memorix 1.1.7 → 1.1.8

package/src/store/project-affinity.ts

@@ -1,195 +1,195 @@
-/**
- * Project Affinity Scoring
- *
- * Prevents cross-project memory pollution by scoring search results
- * based on how well they match the current project context.
- *
- * Inspired by mcp-memory-service's memory-scorer.js:
- * - High affinity: content mentions project name → full score
- * - Medium affinity: related concepts but no direct mention → 0.7x
- * - Low affinity: no project reference → 0.3x (heavily penalized)
- *
- * This runs AFTER projectId filtering, as a second layer of defense
- * against memories that were stored under the correct projectId but
- * contain content about a different project (e.g., discussing Memorix
- * development while in a test project workspace).
- */
-
-export interface AffinityContext {
-  /** Current project name (e.g., "for_memmcp_test", "memorix") */
-  projectName: string;
-  /** Current project ID (e.g., "local/for_memmcp_test", "AVIDS2/memorix") */
-  projectId: string;
-  /** Optional: keywords that indicate project relevance */
-  projectKeywords?: string[];
-}
-
-export interface MemoryContent {
-  title: string;
-  narrative?: string;
-  facts?: string[];
-  concepts?: string[];
-  entityName?: string;
-  filesModified?: string[];
-}
-
-export interface AffinityResult {
-  /** Affinity score 0-1 (1 = high affinity, 0 = no affinity) */
-  score: number;
-  /** Affinity level for debugging */
-  level: 'high' | 'medium' | 'low' | 'none';
-  /** Reason for the score */
-  reason: string;
-}
-
-/**
- * Calculate project affinity score for a memory.
- *
- * @param memory - The memory content to evaluate
- * @param context - Current project context
- * @returns AffinityResult with score, level, and reason
- */
-export function calculateProjectAffinity(
-  memory: MemoryContent,
-  context: AffinityContext,
-): AffinityResult {
-  const projectName = context.projectName.toLowerCase();
-  const projectId = context.projectId.toLowerCase();
-  
-  // Extract base name from projectId (e.g., "memorix" from "AVIDS2/memorix")
-  const projectBaseName = projectId.split('/').pop() ?? projectName;
-  
-  // Build searchable content string
-  const contentParts = [
-    memory.title,
-    memory.narrative ?? '',
-    memory.entityName ?? '',
-    ...(memory.facts ?? []),
-    ...(memory.concepts ?? []),
-    ...(memory.filesModified ?? []),
-  ];
-  const content = contentParts.join(' ').toLowerCase();
-  
-  // Check for direct project name mention
-  if (content.includes(projectName) || content.includes(projectBaseName)) {
-    return { score: 1.0, level: 'high', reason: 'project_name_in_content' };
-  }
-  
-  // Check for project keywords (if provided)
-  if (context.projectKeywords && context.projectKeywords.length > 0) {
-    const keywordsLower = context.projectKeywords.map(k => k.toLowerCase());
-    const matchedKeywords = keywordsLower.filter(k => content.includes(k));
-    if (matchedKeywords.length >= 2) {
-      return { score: 0.9, level: 'high', reason: `keywords_matched: ${matchedKeywords.join(', ')}` };
-    }
-    if (matchedKeywords.length === 1) {
-      return { score: 0.7, level: 'medium', reason: `keyword_matched: ${matchedKeywords[0]}` };
-    }
-  }
-  
-  // Check for file paths that suggest project relevance
-  const files = memory.filesModified ?? [];
-  if (files.some(f => f.toLowerCase().includes(projectName) || f.toLowerCase().includes(projectBaseName))) {
-    return { score: 0.85, level: 'high', reason: 'project_in_file_path' };
-  }
-  
-  // Check entity name
-  if (memory.entityName) {
-    const entityLower = memory.entityName.toLowerCase();
-    if (entityLower.includes(projectName) || entityLower.includes(projectBaseName)) {
-      return { score: 0.8, level: 'high', reason: 'project_in_entity' };
-    }
-  }
-  
-  // Check concepts for project-related terms
-  const concepts = memory.concepts ?? [];
-  if (concepts.some(c => c.toLowerCase().includes(projectName) || c.toLowerCase().includes(projectBaseName))) {
-    return { score: 0.75, level: 'medium', reason: 'project_in_concepts' };
-  }
-  
-  // No project reference found — this memory might be about a different project
-  // Apply penalty but don't completely filter (might still be relevant)
-  return { score: 0.65, level: 'low', reason: 'no_project_reference' };
-}
-
-/**
- * Apply project affinity scoring to search results.
