[nerdalytics]

Description

Every state(obj, hooks) call on a non-array target ran one Object.defineProperty for the module-private HOOKS symbol, and nothing read it. A hooked non-extensible target was also written into a WeakMap that had no reader. Hooks reach every trap through the handler closures, so both writes were dead. This removes them and closes HOOKS_TODO.md item C1.

• The only runtime change is in hooksSlotFor, which now returns undefined for a non-array target. The array branch, definePropertyOnTarget and the internal-symbol guard are unchanged.
• The catch arm of defineProxyProperties stays. An extensible target whose defineProperty trap rejects symbols still reaches it, and tests/property-frozen-sealed.test.ts covers that path.
• No consumer-visible behavior changes. The symbol is module-private, hooks fire as before on extensible and non-extensible targets, and a non-extensible target is still registered in proxyCache.
• No version bump. The publish workflow is gated on a version change, so merging publishes nothing.

Changes Made

• src/index.ts: hooksSlotFor returns undefined for non-array targets, HooksSlot narrows to DirectArrayContext | undefined, and frozenHooksCache with storeHooksForFrozenTarget is deleted.
• scripts/benchmark.ts: a new state creation (hooks) case in the blind-spots group. It is its own commit, so the baseline below comes from the unchanged library.
• tests/state-hooks.test.ts: one test asserts that a hooked plain object owns one own symbol, a hooked array owns two, and a hookless object owns one. The object case fails on trunk.
• Prose that described the removed storage now matches the code, in src/AGENTS.md, src/hooks/AGENTS.md, tests/AGENTS.md, two test comments, and HOOKS_TODO.md, where C1 moves to shipped.

Performance Impact

• state creation (hooks) median went from 21.49ms to between 13.24ms and 13.39ms across five runs on Node v24.21.0, Apple M4 Pro.
• The gate is a 1% per-case ceiling on slowdowns. In two 10-cycle runs per commit, ordered baseline, final, final, baseline, 9 cases read over 1% slower in one pairing and under 1% or faster in the other (classic loop, state no subs, state + derive + 2 effects, effect triggers, derive chain depth 10, sealed write 1 sub, state no subs after seal, state write equal value, array push/pop (undefined)). The two baseline runs of identical code differed by more than 1% on 8 cases, so the suite cannot resolve 1% on them.
• Two cases read over 1% slower in both pairings, state creation (+1.4%, +1.1%) and array push/pop (hooks) (+2.0%, +1.1%). Neither path is in the diff. A targeted probe with 8 alternating processes per commit, 15 samples each, measured them at -0.59% and -0.37%, while the baseline's own processes spread 2.6%. The full tables are in the ticket reports under openspec/changes/drop-unread-hooks-slot-on-objects/tickets/.

Test Coverage

• npm test passes 308 tests. npm run test:coverage reports 100% branches, 100% functions, 100% lines.
• npm run check and npm run build pass.
• To verify by hand, call state(raw, { onRead() {} }) and check Object.getOwnPropertySymbols(raw).length. It is 1 for a plain object and 2 for an array.

Related Issues

None. Tracked as HOOKS_TODO.md item C1 and decided in .do/inbox/2026-10-25-drop-unread-hooks-slot-on-objects.md (Q80 a).

Checklist

• Tests added/updated for changes
• Documentation updated
• All tests passing
• Code formatting checked
• Performance implications considered
• Breaking changes documented (if applicable)