feat(recall): add diffusion stage and improve retrieval metrics

modules/story-summary/vector/retrieval/metrics.js

@@ -78,7 +78,7 @@ export function createMetrics() {
             inStore: 0,
             considered: 0,
             selected: 0,
-            byRecallType: { direct: 0, related: 0, causal: 0, lexical: 0 },
+            byRecallType: { direct: 0, related: 0, causal: 0, lexical: 0, l0Linked: 0 },
             similarityDistribution: { min: 0, max: 0, mean: 0, median: 0 },
             entityFilter: null,
             causalChainDepth: 0,
@@ -112,6 +112,23 @@ export function createMetrics() {
             assemblyTime: 0,
         },
 
+        // Diffusion (PPR Spreading Activation) - 图扩散
+        diffusion: {
+            seedCount: 0,
+            graphNodes: 0,
+            graphEdges: 0,
+            iterations: 0,
+            convergenceError: 0,
+            pprActivated: 0,
+            cosineGatePassed: 0,
+            cosineGateFiltered: 0,
+            cosineGateNoVector: 0,
+            finalCount: 0,
+            scoreDistribution: { min: 0, max: 0, mean: 0 },
+            byChannel: { who: 0, what: 0, where: 0, how: 0 },
+            time: 0,
+        },
+
         // Formatting - 格式化
         formatting: {
             sectionsIncluded: [],
@@ -140,6 +157,7 @@ export function createMetrics() {
             evidenceRetrieval: 0,
             evidenceRerank: 0,
             evidenceAssembly: 0,
+            diffusion: 0,
             formatting: 0,
             total: 0,
         },
@@ -249,9 +267,6 @@ export function formatMetricsLog(metrics) {
     // Fusion (W-RRF, floor-level)
     lines.push('[Fusion] W-RRF (floor-level) - 多路融合');
     lines.push(`├─ dense_floors: ${m.fusion.denseFloors}`);
-    if (m.fusion.denseAggMethod) {
-        lines.push(`│   └─ aggregation: ${m.fusion.denseAggMethod}`);
-    }
     lines.push(`├─ lex_floors: ${m.fusion.lexFloors}`);
     if (m.fusion.lexDensityBonus > 0) {
         lines.push(`│   └─ density_bonus: ${m.fusion.lexDensityBonus}`);
@@ -291,7 +306,12 @@ export function formatMetricsLog(metrics) {
     lines.push(`│   ├─ direct: ${m.event.byRecallType.direct}`);
     lines.push(`│   ├─ related: ${m.event.byRecallType.related}`);
     lines.push(`│   ├─ causal: ${m.event.byRecallType.causal}`);
-    lines.push(`│   └─ lexical: ${m.event.byRecallType.lexical}`);
+    if (m.event.byRecallType.l0Linked) {
+        lines.push(`│   ├─ lexical: ${m.event.byRecallType.lexical}`);
+        lines.push(`│   └─ l0_linked: ${m.event.byRecallType.l0Linked}`);
+    } else {
+        lines.push(`│   └─ lexical: ${m.event.byRecallType.lexical}`);
+    }
 
     const sim = m.event.similarityDistribution;
     if (sim && sim.max > 0) {
@@ -340,6 +360,32 @@ export function formatMetricsLog(metrics) {
     lines.push(`└─ assembly_time: ${m.evidence.assemblyTime}ms`);
     lines.push('');
 
+    // Diffusion (PPR)
+    lines.push('[Diffusion] PPR Spreading Activation');
+    lines.push(`├─ seeds: ${m.diffusion.seedCount}`);
+    lines.push(`├─ graph: ${m.diffusion.graphNodes} nodes, ${m.diffusion.graphEdges} edges`);
+    if (m.diffusion.graphEdges > 0) {
+        const ch = m.diffusion.byChannel || {};
+        lines.push(`│   └─ by_channel: who=${ch.who || 0}, what=${ch.what || 0}, where=${ch.where || 0}, how=${ch.how || 0}`);
+    }
+    if (m.diffusion.iterations > 0) {
+        lines.push(`├─ ppr: ${m.diffusion.iterations} iterations, ε=${Number(m.diffusion.convergenceError).toExponential(1)}`);
+    }
+    lines.push(`├─ activated (excl seeds): ${m.diffusion.pprActivated}`);
+    if (m.diffusion.pprActivated > 0) {
+        lines.push(`├─ cosine_gate: ${m.diffusion.cosineGatePassed} passed, ${m.diffusion.cosineGateFiltered} filtered`);
+        if (m.diffusion.cosineGateNoVector > 0) {
+            lines.push(`│   └─ no_vector: ${m.diffusion.cosineGateNoVector}`);
+        }
+    }
+    lines.push(`├─ final_injected: ${m.diffusion.finalCount}`);
+    if (m.diffusion.finalCount > 0) {
+        const ds = m.diffusion.scoreDistribution;
+        lines.push(`├─ scores: min=${ds.min}, max=${ds.max}, mean=${ds.mean}`);
+    }
+    lines.push(`└─ time: ${m.diffusion.time}ms`);
+    lines.push('');
+
     // Formatting
     lines.push('[Formatting] 格式化');
     lines.push(`├─ sections: [${(m.formatting.sectionsIncluded || []).join(', ')}]`);
@@ -372,6 +418,7 @@ export function formatMetricsLog(metrics) {
     lines.push(`├─ evidence_retrieval: ${m.timing.evidenceRetrieval}ms`);
     lines.push(`├─ floor_rerank: ${m.timing.evidenceRerank || 0}ms`);
     lines.push(`├─ l1_cosine: ${m.evidence.l1CosineTime}ms`);
+    lines.push(`├─ diffusion: ${m.timing.diffusion}ms`);
     lines.push(`├─ evidence_assembly: ${m.timing.evidenceAssembly}ms`);
     lines.push(`├─ formatting: ${m.timing.formatting}ms`);
     lines.push(`└─ total: ${m.timing.total}ms`);
@@ -578,5 +625,25 @@ export function detectIssues(metrics) {
         issues.push(`Slow L1 cosine scoring (${m.evidence.l1CosineTime}ms) - too many chunks pulled`);
     }
 
+    // ─────────────────────────────────────────────────────────────────
+    // Diffusion 问题
+    // ─────────────────────────────────────────────────────────────────
+
+    if (m.diffusion.graphEdges === 0 && m.diffusion.seedCount > 0) {
+        issues.push('No diffusion graph edges - atoms may lack who/edges fields');
+    }
+
+    if (m.diffusion.pprActivated > 0 && m.diffusion.cosineGatePassed === 0) {
+        issues.push('All PPR-activated nodes failed cosine gate - graph structure diverged from query semantics');
+    }
+
+    if (m.diffusion.cosineGateNoVector > 5) {
+        issues.push(`${m.diffusion.cosineGateNoVector} PPR nodes missing vectors - L0 vectorization may be incomplete`);
+    }
+
+    if (m.diffusion.time > 50) {
+        issues.push(`Slow diffusion (${m.diffusion.time}ms) - graph may be too dense`);
+    }
+
     return issues;
 }