tewip-ai 0.70.2 → 0.70.4

package/README.md

@@ -1035,6 +1035,11 @@ import does not show.
 notice it, then run the **whole** suite to find out which tests really catch it. A failure is dropping a
 catching test.
 
+Re-derived on 2026-10-17 against a suite that had grown to 1,136 tests: **9 judged changes, 0 catching
+tests lost, 5.7% of the suite selected.** The harness refuses to report at all when it could not read
+which tests caught anything, because a selection cannot be shown safe by changes nothing was observed
+to catch.
+
 | Project | Judged cases | Catching test lost | Suite selected |
 | --- | --- | --- | --- |
 | This repository (node --test, 76 test files) | 8 | **0** | 7–9% |

package/dist/engine/notify/note-command.js

@@ -7,6 +7,7 @@
 import { existsSync, mkdirSync, readdirSync, writeFileSync } from 'node:fs';
 import { dirname, join, resolve } from 'node:path';
 import { HistoryIndex } from '../history.js';
+import { loadCorrections } from '../corrections.js';
 import { buildItems } from '../../outlets/items.js';
 import { digestView } from '../../outlets/views/digest.js';
 import { qualityView } from '../../outlets/views/quality.js';
@@ -36,9 +37,13 @@ export async function runNote(cmd) {
             continue;
         totalRuns += runs.length;
         totalFailures += records.length;
-        // The diagnoses were made when each failure happened; nothing is judged again here.
+        // The diagnoses were made when each failure happened; nothing is judged again here - except
+        // where a person overruled one. This note is the report a lead reads, and it was printing
+        // "Diagnosis: LOCATOR_DRIFT, confidence 0.95, source: rules" about a test somebody had corrected
+        // to TIMING, because the corrections were never loaded here at all.
         const index = new HistoryIndex(records, runs);
-        const items = await buildItems(records, index, { dataset: `history: ${name}`, llm: null, reuseStoredDiagnoses: true });
+        const where = join(cmd.root, '.tewip', 'history', name);
+        const items = await buildItems(records, index, { dataset: `history: ${name}`, llm: null, reuseStoredDiagnoses: true, corrections: loadCorrections(where) });
         parts.push(`## ${name}`, '', `${runs.length} run(s), ${records.length} failure(s).`, '');
         if (items.length > 0) {
             parts.push(digestView(items, runs, `history: ${name}`).markdown, '', qualityView(items, `history: ${name}`).markdown, '');

package/dist/outlets/items.js

@@ -4,6 +4,7 @@ import { extractLocator } from '../engine/error-parse.js';
 import { decide } from '../engine/decide.js';
 import { lastFailedAttempts } from '../engine/dataset.js';
 import { failingCodeOf } from '../engine/llm/context.js';
+import { applyCorrection, correctionFor } from '../engine/corrections.js';
 import { diagnosePipeline } from '../engine/pipeline.js';
 import { truncate } from '../engine/signature.js';
 import { designDebt } from './design-debt.js';
@@ -47,7 +48,24 @@ export async function buildItems(records, index, { dataset, llm, reuseStoredDiag
     const stored = reuseStoredDiagnoses ? cases.map((c) => c.diagnosis) : cases.map(() => undefined);
     const missing = cases.filter((_, i) => !stored[i]);
     const fresh = await diagnosePipeline(missing, index, llm, { corrections });
-    const diagnoses = stored.map((d) => d ?? fresh.shift());
+    // A correction outranks a stored diagnosis, which it did not.
+    //
+    // `diagnosePipeline` applies corrections, but only to the failures in `missing` - the ones with no
+    // stored diagnosis. `explain` stores one for every failure it diagnoses, so on any history a team
+    // actually keeps, **every** failure has one and a correction reached none of them. Field-tested:
+    // three real corrections of real diagnoses, `learn` reported the pattern perfectly, and the weekly
+    // note still said "Diagnosis: LOCATOR_DRIFT, confidence 0.95, source: rules" about a test a person
+    // had overruled to TIMING - directly against what `correct` had just promised in print, "Tewip will
+    // answer with your decision the next time that test fails the same way, and say it was yours."
+    //
+    // This is the whole hierarchy of the product: what the team decided beats what the tool decided.
+    // Fixing it here fixes every caller at once, which matters because three of them reuse stored
+    // diagnoses and two of those did not even pass the corrections along.
+    const diagnoses = cases.map((record, i) => {
+        const own = stored[i] ?? fresh.shift();
+        const correction = corrections.length > 0 ? correctionFor(record, corrections) : null;
+        return correction ? applyCorrection(own, correction) : own;
+    });
     const failedAttempts = new Map();
     for (const r of records)
         failedAttempts.set(`${r.runId} ${r.testId}`, [...(failedAttempts.get(`${r.runId} ${r.testId}`) ?? []), r]);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tewip-ai",
-  "version": "0.70.2",
+  "version": "0.70.4",
   "mcpName": "io.github.hmamut39/tewip",
   "description": "Tewip finds why each failing test failed - real bug, flaky test, environment, test data or changed selector - for any stack, from the command line, CI, or your editor.",
   "keywords": [