCtx,
+    );
 
-    return { impact, confidence: combinedConfidence, reasoning };
+    const result: AnalysisResult = {
+      impact,
+      confidence,
+      reasoning,
+      dimensions,
+      correlations,
+      uncertainty,
+    };
+
+    updateContext(event, result);
+    return result;
+  }
+
+  private calculateDimensions(
+    active: BackendResult[],
+  ): NonNullable<AnalysisResult["dimensions"]> {
+    let correctnessScore = 0;
+    let correctnessWeight = 0;
+    let riskScore = 0;
+    let riskWeight = 0;
+    let maintainabilityScore = 0;
+    let maintainabilityWeight = 0;
+    let alignmentScore = 0;
+    let alignmentWeight = 0;
+
+    for (const r of active) {
+      const impactScore =
+        r.impact === "positive" ? 1 : r.impact === "negative" ? -1 : 0;
+      const weightedImpact = impactScore * r.confidence * r.weight;
+
+      switch (r.backend) {
+        case "checks":
+          correctnessScore += weightedImpact * 0.7;
+          correctnessWeight += r.weight * 0.7;
+          riskScore += weightedImpact * 0.3;
+          riskWeight += r.weight * 0.3;
+          break;
+        case "diff":
+          maintainabilityScore += weightedImpact * 0.6;
+          maintainabilityWeight += r.weight * 0.6;
+          riskScore += weightedImpact * 0.4;
+          riskWeight += r.weight * 0.4;
+          break;
+        case "quality":
+          alignmentScore += weightedImpact * 0.7;
+          alignmentWeight += r.weight * 0.7;
+          maintainabilityScore += weightedImpact * 0.3;
+          maintainabilityWeight += r.weight * 0.3;
+          break;
+      }
+    }
+
+    return {
+      correctness:
+        correctnessWeight > 0 ? correctnessScore / correctnessWeight : 0,
+      risk: riskWeight > 0 ? riskScore / riskWeight : 0,
+      maintainability:
+        maintainabilityWeight > 0
+          ? maintainabilityScore / maintainabilityWeight
+          : 0,
+      alignment: alignmentWeight > 0 ? alignmentScore / alignmentWeight : 0,
+    };
+  }
+
+  private detectCorrelations(
+    allResults: BackendResult[],
+    dimensions: NonNullable<AnalysisResult["dimensions"]>,
+  ): NonNullable<AnalysisResult["correlations"]> {
+    const suspiciousPatterns: string[] = [];
+    const reinforcingSignals: string[] = [];
+    const contradictions: string[] = [];
+
+    const active = allResults.filter((r) => r.confidence > 0);
+    const hasChecks = active.some((r) => r.backend === "checks");
+    const hasDiff = active.some((r) => r.backend === "diff");
+    const hasQuality = active.some((r) => r.backend === "quality");
+
+    // Check for suspicious patterns
+    if (hasChecks && hasDiff) {
+      const checksResult = active.find((r) => r.backend === "checks");
+      const diffResult = active.find((r) => r.backend === "diff");
+
+      if (
+        checksResult?.impact === "positive" &&
+        diffResult?.impact === "negative"
+      ) {
+        suspiciousPatterns.push(
+          "Checks pass but diff analysis shows concerns (untested changes?)",
+        );
+      }
+
+      if (
+        checksResult?.impact === "negative" &&
+        diffResult?.impact === "positive"
+      ) {
+        suspiciousPatterns.push(
+          "Clean diff but failing checks (test failures?)",
+        );
+      }
+    }
+
+    if (hasDiff && hasQuality) {
+      const diffResult = active.find((r) => r.backend === "diff");
+      const qualityResult = active.find((r) => r.backend === "quality");
+
+      if (
+        diffResult?.impact === "positive" &&
+        qualityResult?.impact === "negative"
+      ) {
+        suspiciousPatterns.push(
+          "Clean code changes but poor description (documentation debt)",
