test(ms-ai-architect): Spor 1 — akseptansetester (residual scoped, worklist-aktivering, CT5/N4, poison-doc) [skip-docs]

tests/kb-eval/test-eval.test.mjs

@@ -18,6 +18,7 @@ import {
   checkN5,
   checkCT5,
 } from '../../scripts/kb-eval/eval.mjs';
+import { insertHeaderFields, insertToc } from '../../scripts/kb-update/lib/transform.mjs';
 
 test('splitFrontmatter — separates frontmatter from body', () => {
   const c = '---\nname: x\ndescription: hi\n---\n# Body\nline2\n';
@@ -294,3 +295,40 @@ test('checkCT5 — 0.80 is the pass threshold (mirrors the old K8 ratio target)'
   assert.equal(checkCT5(mk(5, 4)).pass, true, '4/5 = 0.80 passes');
   assert.equal(checkCT5(mk(5, 3)).pass, false, '3/5 = 0.60 fails');
 });
+
+// --- Spor 1 acceptance: CT5 sourcedness ACTIVATES and N4 TOC passes on the migrated
+// substrate. Fixtures are produced by the REAL Session-2 primitives (insertHeaderFields
+// + insertToc) applied to legacy files — so this asserts the substrate makes both
+// signals move, not that a hand-forged header does.
+
+// A legacy LARGE reference file (>100 lines), no Type/Source/TOC.
+const LEGACY_LARGE = '# Big Reference\n\n**Status:** GA\n\n---\n\n' +
+  Array.from({ length: 8 }, (_, i) => `## Section ${i + 1}\n\n${'body paragraph.\n'.repeat(14)}`).join('\n');
+const MIGRATED_LARGE = insertToc(insertHeaderFields(LEGACY_LARGE, {
+  type: 'reference', source: 'https://learn.microsoft.com/azure/big',
+}));
+const MIGRATED_SMALL = insertHeaderFields('# Small Reference\n\n**Status:** GA\n\n---\n\n## A\n\ntekst\n', {
+  type: 'reference', source: 'https://learn.microsoft.com/azure/small',
+});
+const MIGRATED_REFS = [
+  { path: 'references/big.md', content: MIGRATED_LARGE },
+  { path: 'references/small.md', content: MIGRATED_SMALL },
+];
+
+test('acceptance — CT5 activates on the migrated substrate (referenceFiles>0, fully sourced, pass)', () => {
+  const r = checkCT5(MIGRATED_REFS);
+  // referenceFiles>0 is the score-layer's "available" predicate — CT5 is no longer dormant.
+  assert.ok(r.referenceFiles > 0, 'CT5 available: at least one type:reference file');
+  assert.equal(r.referenceFiles, 2);
+  assert.equal(r.sourced, 2);
+  assert.equal(r.ratio, 1);
+  assert.equal(r.pass, true);
+});
+
+test('acceptance — N4 passes: the migrated large file carries a recognised ## Innhold TOC', () => {
+  const r = checkN4(MIGRATED_REFS);
+  assert.ok(r.largeFiles >= 1, 'the large file is in N4 scope');
+  assert.equal(r.withToc, r.largeFiles, 'every large file has a TOC');
+  assert.equal(r.ratio, 1);
+  assert.equal(r.pass, true);
+});

tests/kb-update/test-full-pass-worklist.test.mjs

@@ -22,6 +22,12 @@ import {
   classifyFullPassDue,
   buildFullPassReport,
 } from '../../scripts/kb-update/lib/full-pass-worklist.mjs';
+import { normalizeStaleVerified } from '../../scripts/kb-update/lib/transform.mjs';
+import {
+  parseTypeHeader,
+  parseSourceHeader,
+  parseVerifiedHeader,
+} from '../../scripts/kb-update/lib/kb-headers.mjs';
 