- *
- * @param results - Search results with scores
- * @param memories - Full memory content for each result (keyed by ID)
- * @param context - Current project context
- * @param options - Scoring options
- * @returns Results with adjusted scores, sorted by affinity-weighted score
- */
-export function applyProjectAffinity<T extends { id: number; score: number }>(
-  results: T[],
-  memories: Map<number, MemoryContent>,
-  context: AffinityContext,
-  options: {
-    /** Minimum affinity score to include (default: 0, include all) */
-    minAffinity?: number;
-    /** Whether to filter out low-affinity results entirely (default: false) */
-    filterLowAffinity?: boolean;
-  } = {},
-): T[] {
-  const { minAffinity = 0, filterLowAffinity = false } = options;
-  
-  // Calculate affinity for each result
-  const withAffinity = results.map(result => {
-    const memory = memories.get(result.id);
-    if (!memory) {
-      // No memory content available — assume medium affinity
-      return { ...result, affinity: 0.5, affinityLevel: 'medium' as const };
-    }
-    
-    const { score: affinityScore, level } = calculateProjectAffinity(memory, context);
-    return {
-      ...result,
-      score: result.score * affinityScore, // Apply affinity as multiplier
-      affinity: affinityScore,
-      affinityLevel: level,
-    };
-  });
-  
-  // Filter by minimum affinity if requested
-  let filtered = withAffinity;
-  if (filterLowAffinity) {
-    filtered = withAffinity.filter(r => r.affinity >= 0.5);
-  } else if (minAffinity > 0) {
-    filtered = withAffinity.filter(r => r.affinity >= minAffinity);
-  }
-  
-  // Re-sort by affinity-weighted score
-  filtered.sort((a, b) => b.score - a.score);
-  
-  // Return without the extra affinity fields (keep original shape)
-  return filtered.map(({ affinity: _, affinityLevel: __, ...rest }) => rest as T);
-}
-
-/**
- * Extract project keywords from project name and common patterns.
- * Used to improve affinity detection for projects with distinctive names.
- */
-export function extractProjectKeywords(projectName: string, projectId: string): string[] {
-  const keywords: string[] = [projectName];
-  
-  // Add base name from projectId
-  const baseName = projectId.split('/').pop();
-  if (baseName && baseName !== projectName) {
-    keywords.push(baseName);
-  }
-  
-  // Add common variations
-  const variations = [
-    projectName.replace(/-/g, '_'),
-    projectName.replace(/_/g, '-'),
-    projectName.replace(/[_-]/g, ''),
-  ];
-  for (const v of variations) {
-    if (v !== projectName && !keywords.includes(v)) {
-      keywords.push(v);
-    }
-  }
-  
-  return keywords.filter(k => k.length > 2);
-}
+/**
+ * Project Affinity Scoring
+ *
+ * Prevents cross-project memory pollution by scoring search results
+ * based on how well they match the current project context.
+ *
+ * Inspired by mcp-memory-service's memory-scorer.js:
+ * - High affinity: content mentions project name → full score
+ * - Medium affinity: related concepts but no direct mention → 0.7x
+ * - Low affinity: no project reference → 0.3x (heavily penalized)
+ *
+ * This runs AFTER projectId filtering, as a second layer of defense
+ * against memories that were stored under the correct projectId but
+ * contain content about a different project (e.g., discussing Memorix
+ * development while in a test project workspace).
+ */
+
+export interface AffinityContext {
+  /** Current project name (e.g., "for_memmcp_test", "memorix") */
+  projectName: string;
+  /** Current project ID (e.g., "local/for_memmcp_test", "AVIDS2/memorix") */
+  projectId: string;
+  /** Optional: keywords that indicate project relevance */
+  projectKeywords?: string[];
+}
+
+export interface MemoryContent {
+  title: string;
+  narrative?: string;
+  facts?: string[];
+  concepts?: string[];
+  entityName?: string;
+  filesModified?: string[];
+}
+
+export interface AffinityResult {
+  /** Affinity score 0-1 (1 = high affinity, 0 = no affinity) */
+  score: number;
+  /** Affinity level for debugging */
+  level: 'high' | 'medium' | 'low' | 'none';
+  /** Reason for the score */
+  reason: string;
+}
+
+/**
+ * Calculate project affinity score for a memory.
