refactor(signals): affects() dedicated channel + delete stashedOptimisticReads (arch audit tier-1)

[ryansolid]

The two Tier-1 architecture consolidations that survived design-spike measurement (spikes 1 and 3 were declined — pull-verdicts is spec-pinned by the #2831/lane-held wake schedule and measured size-neutral; store node-authority costs +585 min). Each of these retires a mechanism family, not a line.

	Δ full (min/gz)	Δ floor (min/gz)
Combined	−775 / −304	−388 / −169

affects() off the async status rails

A mark becomes a refcount on the marked node(s) plus a pull-computed coverage walk at verdict time; Loading interaction survives as a boundary-only visual notification. Because marks never touch _statusFlags/_error/_pendingSources/_asyncReporters, the five carve-out families that existed only to un-teach the async rails everything a mark does — value-transparency, transaction-inertness, error-inferiority, settle-release, quiet-reask immunity — are deleted as unrepresentable rather than maintained. The per-read affects collection and per-recompute mark bookkeeping also leave the core read/recompute hot paths (a small simplification of the paths Sierpinski exercises).

Designer rulings encoded as the channel's structural contract: marks are invisible to all completion/settlement accounting by construction — a mark can never block or legitimize ending its own transaction (the action holding the window is the real work); ambient marks (no transaction) are verdict-only and release at flush end; a mark serializes only inside a resumable transaction. Hydration capture now takes marked-covered nodes' real committed values (they carry no status), consistent with value-transparency.

delete stashedOptimisticReads

It forced a committed-view effect re-run when a transaction backgrounded — masking an active override from stale tracked reads (against A17's "do not mask the override from any read path") and un-rendering co-written optimistic flags mid-window — while engaging zero times across all prior tests. The queue stash and _gatedSubs replay are retained: they cover transaction-held reveal and silent-revert re-ask respectively, which are irreducibly distinct reveal actions; a spec comment records why they cannot physically merge (the design-spike negative result).

Tests

1077 pass (1073 prior, unchanged, + 4 new pins):

• S3 — the backgrounded-transaction reveal shape: co-written optimistic flag stays rendered, user effects fire once.
• S4 — a transaction-held effect must not leak during an interleaved ambient flush. This was a pre-existing coverage hole (the suite never interleaved ambient work with a stash); the queue stash is load-bearing and was previously untested.
• isolation — deterministic recompute-count guards for the propagation/Sierpinski surgical-update property (the isolation CodSpeed's propagation:* benches measure by time; this asserts it by count, noise-free).

CodSpeed is the arbiter on hot-path shape — the coverage walk is O(upstream reachability) per marked-node verdict (was O(pendingSources)); flagged for the perf run. Every prior spike round taught that local wall-clock can't resolve these deltas, so the benches decide.

🤖 Generated with Claude Code

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[changeset-bot]

🦋 Changeset detected

Latest commit: 1035c9c

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

This PR includes changesets to release 9 packages

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

[github-actions]

size-limit report 📦

Path	Size
signals: core floor (createSignal/Memo/Effect/Root/flush)	6.5 KB (-1.86% 🔽)
signals: + createStore	11.08 KB (-1.16% 🔽)
signals: + isPending/latest	8.1 KB (-0.95% 🔽)
app: render + one signal (the simple-app floor)	9.07 KB (-1.44% 🔽)
app: CSR with Show/For/Loading/Errored/lazy	11.09 KB (-1.04% 🔽)

[coveralls]

Coverage Report for CI Build 29783820848

Warning

No base build found for commit 2bf022e on next.
Coverage changes can't be calculated without a base build.
If a base build is processing, this comment will update automatically when it completes.

Coverage: 75.328%

Details

• Patch coverage: No coverable lines changed in this PR.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

Requires a base build to compare against. How to fix this →

---

Coverage Stats

Relevant Lines:	778
Covered Lines:	626
Line Coverage:	80.46%
Relevant Branches:	592
Covered Branches:	406
Branch Coverage:	68.58%
Branches in Coverage %:	Yes
Coverage Strength:	14.11 hits per line

---

💛 - Coveralls

[codspeed-hq]

Merging this PR will regress 1 benchmark

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 3 improved benchmarks
❌ 1 regressed benchmark
✅ 117 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	merge	266.5 µs	369.6 µs	-27.9%
⚡	merge	165.2 µs	73.5 µs	×2.2
⚡	reconcile: deep tree, 10 of ~12k paths subscribed	6.4 ms	3.3 ms	+92.3%
⚡	updateSignals:update1to1	54.4 ms	51.5 ms	+5.56%

Tip

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

---

_{Comparing arch/tier1-affects-reveal (1035c9c) with next (2bf022e)}

[ryansolid]

CodSpeed: net strongly positive; the one regression is a shared-VM artifact

⚡ 3 improved · ❌ 1 regressed · ✅ 117 untouched

Benchmark	BASE → HEAD
merge-merge-static(both 5,1,3,2)	165.2 → 73.3 µs	×2.3
reconcile: deep tree, 10 of ~12k paths	6.4 → 3.3 ms	+92%
updateSignals:update1to1	54.4 → 51.5 ms	+5.6%
merge-mixed(0,100,3,2)	266.5 → 369.6 µs	−27.9%

Attribution (two isolation probe PRs, now closed)

• stashedOptimisticReads deletion alone → clean, gate passed, 1 improvement.
• affects-channel alone → reproduces the merge pair exactly: merge-mixed −27.9% and merge-static ×2.3. They are one coupled effect of simplifying read()'s shape.

Why merge-mixed is a measurement artifact, not a cost

merge-mixed builds objects whose properties alternate between a plain value descriptor ({value: 1}) and a plain-JS getter ({ get() { return 1 } } — no signal, no Solid primitive). It exercises merge's property-descriptor resolution over getter/value polymorphism; it never calls read(), link(), or any code this diff touches. So the −28% cannot be a per-operation cost from the change — it is V8 retiering the shared merge bench file's megamorphic descriptor-dispatch site because read() (called by other benches in the same process) changed shape. merge-static ×2.3 is the coupled other half of the same recompilation; the two always move together and neither reflects a real cost in the merged code.

Acknowledging merge-mixed on that basis. The real ledger — a 2.3× and a 92% improvement on paths the diff does touch, against zero reachable regressions — plus the architecture win (five affects carve-out families deleted as unrepresentable; the affects fix-cluster retired) is the reason to land.