+        );
+      }
+    }
+
+    // Check for reinforcing signals
+    if (dimensions.correctness > 0.5 && dimensions.maintainability > 0.5) {
+      reinforcingSignals.push(
+        "High correctness and maintainability scores align",
+      );
+    }
+
+    if (dimensions.risk < -0.5 && dimensions.alignment < -0.3) {
+      reinforcingSignals.push("Risk and misalignment indicators converge");
+    }
+
+    // Check for contradictions
+    const positiveCount = active.filter((r) => r.impact === "positive").length;
+    const negativeCount = active.filter((r) => r.impact === "negative").length;
+
+    if (positiveCount > 0 && negativeCount > 0) {
+      contradictions.push(
+        `Mixed backend signals: ${positiveCount} positive, ${negativeCount} negative`,
+      );
+    }
+
+    if (dimensions.correctness > 0.3 && dimensions.risk < -0.3) {
+      contradictions.push("Correct implementation but high risk profile");
+    }
+
+    return {
+      suspiciousPatterns,
+      reinforcingSignals,
+      contradictions,
+    };
+  }
+
+  private calculateOverall(
+    active: BackendResult[],
+    dimensions: NonNullable<AnalysisResult["dimensions"]>,
+    correlations: NonNullable<AnalysisResult["correlations"]>,
+    authorRep: ReturnType<typeof getAuthorReputation>,
+  ): {
+    score: number;
+    confidence: number;
+    uncertainty: NonNullable<AnalysisResult["uncertainty"]>;
+  } {
+    const totalWeight = active.reduce((s, r) => s + r.weight, 0);
+
+    // Calculate weighted average of individual backend scores
+    let baseScore = 0;
+    for (const r of active) {
+      const impactScore =
+        r.impact === "positive" ? 1 : r.impact === "negative" ? -1 : 0;
+      baseScore += impactScore * r.confidence * r.weight;
+    }
+
+    // Guard against division by zero when all weights are 0
+    if (totalWeight === 0) {
+      baseScore = 0;
+      getLogger().debug(
+        "All backend weights are zero, defaulting baseScore to 0",
+      );
+    } else {
+      baseScore /= totalWeight;
+    }
+
+    // Adjust score based on dimensions
+    const dimensionScore =
+      dimensions.correctness * 0.4 +
+      dimensions.risk * 0.2 +
+      dimensions.maintainability * 0.25 +
+      dimensions.alignment * 0.15;
+
+    // Blend backend score with dimension score
+    let finalScore = baseScore * 0.6 + dimensionScore * 0.4;
+
+    // Penalize for contradictions
+    if (correlations.contradictions.length > 0) {
+      finalScore *= 0.85;
+    }
+
+    // Calculate confidence
+    let baseConfidence = 0;
+    if (totalWeight === 0) {
+      baseConfidence = 0;
+      getLogger().debug(
+        "All backend weights are zero, defaulting baseConfidence to 0",
+      );
+    } else {
+      baseConfidence =
+        active.reduce((s, r) => s + r.confidence * r.weight, 0) / totalWeight;
+    }
+
+    // Adjust confidence based on various factors
+    const uniqueImpacts = new Set(active.map((r) => r.impact));
+    if (uniqueImpacts.size === 1) {
+      baseConfidence = Math.min(1, baseConfidence * 1.15);
+    } else if (uniqueImpacts.size === 3) {
+      baseConfidence *= 0.75;
+    }
+
+    if (authorRep.isNew) {
+      baseConfidence *= 0.9;
+    }
+
+    if (correlations.suspiciousPatterns.length > 0) {
+      baseConfidence *= 0.85;
+    }
+
+    // Calculate uncertainty interval
+    const uncertaintyRange = 1 - baseConfidence;
+    const lowerBound = Math.max(0, baseConfidence - uncertaintyRange * 0.5);
+    const upperBound = Math.min(1, baseConfidence + uncertaintyRange * 0.5);
+
+    // Determine primary uncertainty source
+    let primaryUncertaintySource = "Backend confidence variance";
