Tests share their helpers, lean on the baselines, and migration scaffolding carries an expiry

Why

Every file writes its own helpers.

• Nine plugins/*/test/helpers.ts files (601 lines) are the same apart from the plugin they import: places and business differ in exactly two lines. packages/runes/test/helpers.ts is a near-copy.
• 66 test files define their own tree walkers: asTag in 31, findTag in 12, find in 12, findAllTags in 11, textOf/text in 10, and more. There are 34 local render and 16 local run definitions; about 9 CLI tests each spawn the binary with their own run(), and 47 files hand-roll a temp directory.
• t.attributes['data-rune'] === '…' appears 615 times in test predicates. That's one helper call, findRune(tree, 'badge'), written out longhand.

Some unit tests restate a baseline. contracts/seo-baseline/baseline.json (SPEC-130) pins the full JSON-LD for every emitting rune. Many cases in packages/runes/test/seo.test.ts and plugins/*/test/seo.test.ts assert fields of the very same fixture. For example, plugins/storytelling/test/seo.test.ts:21-33 checks Veshra is a Person with a jobTitle, which fixtures/character.md pins in full. Testimonial's JSON-LD is now asserted three times: the marketing SEO test, composed-testimonial.test.ts and the baseline.

Some tests are one table written as seven files. The seven packages/lumina/test/*-preset.test.ts files (~1,000 lines) each check that a preset stays out of typography/structural/status roles, claims chrome + code-surface + syntax, and composes with tideline, niwaki and lumina.

Migration scaffolding has no end date. The composition migration needs tests that compare a composition against the native rune it replaces (SPEC-147 D2), tests that hold the staging site equal to main, and policy tests over the staging convention. They are right while the migration runs and dead after it, but nothing marks which is which:

• packages/content/test/composed-{character,lore,bond,realm-faction}.test.ts, the composed-storytelling.ts helper and its 19 fixtures: ~1,500–2,000 lines that die with the native storytelling runes. About 170 of those lines already only record spike findings about what a spec example "does not construct" (e.g. composed-realm-faction.test.ts:549-662).
• staging-site.test.ts and staging-convention.test.ts (~250 lines): dead once site/staged-runes/ is empty.
• composed-hero/cta/testimonial.test.ts each keep a tombstone test asserting the native rune is gone, and still build a fixture copy of the plugin through composed-marketing.ts even though the runes are live.
• imperative-schema-migration.test.ts has PENDING = {}: its migration is finished, and what remains is a regex lint.

Plan's equivalence snapshot is 11,089 lines. plugins/plan/test/emits-equivalence.test.ts was the SPEC-143 D7 gate for moving work, bug and decision to emits. That migration is done. But its header says why it can't simply go: the SEO baseline doesn't cover these runes and the structure contract is derived from config, so this snapshot is the only thing that would notice a change in what their schemas build. It is so large because of JSON.stringify(out, null, 1).

Decisions

D1 — one test-utils module

A private workspace module (e.g. packages/test-utils, not published) exports:

• makeRuneTestHelpers(plugin), which replaces the nine plugin helpers.ts files;
• tree walkers (asTag, findTag, findAll, findRune, textOf, findByDataName);
• runCli(args, opts) for the CLI tests and withTempDir().

Files move to it as they are touched; a single sweep is a work item but not required before other work. A lint test reports new local definitions of the exported names.

D2 — a baseline-pinned fact is not asserted again

A unit test that asserts a field of a rune's structured data, where the SEO baseline has a fixture for that rune with the same input, is removed. What stays is what the baseline can't express: intent ("a character with a role becomes a Person"), defects the baseline records but a test wants to see fixed, and inputs with no fixture. The same rule applies to BEM classes and data attributes already pinned by contracts/structures.json.

D3 — the preset tests are one table

The seven preset files become one describe.each over { preset, modes, extendedRoles }, with each preset's specific invariants (Solarized's symmetry, Gruvbox's warmth) kept as their own cases.

D4 — migration tests declare when they expire

A test that exists only for a migration carries a marker naming its exit condition:

describe.expires('storytelling goes live (WORK-682)', () => { … });
// or a header comment: `@expires <work-id>: <condition>`

refrakt plan validate (or a test) lists expiring tests whose named work item is done, so going live includes deleting the scaffolding. Applied now:

• the storytelling comparisons and the composed-storytelling helper expire with the storytelling go-live;
• staging-site.test.ts and staging-convention.test.ts expire when site/staged-runes/ is empty;
• the steps comparison (composed-steps.test.ts:144-189) and plugins/marketing/test/steps.test.ts expire with steps' go-live.

Deleted now: the ~170 lines of spike-finding tests, the three tombstone tests, and imperative-schema-migration.test.ts (or reduced to the lint it actually is). The composed-hero/cta/testimonial tests use the shipping marketing plugin and move to plugins/marketing/test.

D5 — the plan equivalence snapshot is replaced, not deleted

The snapshot keeps pinning what work, bug and decision build, in a form a reviewer can read: the identity-transformed HTML per fixture input (as the SEO baseline's rendered channel does), written as compact files. A change to the snapshot should be a reviewable diff; 11k lines of indented JSON isn't one.

Not in scope

• Algorithmic tests with many distinct cases: resolver.test.ts, engine-core.test.ts, schema-table.test.ts, anchor.test.ts, plugin-validate.test.ts, the plan CLI tests and behaviors. Their it() blocks test different logic; tabulating them would lose clarity.
• The duplicate structures.json, which is data, not a test (SPEC-192).
• css-coverage.test.ts's stale allowlist (SPEC-192).

Acceptance Criteria

• A private test-utils module exports the rune helpers factory, tree walkers, runCli and withTempDir
• The nine plugin helpers.ts files are replaced by makeRuneTestHelpers(plugin)
• SEO unit tests that restate a baseline fixture are removed, and every remaining case says what the baseline doesn't
• The seven preset tests are one table-driven test
• Migration-only tests carry an expiry marker, and a check lists those whose work item is done
• Spike-finding tests, tombstone tests and imperative-schema-migration.test.ts are removed
• The live marketing composition tests use the shipping plugin and live in plugins/marketing/test
• emits-equivalence pins rendered output in a compact, reviewable form, and the 11k-line snapshot is deleted

References

• SPEC-147: D2, the comparison tests D4 gives an expiry.
• SPEC-181: the staging convention its policy tests hold.
• SPEC-143: D7, the gate behind the plan snapshot.
• SPEC-130: the SEO baseline D2 defers to.