@hmharness/evolution 0.14.4 → 0.14.7

package/dist/candidates.d.ts

@@ -46,6 +46,12 @@ export interface ExperimentArm {
     n: number;
     passRate: number;
     tokens: number;
+    /** per-case rows (day-56): which case flipped is the point of a diff */
+    results?: Array<{
+        name: string;
+        pass: boolean;
+        tokens: number;
+    }>;
 }
 export interface ExperimentReport {
     candidateId: string;

package/dist/candidates.js

@@ -105,23 +105,29 @@ export async function runCandidateExperiment(home, candidateId, opts) {
     // holdout cases are excluded from promotion gates (bench.ts anti-memorization)
     const gate = opts.cases.filter((c) => !c.holdout).slice(0, Math.max(1, opts.maxCases ?? 12));
     let cPass = 0, cTok = 0, tPass = 0, tTok = 0;
+    // per-case rows (day-56 observability: which case flipped is the whole
+    // point of a diff - aggregates alone cannot answer it)
+    const ctlRows = [];
+    const trtRows = [];
     for (const c of gate) {
         const ctl = await opts.runCase(c, 'control');
         if (ctl.pass)
             cPass++;
         cTok += ctl.tokens;
+        ctlRows.push({ name: c.name, pass: ctl.pass, tokens: ctl.tokens });
         const trt = await opts.runCase(c, 'treatment');
         if (trt.pass)
             tPass++;
         tTok += trt.tokens;
+        trtRows.push({ name: c.name, pass: trt.pass, tokens: trt.tokens });
     }
     const v = verdictFor({ pass: cPass, n: gate.length }, { pass: tPass, n: gate.length });
     const report = {
         candidateId,
         ranAt: new Date().toISOString(),
         cases: gate.length,
-        control: { pass: cPass, n: gate.length, passRate: gate.length ? cPass / gate.length : 0, tokens: cTok },
-        treatment: { pass: tPass, n: gate.length, passRate: gate.length ? tPass / gate.length : 0, tokens: tTok },
+        control: { pass: cPass, n: gate.length, passRate: gate.length ? cPass / gate.length : 0, tokens: cTok, results: ctlRows },
+        treatment: { pass: tPass, n: gate.length, passRate: gate.length ? tPass / gate.length : 0, tokens: tTok, results: trtRows },
         ...v,
     };
     const dir = join(home, 'evolution', 'experiments', candidateId);

package/dist/evolve.js

@@ -167,14 +167,18 @@ export async function runEvolution(opts) {
     }
     else {
         proposals = await proposeSkills(provider, signals, say, ancestor ?? undefined, mergeWith ?? undefined);
-        if (proposals.length === 0) {
-            try {
-                const { workflowProposals } = await import("./workflows.js");
-                proposals = await workflowProposals(provider, home, say);
-            }
-            catch (err) {
-                say(`  awm skipped: ${String(err).slice(0, 80)}`);
-            }
+        // AWM is ADDITIVE, not fallback-only (design intent: "higher abstraction
+        // than per-mistake reflection"). With batch-produced task archetypes
+        // repeating 100+ times/day, the workflow path must get its turn even
+        // when the meta-model also proposes - the -workflow naming + gate still
+        // decides promotion. Guard: only append when clusters are real (>=5).
+        try {
+            const { workflowProposals } = await import("./workflows.js");
+            const wf = await workflowProposals(provider, home, say);
+            proposals = [...proposals, ...wf];
+        }
+        catch (err) {
+            say(`  awm skipped: ${String(err).slice(0, 80)}`);
         }
     }
     report.proposals = proposals;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hmharness/evolution",
-  "version": "0.14.4",
+  "version": "0.14.7",
   "description": "hmharness evolution subsystem: persistent memory, insight capture, skill library, and the bench that gives evolution its fitness signal. First-class kernel citizen, not a plugin.",
   "type": "module",
   "main": "dist/index.js",