fix(signals): A28 at the store backing — an unflushed write is invisible to owner-context readers until the carrying flush (S6)

[ryansolid]

Stacked on #3525 (← #3523). The S6 fix, alone, so its perf and bytes are judged apart from the consolidation.

The bug (S6)

Core serve() serves an unflushed node's committed value and re-runs the reader in the carrying flush (A28). The store's backing selection served the pending backing to every owner-context reader at once. A derivation created inside a foreign action that adopted a mainline setStore published the unflushed 1; the signal published 0. Ruled 2026-09-17: store rules follow signal rules.

The fix

• ensurePB stamps the target with clock when the draft opens outside a flush from outside any owner — A28 (4)'s promoted-write exemption (a projection derive's writes stay visible to its block).
• pendingBackingVisible's owner-context branch serves committed for a backing stamped this tick and marks the reader late linker.
• getNode gains a third born kind: a node born in the unflushed window stages the pending backing's value, so the carrying flush commits it through the node and the late-linked reader finds it there. stageHeldKey generalizes to one stageKey(node, nv, txn | null).
• deep() / snapshot() keep their speculative-peek contract; nodeValue already has the arm through serve().

Gate

check	result
signals suite	3011 pass; S6 flips from recorded divergence to parity pin (three shapes, with [0] → [0, 1] at the commit)
oracles	identical
posture matrix	exactly the 12 S6 cells move, to the signal's values; all six signal/store pairs 0 diffs
size	+283 B minified store (was +402 at the old twin); one brotli cap ratcheted (+23 B)

Net with the consolidation PR: −30 B minified store, S6 closed.

Made with Cursor

[changeset-bot]

🦋 Changeset detected

Latest commit: 4be36d0

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

[coveralls]

Coverage Report for CI Build 35285034403

Coverage remained the same at 71.304%

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:	1035
Covered Lines:	783
Line Coverage:	75.65%
Relevant Branches:	798
Covered Branches:	524
Branch Coverage:	65.66%
Branches in Coverage %:	Yes
Coverage Strength:	15.06 hits per line

---

💛 - Coveralls

[codspeed]

Merging this PR will degrade performance by 5.7%

⚡ 2 improved benchmarks
❌ 15 regressed benchmarks
✅ 155 untouched benchmarks
⏩ 3 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	class store object, toggled in place × 500	38.9 ms	44.9 ms	-13.43%
❌	style store object, mutated in place × 500	39.7 ms	45.5 ms	-12.76%
❌	reconcile: deep tree, 10 of ~12k paths subscribed	3.2 ms	3.6 ms	-11.27%
❌	store reconcile tree shuffle: 1111 keyed nodes	32.4 ms	35.6 ms	-8.86%
❌	style store object, replaced × 500	43.9 ms	47.9 ms	-8.36%
❌	reconcile: deep tree, all ~12k paths subscribed	72.8 ms	79.4 ms	-8.3%
❌	store reconcile tree reverse: 1111 keyed nodes	29 ms	31.5 ms	-7.87%
❌	dbmon full tick — deep reconcile	681.2 ms	730.3 ms	-6.72%
❌	spread store record × 500	31.9 ms	34.2 ms	-6.71%
❌	dbmon full tick	696.9 ms	746.3 ms	-6.62%
❌	dbmon partial tick — deep reconcile	710.7 ms	760.9 ms	-6.61%
❌	dbmon partial tick	724.2 ms	771.9 ms	-6.19%
❌	steady state: owned backings (#3360)	8 ms	8.5 ms	-6.15%
❌	class store object, replaced × 500	41.7 ms	44.3 ms	-6.06%
❌	dbmon full tick — shallow reconcile	533.2 ms	561.6 ms	-5.06%
⚡	storeWrap:shallow 1k records	4.5 ms	3.6 ms	+24.2%
⚡	storeWrap:deep 1k records	12.3 ms	11.6 ms	+5.44%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing fix/store-a28-backing (4be36d0) with next (dd908fc)}

Footnotes

1. 3 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[ryansolid]

Not shipping for now. The fix is right by the ruling (store follows signal) but needs one bit of per-target state — "was this backing opened this tick, outside a flush, by imperative code" — and every home for it regressed a hot path on CodSpeed:

stamp	write floor	dbmon
WeakMap per trap write	−7…−19%	flat
Map per draft open	−7…−9% (set/delete churn)	flat
target field uf	~flat	−5…−7% — the 24th named field flips array targets to dictionary mode on CI's V8 (the target.ts shape rule holds there; it did not reproduce on Node 26 locally)

Untried: a symbol on the pb container (free on the hot paths, but the symbol rides into the committed object and every Reflect.ownKeys over a backing has to skip it). Not worth the surface for this shape — an imperative write read by a derivation created inside a foreign action before the flush.

S6 stays pinned on next at the store's current value in posture-store-parity, ruled store-follows-signal and deferred; it closes for free if the store's value selection ever moves to one home, where the node carries the stamp. The branch stays for reference.

— Claude via Cursor