 const TODAY = '2026-06-29';
 const MAX_AGE = 90; // threshold = 2026-03-31 (verified strictly before → cadence-due)
@@ -195,3 +201,46 @@ test('buildFullPassReport — no source map still works (annotation null)', () =
   assert.equal(r.worklist.length, 1);
   assert.equal(r.worklist[0].newestSourceLastmod, null);
 });
+
+// --- Spor 1 acceptance: worklist ACTIVATION + the **Verified:** MCP poison it fixes ---
+
+test('acceptance — migrated reference headers (type+source, verified:null) land in never-verified; unmigrated===0', () => {
+  const reader = (p) => ({
+    'm1.md': { type: 'reference', source: 'https://learn.microsoft.com/1', verified: null },
+    'm2.md': { type: 'reference', source: 'https://learn.microsoft.com/2', verified: null },
+  }[p] ?? null);
+  const r = buildFullPassReport(['m1.md', 'm2.md'], reader, new Map(), opts);
+  assert.equal(r.counts.neverVerified, 2);
+  assert.ok(r.counts.due > 0, 'the worklist activates — due>0');
+  assert.equal(r.counts.unmigrated, 0, 'no legacy no-Type files remain');
+  assert.ok(r.worklist.every((e) => e.reason === 'never-verified'));
+});
+
+// The poison, exercised through the REAL parser + normalizer (not a hand-forged
+// header value): a header-region `**Verified:** MCP <date>` parses to the non-date
+// token 'MCP', which the cadence classifier reads as a recent verification and drops
+// the file to `fresh` — omitted from the worklist AND falsely "verified". This is the
+// exact failure mode normalizeStaleVerified removes.
+const POISONED =
+  '# Poisoned Ref\n\n**Status:** GA\n**Verified:** MCP 2026-06\n**Type:** reference\n' +
+  '**Source:** https://learn.microsoft.com/x\n\n---\n\n## A\n\ntekst\n';
+const hdr = (c) => ({ type: parseTypeHeader(c), source: parseSourceHeader(c), verified: parseVerifiedHeader(c) });
+
+test('acceptance — **Verified:** MCP poison drops a never-verified ref to fresh (the bug the normalizer fixes)', () => {
+  assert.equal(parseVerifiedHeader(POISONED), 'MCP'); // the poison: a non-date token read as "verified"
+  const r = buildFullPassReport(['p.md'], (p) => ({ 'p.md': hdr(POISONED) }[p] ?? null), new Map(), opts);
+  assert.equal(r.counts.fresh, 1, 'wrongly counted fresh');
+  assert.equal(r.worklist.length, 0, 'omitted from the worklist — the poison');
+  assert.equal(r.counts.unmigrated, 0);
+});
+
+test('acceptance — normalizeStaleVerified restores the poisoned file to a due never-verified entry', () => {
+  const fixed = normalizeStaleVerified(POISONED);
+  assert.equal(parseVerifiedHeader(fixed), null); // stale token stripped
+  assert.equal(parseTypeHeader(fixed), 'reference'); // Type/Source untouched
+  assert.equal(parseSourceHeader(fixed), 'https://learn.microsoft.com/x');
+  const r = buildFullPassReport(['p.md'], (p) => ({ 'p.md': hdr(fixed) }[p] ?? null), new Map(), opts);
+  assert.equal(r.worklist.length, 1);
+  assert.equal(r.worklist[0].reason, 'never-verified');
+  assert.ok(r.counts.due > 0);
+});

tests/kb-update/test-validate-kb-file.test.mjs

@@ -10,7 +10,12 @@
 import { test } from 'node:test';
 import assert from 'node:assert/strict';
 import { validatePaths } from '../../scripts/kb-update/validate-kb-file.mjs';
-import { buildKbHeader } from '../../scripts/kb-update/lib/transform.mjs';
+import {
+  buildKbHeader,
+  insertHeaderFields,
+  validateKbFile,
+} from '../../scripts/kb-update/lib/transform.mjs';
+import { parseTypeHeader, parseSourceHeader } from '../../scripts/kb-update/lib/kb-headers.mjs';
 
 const GOOD = buildKbHeader({
   title: 'Azure AI Search - Hybrid Retrieval',
@@ -78,3 +83,44 @@ test('validatePaths — no paths → ok:true (nothing to gate)', () => {
   assert.equal(r.ok, true);
   assert.deepEqual(r.results, []);
 });
+
+// --- Spor 1 acceptance: the residual criterion after the substrate migration ---
+// These generate the migrated file through the REAL Session-2 primitive
+// (insertHeaderFields), then assert the create-guard's residual — proving the
+// substrate closes the worklist-relevant fields (type+source) even where base
+// fields lag. `verified`/`verified_by` stay open by design (the judge fills them,
+// next brief), so the residual for a conformant file is EXACTLY those two.
+
+// A base-field-CONFORMANT legacy file: title + English **Last updated:** + **Status:**,
+// small (no TOC required). Migrated → Type + Source stamped, still no Verified.
+const LEGACY_CONFORMANT =
+  '# Azure AI Foundry\n\n**Last updated:** 2026-06\n**Status:** GA\n\n---\n\n## Oversikt\n\ntekst\n';
+// A base-field-INCOMPLETE legacy file: Norwegian **Sist oppdatert:** (RE_LAST_UPDATED is
+// English-only) + **Kategori:** but NO **Status:**. Migrated → Type + Source stamped.
+const LEGACY_INCOMPLETE =
+  '# Copilot Studio\n\n**Kategori:** Styring\n**Sist oppdatert:** 2026-06\n\n---\n\n## Oversikt\n\ntekst\n';
+
+test('acceptance — a migrated base-field-conformant reference has residual EXACTLY [verified, verified_by]', () => {
+  const migrated = insertHeaderFields(LEGACY_CONFORMANT, {
+    type: 'reference',
+    source: 'https://learn.microsoft.com/azure/ai-foundry',
+  });
+  assert.equal(parseTypeHeader(migrated), 'reference');
+  assert.equal(parseSourceHeader(migrated), 'https://learn.microsoft.com/azure/ai-foundry');
+  assert.deepEqual(validateKbFile(migrated).missing, ['verified', 'verified_by']);
+});
+
+test('acceptance — a migrated base-field-incomplete reference has residual ⊆ the 4-set, with type+source present', () => {
+  const migrated = insertHeaderFields(LEGACY_INCOMPLETE, {
+    type: 'reference',
+    source: 'https://learn.microsoft.com/power-platform/copilot',
+  });
+  const { missing } = validateKbFile(migrated);
+  const FOUR_SET = ['last_updated', 'status', 'verified', 'verified_by'];
+  assert.ok(missing.every((m) => FOUR_SET.includes(m)), `residual ${JSON.stringify(missing)} ⊆ 4-set`);
+  // The substrate DID complete the worklist-relevant fields — type + source are present,
+  // never in the residual, even though base fields (Norwegian date / no Status) lag.
+  assert.ok(!missing.includes('source'));
+  assert.equal(parseTypeHeader(migrated), 'reference');
+  assert.equal(parseSourceHeader(migrated), 'https://learn.microsoft.com/power-platform/copilot');
+});