+    if (uniqueImpacts.size > 1) {
+      primaryUncertaintySource = "Mixed backend signals";
+    } else if (authorRep.isNew) {
+      primaryUncertaintySource = "Limited author history";
+    } else if (active.length < this.backends.length) {
+      primaryUncertaintySource = "Partial backend coverage";
+    }
+
+    return {
+      score: finalScore,
+      confidence: baseConfidence,
+      uncertainty: {
+        confidenceInterval: [lowerBound, upperBound],
+        primaryUncertaintySource,
+      },
+    };
+  }
+
+  private generateAnalyticalReasoning(
+    event: WebhookEvent,
+    _allResults: BackendResult[],
+    activeResults: BackendResult[],
+    dimensions: NonNullable<AnalysisResult["dimensions"]>,
+    correlations: NonNullable<AnalysisResult["correlations"]>,
+    score: number,
+    confidence: number,
+    uncertainty: NonNullable<AnalysisResult["uncertainty"]>,
+    authorRep: ReturnType<typeof getAuthorReputation>,
+    repoCtx: ReturnType<typeof getRepoContext>,
+  ): string {
+    const parts: string[] = [];
+
+    // Summary of dimensional analysis
+    parts.push(`Dimensional Analysis:`);
+    parts.push(
+      `  Correctness: ${(dimensions.correctness * 100).toFixed(0)}% | ` +
+        `Risk: ${(dimensions.risk * 100).toFixed(0)}% | ` +
+        `Maintainability: ${(dimensions.maintainability * 100).toFixed(0)}% | ` +
+        `Alignment: ${(dimensions.alignment * 100).toFixed(0)}%`,
+    );
+
+    // Backend breakdown
+    parts.push(`\nBackend Results:`);
+    for (const r of activeResults) {
+      const icon =
+        r.impact === "positive"
+          ? "[+]"
+          : r.impact === "negative"
+            ? "[-]"
+            : "[~]";
+      parts.push(
+        `  ${icon} ${r.backend}: ${r.impact} (${(r.confidence * 100).toFixed(0)}%) - ${r.reasoning}`,
+      );
+    }
+
+    // Correlation analysis
+    const hasPatterns =
+      correlations.suspiciousPatterns.length > 0 ||
+      correlations.reinforcingSignals.length > 0 ||
+      correlations.contradictions.length > 0;
+
+    if (hasPatterns) {
+      parts.push(`\nPattern Analysis:`);
+
+      if (correlations.reinforcingSignals.length > 0) {
+        for (const signal of correlations.reinforcingSignals) {
+          parts.push(`  [^] ${signal}`);
+        }
+      }
+
+      if (correlations.suspiciousPatterns.length > 0) {
+        for (const pattern of correlations.suspiciousPatterns) {
+          parts.push(`  [!] ${pattern}`);
+        }
+      }
+
+      if (correlations.contradictions.length > 0) {
+        for (const contradiction of correlations.contradictions) {
+          parts.push(`  [x] ${contradiction}`);
+        }
+      }
+    }
+
+    // Context information
+    parts.push(`\nContext:`);
+    parts.push(
+      `  Author: ${event.author} (${authorRep.isNew ? "new" : `reputation: ${(authorRep.reputation * 100).toFixed(0)}%`})`,
+    );
+    parts.push(
+      `  Repository: ${repoCtx.maturity} (${repoCtx.communitySize} active contributors)`,
+    );
+
+    // Confidence and uncertainty
+    parts.push(`\nConfidence Assessment:`);
+    parts.push(`  Overall: ${(confidence * 100).toFixed(0)}%`);
+    parts.push(
+      `  Interval: [${(uncertainty.confidenceInterval[0] * 100).toFixed(0)}%, ${(uncertainty.confidenceInterval[1] * 100).toFixed(0)}%]`,
+    );
+    parts.push(
+      `  Primary uncertainty: ${uncertainty.primaryUncertaintySource}`,
+    );
+
+    // Final assessment
+    parts.push(
+      `\nAssessment: ${score > 0.2 ? "POSITIVE" : score < -0.2 ? "NEGATIVE" : "NEUTRAL"} (score: ${score.toFixed(2)})`,
+    );
+
+    return parts.join("\n");
   }
 }
 