+ *
+ * @param memory - The memory content to evaluate
+ * @param context - Current project context
+ * @returns AffinityResult with score, level, and reason
+ */
+export function calculateProjectAffinity(
+  memory: MemoryContent,
+  context: AffinityContext,
+): AffinityResult {
+  const projectName = context.projectName.toLowerCase();
+  const projectId = context.projectId.toLowerCase();
+  
+  // Extract base name from projectId (e.g., "memorix" from "AVIDS2/memorix")
+  const projectBaseName = projectId.split('/').pop() ?? projectName;
+  
+  // Build searchable content string
+  const contentParts = [
+    memory.title,
+    memory.narrative ?? '',
+    memory.entityName ?? '',
+    ...(memory.facts ?? []),
+    ...(memory.concepts ?? []),
+    ...(memory.filesModified ?? []),
+  ];
+  const content = contentParts.join(' ').toLowerCase();
+  
+  // Check for direct project name mention
+  if (content.includes(projectName) || content.includes(projectBaseName)) {
+    return { score: 1.0, level: 'high', reason: 'project_name_in_content' };
+  }
+  
+  // Check for project keywords (if provided)
+  if (context.projectKeywords && context.projectKeywords.length > 0) {
+    const keywordsLower = context.projectKeywords.map(k => k.toLowerCase());
+    const matchedKeywords = keywordsLower.filter(k => content.includes(k));
+    if (matchedKeywords.length >= 2) {
+      return { score: 0.9, level: 'high', reason: `keywords_matched: ${matchedKeywords.join(', ')}` };
+    }
+    if (matchedKeywords.length === 1) {
+      return { score: 0.7, level: 'medium', reason: `keyword_matched: ${matchedKeywords[0]}` };
+    }
+  }
+  
+  // Check for file paths that suggest project relevance
+  const files = memory.filesModified ?? [];
+  if (files.some(f => f.toLowerCase().includes(projectName) || f.toLowerCase().includes(projectBaseName))) {
+    return { score: 0.85, level: 'high', reason: 'project_in_file_path' };
+  }
+  
+  // Check entity name
+  if (memory.entityName) {
+    const entityLower = memory.entityName.toLowerCase();
+    if (entityLower.includes(projectName) || entityLower.includes(projectBaseName)) {
+      return { score: 0.8, level: 'high', reason: 'project_in_entity' };
+    }
+  }
+  
+  // Check concepts for project-related terms
+  const concepts = memory.concepts ?? [];
+  if (concepts.some(c => c.toLowerCase().includes(projectName) || c.toLowerCase().includes(projectBaseName))) {
+    return { score: 0.75, level: 'medium', reason: 'project_in_concepts' };
+  }
+  
+  // No project reference found — this memory might be about a different project
+  // Apply penalty but don't completely filter (might still be relevant)
+  return { score: 0.65, level: 'low', reason: 'no_project_reference' };
+}
+
+/**
+ * Apply project affinity scoring to search results.
+ *
+ * @param results - Search results with scores
+ * @param memories - Full memory content for each result (keyed by ID)
+ * @param context - Current project context
+ * @param options - Scoring options
+ * @returns Results with adjusted scores, sorted by affinity-weighted score
+ */
+export function applyProjectAffinity<T extends { id: number; score: number }>(
+  results: T[],
+  memories: Map<number, MemoryContent>,
+  context: AffinityContext,
+  options: {
+    /** Minimum affinity score to include (default: 0, include all) */
+    minAffinity?: number;
+    /** Whether to filter out low-affinity results entirely (default: false) */
+    filterLowAffinity?: boolean;
+  } = {},
+): T[] {
+  const { minAffinity = 0, filterLowAffinity = false } = options;
+  
+  // Calculate affinity for each result
+  const withAffinity = results.map(result => {
+    const memory = memories.get(result.id);
+    if (!memory) {
+      // No memory content available — assume medium affinity
+      return { ...result, affinity: 0.5, affinityLevel: 'medium' as const };
+    }
+    
+    const { score: affinityScore, level } = calculateProjectAffinity(memory, context);
+    return {
+      ...result,
+      score: result.score * affinityScore, // Apply affinity as multiplier
+      affinity: affinityScore,
+      affinityLevel: level,
+    };
+  });
+  
+  // Filter by minimum affinity if requested
+  let filtered = withAffinity;
+  if (filterLowAffinity) {
+    filtered = withAffinity.filter(r => r.affinity >= 0.5);
+  } else if (minAffinity > 0) {
+    filtered = withAffinity.filter(r => r.affinity >= minAffinity);
+  }
+  
+  // Re-sort by affinity-weighted score
+  filtered.sort((a, b) => b.score - a.score);
+  
+  // Return without the extra affinity fields (keep original shape)
+  return filtered.map(({ affinity: _, affinityLevel: __, ...rest }) => rest as T);
+}
+
+/**
+ * Extract project keywords from project name and common patterns.
+ * Used to improve affinity detection for projects with distinctive names.
+ */
+export function extractProjectKeywords(projectName: string, projectId: string): string[] {
+  const keywords: string[] = [projectName];
+  
+  // Add base name from projectId
+  const baseName = projectId.split('/').pop();
+  if (baseName && baseName !== projectName) {
+    keywords.push(baseName);
+  }
+  
+  // Add common variations
+  const variations = [
+    projectName.replace(/-/g, '_'),
+    projectName.replace(/_/g, '-'),
+    projectName.replace(/[_-]/g, ''),
+  ];
+  for (const v of variations) {
+    if (v !== projectName && !keywords.includes(v)) {
+      keywords.push(v);
+    }
+  }
+  
+  return keywords.filter(k => k.length > 2);
+}