test(signals): store oracle + four signal states; latest() seed leak and body-end visibility fixes

[ryansolid]

Follow-up to #3449 / #3451 (both merged). One commit, rebased onto next.

Oracle: 90 → 207 cells

The harness (cell vocabulary, nine readers, hold management, runner) moves to tests/visibility-oracle.harness.ts, shared by:

• signal oracle — 14 states now (+ body ended 2.0.0-rc.8 optimistic update first reveals after its action body has finished #3427, un-superseded A18, held truth 2.0.0-rc until doesnt entangle the update resolving it with the action update #3164, loading window over a held input), 126 cells
• store oracle (new) — 8 states through plain stores, optimistic stores and projections, 72 cells
• plus the dedicated pins below

The store side tracked the signal side far more closely than expected (67/72 on the first run): A18 supersession — tracked pass derives from the landed truth, fresh reader born held, overlay stays displayed — works through store nodes exactly as through signals. #3434's store carve-out is about the body-end settle order, not landing supersession.

Runtime fixes

latest() leaked a projection's seed. latest(() => store.key) on an unresolved createStore(fn, seed) returned the seed; untracked and tracked reads correctly threw (A25). read() routes a latest() read to the companion before its firewall/status logic, and the leaf's own _value is the seed. latestRead now judges "uninitialized" on the leaf's owner (the firewall). verdict.ts only.

Body-end supersession read differently from landing supersession (3 oracle violations). After the action body ends and the override is superseded by the committed truth (#3427): a stale reader re-run by an unrelated write showed the truth while the display showed the override; a fresh mainline memo published the truth instead of being held; isPending read false while latest read the truth. One cause: the node had no _transition stamp — an override written inside an action goes straight into the transaction's _optimisticNodes without the adoption loop that stamps, and body-end stages nothing that would queue it. #2912 already ruled that an override node's owner is _overrideOwner; supersededRead and the verdict now resolve it (resolveTransition). A superseded read enters as a staged read whether the truth is staged or committed.

A latest() call captured the caller's block. Found while pinning the above: the latest() shadow's pull reads through supersededRead → enterStagedRead; verdict pulls are excluded from born-held and so fell to the initTransition path — from an event handler, outside any flush. Verdict pulls are observations and now never enter from mainline. tests/body-end-supersession-visibility.test.ts reads the verdicts before the display checks on purpose.

Rulings recorded (2026-09-15)

• A27: verdict-quiet through the window is a hydration invariant (server answers false; client must match on creation); the window's first real landing is initial-load class — like a boundary's first content reveal — and commits on arrival even when the re-asking input is action-held. From that landing on the node is an ordinary memo.
• A18 body-end visibility: identical to a landing supersession; owner via _overrideOwner.
• A29: verdict pulls never enter from mainline.
• A7: uninitialized judged on the owner.

Verification

• @solidjs/signals 178 files / 2014 · solid-js 618 · @solidjs/web 783 — green (one attribution-holds timing flake on one run; clean on two reruns and in isolation)
• Oracle: 207 cells, zero violations; 15 observed (12 pre-flush cells parked on A28/fix(signals): writes become visible at flush — latest() reads the flushed staged world (A28) #3337; un-superseded/held-truth fresh-reader cells; held-truth isPending)
• Brotli: all signals caps pass (core floor 8,806 of 8,850; verdict and supersession changes live outside the floor)

Co-authored-by: Cursor cursoragent@cursor.com

Made with Cursor

[changeset-bot]

🦋 Changeset detected

Latest commit: e15b322

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 11 packages

Name	Type
@solidjs/signals	Patch
test-integration	Patch
@solidjs/web	Patch
@solidjs/babel-plugin	Patch
@solidjs/compiler	Patch
@solidjs/diagnostics	Patch
@solidjs/element	Patch
@solidjs/h	Patch
@solidjs/html	Patch
solid-js	Patch
@solidjs/universal	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[codspeed]

Merging this PR will not alter performance

✅ 172 untouched benchmarks

---

_{Comparing test/oracle-stores (e15b322) with next (0d8347a)}

[coveralls]

Coverage Report for CI Build 34955159006

Coverage remained the same at 71.898%

Details

• Coverage remained the same as the base build.
• Patch coverage: No coverable lines changed in this PR.
• No coverage regressions found.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

No coverage regressions found.

---

Coverage Stats

Relevant Lines:	1007
Covered Lines:	772
Line Coverage:	76.66%
Relevant Branches:	790
Covered Branches:	520
Branch Coverage:	65.82%
Branches in Coverage %:	Yes
Coverage Strength:	15.1 hits per line

---

💛 - Coveralls