src/polling.ts

@@ -1,6 +1,7 @@
-import type { Config, WebhookEvent, RepoConfig } from './types.js';
+import type { Config, WebhookEvent, RepoPattern } from "./types.js";
 import {
   listRecentComments,
+  listAccessibleRepositories,
   fetchPR,
   fetchIssue,
   hasExistingComment,
@@ -9,15 +10,16 @@ import {
   formatComment,
   createReaction,
   type RecentComment,
-} from './github.js';
+} from "./github.js";
-import { createEngine } from './engine/index.js';
+import { createEngine } from "./engine/index.js";
-import { getLogger } from './logger.js';
+import { getLogger } from "./logger.js";
-import { recordEvent } from './events.js';
+import { recordEvent } from "./events.js";
-import { readFileSync, writeFileSync, existsSync } from 'fs';
+import { readFileSync, writeFileSync, existsSync } from "fs";
 
 interface ProcessedComment {
   id: number;
   timestamp: number;
+  failures?: number;
 }
 
 interface PollingState {
@@ -26,24 +28,27 @@ interface PollingState {
 
 const processedComments: Map<string, ProcessedComment> = new Map();
 const MAX_PROCESSED_CACHE = 1000;
+const MAX_RETRY_ATTEMPTS = 3;
 
 let pollingState: PollingState = { lastProcessedAt: {} };
 
 function loadPollingState(stateFile: string): void {
   if (existsSync(stateFile)) {
     try {
-      const data = readFileSync(stateFile, 'utf-8');
+      const data = readFileSync(stateFile, "utf-8");
       const parsed = JSON.parse(data);
       // Validate that parsed data has expected structure
       if (
         parsed &&
-        typeof parsed === 'object' &&
+        typeof parsed === "object" &&
         parsed.lastProcessedAt &&
-        typeof parsed.lastProcessedAt === 'object'
+        typeof parsed.lastProcessedAt === "object"
       ) {
         pollingState = parsed;
       } else {
-        getLogger().warn('Invalid polling state format, resetting to empty state');
+        getLogger().warn(
+          "Invalid polling state format, resetting to empty state",
+        );
         pollingState = { lastProcessedAt: {} };
       }
     } catch {
@@ -57,7 +62,7 @@ function savePollingState(stateFile: string): void {
   try {
     writeFileSync(stateFile, JSON.stringify(pollingState, null, 2));
   } catch (err) {
-    getLogger().warn('Failed to save polling state', err);
+    getLogger().warn("Failed to save polling state", err);
   }
 }
 
@@ -86,7 +91,11 @@ function isProcessed(owner: string, repo: string, commentId: number): boolean {
 
 function markProcessed(owner: string, repo: string, commentId: number): void {
   const key = getCacheKey(owner, repo, commentId);
-  processedComments.set(key, { id: commentId, timestamp: Date.now() });
+  processedComments.set(key, {
+    id: commentId,
+    timestamp: Date.now(),
+    failures: 0,
+  });
 
   // Clean up old entries if cache is too large
   if (processedComments.size > MAX_PROCESSED_CACHE) {
@@ -99,13 +108,31 @@ function markProcessed(owner: string, repo: string, commentId: number): void {
   }
 }
 
+function recordFailure(
+  owner: string,
+  repo: string,
+  commentId: number,
+): boolean {
+  const key = getCacheKey(owner, repo, commentId);
+  const existing = processedComments.get(key);
+
+  const failures = (existing?.failures || 0) + 1;
+  processedComments.set(key, {
+    id: commentId,
+    timestamp: Date.now(),
+    failures,
+  });
+
+  return failures >= MAX_RETRY_ATTEMPTS;
+}
+
 function containsMention(body: string): boolean {
-  return body.includes('@troutbot');
+  return body.includes("@troutbot");
 }
 
 async function analyzeAndComment(
   event: WebhookEvent,
-  config: Config
+  config: Config,
 ): Promise<Record<string, unknown>> {
   const logger = getLogger();
   const engine = createEngine(config.engine);
@@ -113,36 +140,50 @@ async function analyzeAndComment(
   // Run analysis
   const analysis = await engine.analyze(event);
   logger.info(
-    `Analyzed ${event.owner}/${event.repo}#${event.number}: impact=${analysis.impact}, confidence=${analysis.confidence.toFixed(2)}`
+    `Analyzed ${event.owner}/${event.repo}#${event.number}: impact=${analysis.impact}, confidence=${analysis.confidence.toFixed(2)}`,
   );
 
