Skip to content

Batch 2026-11-22t: ADRs 1173, 1176, 1179, 1181, 1166 and 1180 (a props spread forward's sources; a caught read in a local closure; an exact forwarded wrapper; the parameter of a one-element map; a named callback's reporting run; a parameter default run by a call) - #318

Merged
yumemi-thomas merged 37 commits into
mainfrom
claude/batch-2026-11-22t
Nov 22, 2026

Conversation

@yumemi-thomas

@yumemi-thomas yumemi-thomas commented Nov 22, 2026 •

Copy link
Copy Markdown
Owner

Bundle t (Mac 24c), from main 01d5fdc (#317). Six accepted items, merged in the coordinator's order: ADR 1173 (claude/prop-member-dispatch-1173 a8810be, base a63e130), ADR 1176 (claude/caught-read-reaction-1176 df6549e, N5 only; its N8 attempt is reverted on the branch), ADR 1179 (claude/forwarded-wrapper-exact-1179 3c585cd, base a63e130) ADR 1181 (claude/array-callback-store-read-1181 ac7550e, base 960bfa1), ADR 1166 r3 (claude/named-callback-1166 48b48f1; r1 and r2 are on the same branch, none of the three is on main) and ADR 1180 (claude/param-default-1180 e549af2, base a63e130), plus my own commits. None of the four ran make verify, make corpus-delta or the fuzz smoke (1179's whole-grid mx-recall sample timed out with 0 programs judged); this bundle runs all of it, the census and rest legs of the smoke included, judges every moved-up program against Chrome and scans ADR 1181's store-map-* cells whole.

In the bundle (each accepted by the coordinator on evidence)

  • ADR 1173 (R4 N2): a component reached only through <Inner {...props} /> kept SC9012 closed-parameter-member-dispatch on props.setW1(7) and SC1001 read-backing-unproven on the prop read. For a component whose only escapes are such forwards (spread_of_own_props accepts each, props root current through its spreads), passed_sources takes each forwarder's call-site sources and prop_use/accessor_value_use prove a name static through the forwards; the reactive direction stays ADR 0919's. Author's probe (a63e130, mx-src and lg-src x 6 codes, 68,544 programs): mx-src 78 recall-gap -> agree, lg-src 273 U -> clean, no violation lost. Fixture prop-spread-forward-sources (Body SC2001 violation), Chrome case covered-violation. spread-prop-backing StaticOuter and competing-check-sync-memo lose an SC1001 U row (proven static).
  • ADR 1176 (R4 N5): a caught read in a local closure the computation calls in place is the computation's: SC5008 violation. Probe vs a63e130 (1,266 recall programs, every finding of every code): 14 programs, each one added SC5008 violation (12 lose the site's caught-read-unproven row); Chrome rc.21 CAUGHT_REACTIVE_READ at both violations of caught-read-kept-closure. N8 (a read in a createReaction closure) was tried and reverted by the author (4 V -> U): not built.
  • ADR 1179: ADR 0200's forwarding (forwarded_compute_slot) named its wrapper through function_for_symbol / function_called_at (the first function of a let written later, the literal handed to a factory whose result the name holds), so where the function the call runs never calls the literal main claimed a violation Chrome never reports: 3 fixture false positives (forwarded-wrapper-exact DropApp, FactoryDropApp, MemberDropApp, SC5001) and 6 legal generator cells (barm-fwdwrap-let, barm-fwdwrap-factory, SC2001 at a write that never runs). Now the wrapper is project_callee_at_parameter's exact answer (ADR 0278); the member census also follows a data slot naming a function declaration. Named fail-closed V -> U, ruled by the coordinator: loading-member-callee WrapLet and ViaWrap (SC5003, regress.json rows), forwarded-wrapper-exact SwapApp, FactoryApp, MemberSwapApp and PassApp (SC5001); the follow-up (every value the binding can hold is an exact function that calls the literal) is in open-work. Joined with ADR 1155 r4 (bundle s) by the author: no coverage moves.
  • ADR 1181 (R5, G6): in [s].map((x) => x.v) (and forEach, filter, some, every, find) the Get through the callback's parameter had no SC1001 row, so the value-use census kept an SC9012 value-in-carrier row at the literal. Step 1 (local_access::array_callback_parameter): the parameter of a one-element literal's standard inline callback is the store, so x.v takes the store's read row (callback-runs-1109 MapParamMember becomes the violation its README says Chrome reports; new fixture array-callback-store-read). Step 2 (G6 step 3, value_use::carried_parameter_rowed): where every use of the parameter is a Get a read row stands on, the census follows the position instead of opening it; a census silence with no inventoried arm, its witness pair the generator's store-map-member cells. Author's probe (SRU store-map-member, -unread, -apart): step 1 mx 238 recall-gap -> agree (each one added SC1001 violation), no row removed, lg no violation added; step 2 lg 227 U -> clean (1,434 value-in-carrier rows gone), mx 3,503 such rows gone with no outcome moved.
  • ADR 1166 r3 (the branch carries r1, r2 and r3; r1 was dropped from bundle q for two census false cleans, r2 from bundle s for the kui corpus finding): a named callback's read is placed by a reporting run (a direct call, a forEach/map carrier): the reporting run places strict reads only (r2: the generic timing paths stay main's, so SC2004/SC2005 keep their rows), and r3 reduces the kui feature-consumers/src/GroupRoute.tsx:80 finding (a callback declared in an effect's apply, run directly there and again from a timer inside itself) to fixture reporting-run-apply-timer, which Chrome confirms (STRICT_READ_UNTRACKED, covered-violation, nothing static-only). The corpus therefore gets new violations by design (the two SC1001 rows at that line in kui feature-consumers and shell; corpus-changes.json has no row kind for an added violation, so ADR 1166 records it). Coverage: construct-only-class:149 and named-callback-escaped-entry 81/89/97 become SC1001 violations (Chrome covered-violation), accessor-provenance:16 stays U. Named rise of foreach-carrier (SC1001 read-backing-unproven 0 to 4); a stated boundary gap for the reporting-run arm.
  • ADR 1180: a direct call with no spread that passes fewer arguments than a parameter's index runs its default on its own stack; a default that is a whole call (flush, a creation, onCleanup, a helper) is walked as a body call, guaranteed unless an earlier default calls (its span is exempt from ADR 0454's parameter-list demotion; a passed argument proves nothing). Author's probe (the 7 leaf/PD codes, release against a63e130): mx-recall 450 recall-gap -> agree, 0 V -> U, 0 clean -> U; lg-recall 924 unchanged. Fixture param-default-leaf with a Chrome case: creation and cleanup covered-violation; the 3 flush violations throw errors that carry no frames, so the judge ledgers them as attribution (Chrome confirms the throw, not its line).
  • Facts schema stays 108 (none of the six touches it).

