// tests/synthetic/plan-determinism.test.mjs // SC7 plan-determinism floor — Jaccard pipeline test. // // Reads two synthetic plan-run fixtures and asserts that // jaccardSimilarity(stepsTokens(planA), stepsTokens(planB)) >= 0.833. // // This exercises the determinism pipeline (parser + jaccard) on a known // input pair. It does NOT measure real-LLM determinism — that is deferred // to a future run of the pipeline against examples/01-add-verbose-flag/. import { test } from 'node:test'; import { strict as assert } from 'node:assert'; import { readFileSync } from 'node:fs'; import { join, dirname } from 'node:path'; import { fileURLToPath } from 'node:url'; import { jaccardSimilarity } from '../../lib/parsers/jaccard.mjs'; import { parseDocument } from '../../lib/util/frontmatter.mjs'; const HERE = dirname(fileURLToPath(import.meta.url)); const ROOT = join(HERE, '..', '..'); const SC7_THRESHOLD = 0.833; function loadSteps(rel) { const text = readFileSync(join(ROOT, rel), 'utf-8'); const doc = parseDocument(text); assert.ok(doc.valid, `frontmatter of ${rel} did not parse: ${(doc.errors || []).map(e => e.message).join(', ')}`); const steps = doc.parsed.frontmatter && doc.parsed.frontmatter.steps; assert.ok(Array.isArray(steps), `frontmatter.steps of ${rel} is not an array`); return steps; } test('plan determinism — Jaccard of synthetic plan-run-A vs plan-run-B meets SC7 threshold (0.833)', () => { const a = loadSteps('tests/synthetic/plan-run-A.md'); const b = loadSteps('tests/synthetic/plan-run-B.md'); const sim = jaccardSimilarity(a, b); assert.ok( sim >= SC7_THRESHOLD, `jaccardSimilarity(stepsTokens(planA), stepsTokens(planB)) = ${sim} < ${SC7_THRESHOLD} (SC7 floor). ` + `Fixtures may have drifted — re-tune step titles to restore the overlap.`, ); }); test('plan determinism — both fixtures contain at least 30 unique step titles', () => { for (const rel of ['tests/synthetic/plan-run-A.md', 'tests/synthetic/plan-run-B.md']) { const steps = loadSteps(rel); assert.ok( new Set(steps).size >= 30, `${rel}: < 30 unique step titles (got ${new Set(steps).size}). Synthetic fixtures must reflect a substantial plan.`, ); } }); test('plan determinism — no duplicate step titles within run', () => { for (const rel of ['tests/synthetic/plan-run-A.md', 'tests/synthetic/plan-run-B.md']) { const steps = loadSteps(rel); assert.strictEqual( new Set(steps).size, steps.length, `${rel}: contains duplicate step titles (${steps.length} entries vs ${new Set(steps).size} unique)`, ); } }); // --- v4.1 forward-compat block (SC #10) --- // // Adding the optional frontmatter key `profile_used` (Step 3 OPTIONAL_STRING_KEYS) // must not break parsing of EITHER: // - Existing plans WITHOUT profile_used (plan-run-A.md, plan-run-B.md) // - New plans WITH profile_used (profile-plan-run-{economy,premium}-*.md) // // This is the forward-compat assertion required by Step 19. Extend-in-place // keeps the determinism + forward-compat checks colocated. test('plan determinism — forward-compat: legacy fixtures (no profile_used) parse cleanly', () => { for (const rel of ['tests/synthetic/plan-run-A.md', 'tests/synthetic/plan-run-B.md']) { const text = readFileSync(join(ROOT, rel), 'utf-8'); const doc = parseDocument(text); assert.ok(doc.valid, `${rel}: frontmatter parse failed: ${(doc.errors || []).map((e) => e.message).join(', ')}`); assert.equal( doc.parsed.frontmatter.profile_used, undefined, `${rel}: legacy fixture must NOT have profile_used set`, ); assert.ok( Array.isArray(doc.parsed.frontmatter.steps), `${rel}: steps array still loads after parser extension`, ); } }); test('plan determinism — forward-compat: new fixtures with profile_used parse cleanly', () => { const cases = [ { rel: 'tests/synthetic/profile-plan-run-economy-1.md', profile: 'economy' }, { rel: 'tests/synthetic/profile-plan-run-economy-2.md', profile: 'economy' }, { rel: 'tests/synthetic/profile-plan-run-premium-1.md', profile: 'premium' }, { rel: 'tests/synthetic/profile-plan-run-premium-2.md', profile: 'premium' }, ]; for (const { rel, profile } of cases) { const text = readFileSync(join(ROOT, rel), 'utf-8'); const doc = parseDocument(text); assert.ok(doc.valid, `${rel}: frontmatter parse failed: ${(doc.errors || []).map((e) => e.message).join(', ')}`); assert.equal( doc.parsed.frontmatter.profile_used, profile, `${rel}: profile_used must be ${profile}`, ); assert.ok( Array.isArray(doc.parsed.frontmatter.steps) && doc.parsed.frontmatter.steps.length >= 10, `${rel}: steps array must be non-empty`, ); } }); test('plan determinism — forward-compat: synthetic v1.7 plan validates with --strict (no PLAN_VERSION_MISMATCH)', async () => { // Sanity check that adding profile_used to manifest-yaml schema doesn't // regress full plan-validator strict-mode behaviour on a v1.7 plan with // standard step + manifest structure. Uses a committed synthetic fixture // (plan-run-C.md) instead of a gitignored project plan so the assertion // is stable across worktrees and headless runs. const fixturePlan = 'tests/synthetic/plan-run-C.md'; const { validatePlan } = await import('../../lib/validators/plan-validator.mjs'); const result = await validatePlan(join(ROOT, fixturePlan), { strict: true }); assert.equal(result.valid, true, `synthetic plan must validate strict: ${JSON.stringify(result.errors)}`); const versionMismatch = (result.warnings || []).find((w) => w.code === 'PLAN_VERSION_MISMATCH'); assert.equal(versionMismatch, undefined, 'synthetic plan must NOT emit PLAN_VERSION_MISMATCH warning'); });