   // Check for existing comment
   const { commentMarker, allowUpdates } = config.response;
-  const existing = await hasExistingComment(event.owner, event.repo, event.number, commentMarker);
+  const existing = await hasExistingComment(
+    event.owner,
+    event.repo,
+    event.number,
+    commentMarker,
+  );
 
   if (existing.exists && !allowUpdates) {
-    logger.info(`Already commented on ${event.owner}/${event.repo}#${event.number}, skipping`);
+    logger.info(
-    const result = { skipped: true, reason: 'Already commented' };
+      `Already commented on ${event.owner}/${event.repo}#${event.number}, skipping`,
+    );
+    const result = { skipped: true, reason: "Already commented" };
     recordEvent(event, result, analysis);
     return result;
   }
 
+  // Generate comment with analysis
   const body = formatComment(
     config.response,
     event.type,
     analysis.impact,
     analysis.confidence,
-    analysis.reasoning
+    analysis.reasoning,
   );
 
   if (existing.exists && allowUpdates && existing.commentId) {
-    logger.info(`Updating existing comment on ${event.owner}/${event.repo}#${event.number}`);
+    logger.info(
+      `Updating existing comment on ${event.owner}/${event.repo}#${event.number}`,
+    );
     await updateComment(event.owner, event.repo, existing.commentId, body);
   } else {
     await postComment(event.owner, event.repo, event.number, body);
   }
 
-  const result = { processed: true, impact: analysis.impact, confidence: analysis.confidence };
+  const result = {
+    processed: true,
+    impact: analysis.impact,
+    confidence: analysis.confidence,
+  };
   recordEvent(event, result, analysis);
   return result;
 }
@@ -155,18 +196,38 @@ function isAuthorized(username: string, authorizedUsers?: string[]): boolean {
   return authorizedUsers.some((u) => u.toLowerCase() === normalizedUsername);
 }
 
-function isRepoAuthorized(owner: string, repo: string, pollingRepos?: RepoConfig[]): boolean {
+function isRepoAuthorized(
-  if (!pollingRepos || pollingRepos.length === 0) {
+  owner: string,
-    return true; // No restrictions, use global repos
+  repo: string,
+  pollingPatterns?: RepoPattern[],
+): boolean {
+  if (!pollingPatterns || pollingPatterns.length === 0) {
+    return true; // No restrictions, accept all repos
   }
-  return pollingRepos.some((r) => r.owner === owner && r.repo === repo);
+
+  // Check if repo matches any pattern
+  for (const pattern of pollingPatterns) {
+    const ownerMatch =
+      pattern.owner === "*" ||
+      pattern.owner.toLowerCase() === owner.toLowerCase();
+    const repoMatch =
+      pattern.repo === "*" ||
+      pattern.repo === undefined ||
+      pattern.repo.toLowerCase() === repo.toLowerCase();
+
+    if (ownerMatch && repoMatch) {
+      return true;
+    }
+  }
+
+  return false;
 }
 
 async function processComment(
   comment: RecentComment,
   owner: string,
   repo: string,
-  config: Config
+  config: Config,
 ): Promise<void> {
   const logger = getLogger();
 
@@ -175,7 +236,9 @@ async function processComment(
   }
 
   if (isProcessed(owner, repo, comment.id)) {
-    logger.debug(`Comment ${owner}/${repo}#${comment.id} already processed, skipping`);
+    logger.debug(
+      `Comment ${owner}/${repo}#${comment.id} already processed, skipping`,
+    );
     return;
   }
 
@@ -183,9 +246,9 @@ async function processComment(
   const pollingRepos = config.polling?.repositories;
   if (!isRepoAuthorized(owner, repo, pollingRepos)) {
     logger.info(
-      `Unauthorized repo ${owner}/${repo} for polling, ignoring mention from ${comment.author}`
+      `Unauthorized repo ${owner}/${repo} for polling, ignoring mention from ${comment.author}`,
     );
-    await createReaction(owner, repo, comment.id, 'thumbs_down');
+    await createReaction(owner, repo, comment.id, "thumbs_down");
     markProcessed(owner, repo, comment.id);
     return;
   }
@@ -194,13 +257,15 @@ async function processComment(
   const authorizedUsers = config.polling?.authorizedUsers;
   if (!isAuthorized(comment.author, authorizedUsers)) {
     logger.info(
-      `Unauthorized user ${comment.author} attempted on-demand analysis in ${owner}/${repo}#${comment.issueNumber}`
+      `Unauthorized user ${comment.author} attempted on-demand analysis in ${owner}/${repo}#${comment.issueNumber}`,
     );
     markProcessed(owner, repo, comment.id);
     return;
   }
 
-  logger.info(`Found @troutbot mention in ${owner}/${repo}#${comment.issueNumber}`);
+  logger.info(
+    `Found @troutbot mention in ${owner}/${repo}#${comment.issueNumber}`,
+  );
 
   try {
     // First, try to fetch as a PR to check if it's a pull request
@@ -211,8 +276,8 @@ async function processComment(
     if (prData) {
       // It's a pull request
       event = {
-        action: 'on_demand',
+        action: "on_demand",
-        type: 'pull_request',
+        type: "pull_request",
         number: comment.issueNumber,
         title: prData.title,
         body: prData.body,
@@ -227,13 +292,15 @@ async function processComment(
       // It's an issue
       const issueData = await fetchIssue(owner, repo, comment.issueNumber);
       if (!issueData) {
-        logger.warn(`Could not fetch issue ${owner}/${repo}#${comment.issueNumber}`);
+        logger.warn(
+          `Could not fetch issue ${owner}/${repo}#${comment.issueNumber}`,
+        );
         return;
       }
 
       event = {
-        action: 'on_demand',
+        action: "on_demand",
-        type: 'issue',
+        type: "issue",
         number: comment.issueNumber,
         title: issueData.title,
         body: issueData.body,
@@ -248,10 +315,21 @@ async function processComment(
     markProcessed(owner, repo, comment.id);
 
     logger.info(
-      `Successfully processed on-demand analysis for ${owner}/${repo}#${comment.issueNumber}`
+      `Successfully processed on-demand analysis for ${owner}/${repo}#${comment.issueNumber}`,
     );
   } catch (err) {
-    logger.error(`Failed to process mention in ${owner}/${repo}#${comment.issueNumber}`, err);
+    logger.error(
+      `Failed to process mention in ${owner}/${repo}#${comment.issueNumber}`,
+      err,
+    );
+    // Track failures and mark as processed after max retries
+    const shouldStop = recordFailure(owner, repo, comment.id);
+    if (shouldStop) {
+      logger.warn(
+        `Max retry attempts (${MAX_RETRY_ATTEMPTS}) reached for comment ${comment.id}, marking as processed`,
+      );
+      markProcessed(owner, repo, comment.id);
+    }
   }
 }
 
@@ -260,13 +338,15 @@ async function pollRepository(
   repo: string,
   config: Config,
   since: Date,
-  stateFile?: string
+  stateFile?: string,
 ): Promise<void> {
   const logger = getLogger();
 
   try {
     const comments = await listRecentComments(owner, repo, since);
-    logger.debug(`Fetched ${comments.length} recent comments from ${owner}/${repo}`);
+    logger.debug(
+      `Fetched ${comments.length} recent comments from ${owner}/${repo}`,
+    );
 
     let latestCommentDate = since;
 
@@ -293,26 +373,60 @@ export async function startPolling(config: Config): Promise<void> {
   const pollingConfig = config.polling;
 
   if (!pollingConfig || !pollingConfig.enabled) {
-    logger.info('Polling is disabled');
+    logger.info("Polling is disabled");
     return;
   }
 
-  if (config.repositories.length === 0) {
+  // Determine repos to poll
-    logger.warn('Polling enabled but no repositories configured');
+  let reposToPoll: Array<{ owner: string; repo: string }>;
+  const pollingPatterns = pollingConfig.repositories;
+
+  if (!pollingPatterns || pollingPatterns.length === 0) {
+    // No patterns configured - poll all accessible repos
+    reposToPoll = await listAccessibleRepositories();
+    logger.info(
+      `Polling all accessible repositories (${reposToPoll.length} repos)`,
+    );
+  } else {
+    // Build repo list from patterns
+    reposToPoll = [];
+
+    // Start with explicitly configured global repos (for webhooks)
+    for (const repo of config.repositories) {
+      // Only include repos that match polling patterns
+      if (isRepoAuthorized(repo.owner, repo.repo, pollingPatterns)) {
+        reposToPoll.push({ owner: repo.owner, repo: repo.repo });
+      }
+    }
+
+    // If no explicit repos configured, fetch accessible ones and filter by patterns
+    if (config.repositories.length === 0) {
+      const accessibleRepos = await listAccessibleRepositories();
+      for (const repo of accessibleRepos) {
+        if (isRepoAuthorized(repo.owner, repo.repo, pollingPatterns)) {
+          reposToPoll.push({ owner: repo.owner, repo: repo.repo });
+        }
+      }
+    }
+
+    logger.info(`Polling ${reposToPoll.length} repositories matching patterns`);
+  }
+
+  if (reposToPoll.length === 0) {
+    logger.warn("No repositories match polling patterns");
     return;
   }
 
   const intervalMs = pollingConfig.intervalMinutes * 60 * 1000;
   const lookbackMs = pollingConfig.lookbackMinutes * 60 * 1000;
 
-  logger.info(`Starting polling for ${config.repositories.length} repositories`);
   logger.info(
-    `Poll interval: ${pollingConfig.intervalMinutes} minutes, lookback: ${pollingConfig.lookbackMinutes} minutes`
+    `Poll interval: ${pollingConfig.intervalMinutes} minutes, lookback: ${pollingConfig.lookbackMinutes} minutes`,
   );
 
   // Load persisted state if backfill is enabled
   const stateFile = pollingConfig.backfill
-    ? pollingConfig.stateFile || '.troutbot-polling-state.json'
+    ? pollingConfig.stateFile || ".troutbot-polling-state.json"
     : undefined;
   if (stateFile) {
     loadPollingState(stateFile);
@@ -320,22 +434,28 @@ export async function startPolling(config: Config): Promise<void> {
   }
 
   // Do an initial poll - use persisted timestamp if available, otherwise use lookback
-  for (const repo of config.repositories) {
+  for (const repo of reposToPoll) {
     const lastProcessed = getLastProcessedAt(repo.owner, repo.repo);
     const initialSince = lastProcessed || new Date(Date.now() - lookbackMs);
     if (lastProcessed) {
       logger.info(
-        `Resuming polling for ${repo.owner}/${repo.repo} from ${lastProcessed.toISOString()}`
+        `Resuming polling for ${repo.owner}/${repo.repo} from ${lastProcessed.toISOString()}`,
       );
     }
-    await pollRepository(repo.owner, repo.repo, config, initialSince, stateFile);
+    await pollRepository(
+      repo.owner,
+      repo.repo,
+      config,
+      initialSince,
+      stateFile,
+    );
   }
 
   // Set up recurring polling
   setInterval(async () => {
     const since = new Date(Date.now() - lookbackMs);
 
-    for (const repo of config.repositories) {
+    for (const repo of reposToPoll) {
       await pollRepository(repo.owner, repo.repo, config, since, stateFile);
     }
   }, intervalMs);