Moves, read against main 01d5fdc

  • Coverage (make coverage, 1,780 fixture projects, 40,841 findings): nothing moved after the merge; the proof gate and the five CI guards pass on the joined tree (count 310 keys / 37,889 rows / 500 drops, rises 3 of 3 named, regress 1 of 1 named, base and proved 0 problems). The ten existing snapshots the bundle changes, read row by row (snapdiff2.mjs), each belongs to one branch: ADR 1173: competing-check-sync-memo and spread-prop-backing lose one SC1001 read-backing-unproven row each (proven static). ADR 1179: loading-member-callee WrapLet and ViaWrap SC5003 V -> U (named in regress.json; the two sites also gain SC1001 function-run-unplaced, the named rise 19 to 21; a byte shift from corrected comments). ADR 1181: callback-runs-1109 gains the SC1001 violation Chrome reports at MapParamMember and loses its three value-in-carrier rows. ADR 1166 r3: construct-only-class, destructure-store-slots (SC1003), effect-compute-also-called, named-callback-escaped-entry (6 rows) and named-fn-expr-calls each turn function-run-unplaced U rows into violations (one for one), and foreach-carrier relabels four function-run-unplaced rows as read-backing-unproven (the named rise 0 to 4). New fixtures: prop-spread-forward-sources (1173), caught-read-kept-closure (1176), forwarded-wrapper-exact (1179: SwapApp, FactoryApp, MemberSwapApp, PassApp V -> U, named), array-callback-store-read (1181), named-callback-reporting-run and reporting-run-apply-timer (1166), param-default-leaf (1180: creation, cleanup and flush V).
  • Fuzz cover scan, code level (checker only, main's release against this build, the smoke's rest-leg cover, 45,634 programs; compared by site class and by the uncertifiable kinds at the site): 40,284 identical, 5,350 changed. U -> V 3,957 (mx-recall STRICT_READ_UNTRACKED: ADR 1166 r3's named-callback reads placed by a reporting run, plus one ADR 1181 carrier program), U -> clean 3, V -> U 6 (lg-recall barm-fwdwrap-let REACTIVE_WRITE_IN_OWNED_SCOPE: ADR 1179's legal cells, the false positives gone). 1,345 programs keep class V with a function-run-unplaced row replaced by one more violation (ADR 1166 r3); the rest are relabels at a site that stays uncertifiable (class kept). The 3 U -> clean: two mx FLUSH_IN_TRACKED_QUEUE_CALLBACK pd- programs (ADR 1180: the violation stands at the flush in the parameter default, a remote site the scan does not look at: the judge below places it) and one lg-recall legal program. Oracle judge (Happy DOM against Chrome, --runtime-diff all) on all 3,960 moved up: mx-recall 3,959 agree, lg-recall 1 clean, 0 false positive, 0 unexecuted, 0 behaviour, attribution or inconclusive differences.
  • ADR 1181's whole store-map- cells* (the coordinator's check on its census silence; mx-recall 5,782 and lg-recall 4,257 programs, not just the cover): mx-recall 5,440 U -> U and 342 U -> V; lg-recall 3,875 U -> U, 328 U -> clean, 54 clean -> clean. No misuse program is clean in the bundle (0, and 0 moved to clean from V or U); the legal programs that became clean are the ADR's step 2.
  • Smoke, census leg and rest leg exactly as CI runs them (focus.mjs --smoke --smoke-leg census|rest --cover all --cover-seed 1 --sample 0): census exit 0, nothing needs a look (44,641 programs, 69 more than main's: ADR 1179's two generator contexts barm-fwdwrap-let and barm-fwdwrap-factory, all not-fired in the census; every other program keeps main's outcome, 10 agree programs change their findings only: agree 7,723, recall-gap 5,419; 0 worse, proved verdicts 37 lost and all 37 named, 719 newly proved); rest exit 0, nothing needs a look (mx-recall 34,294 programs: agree 11,909 where main had 7,665, recall-gap 22,053 where it had 26,297; 12 lost, 12 named; lg-recall 1,913 clean where main had 1,912).
  • monotone (textbook and fixtures, 1,801 programs): 0 programs U -> certified; the 6 programs that moved down or sideways are all named: forwarded-wrapper-exact SC5001 V -> U x3 and loading-member-callee SC5003 V -> U x2 and SC1001 clean -> U (ADR 1179's named fail-closed moves), the rest are new fixtures' sites (clean -> V).
  • corpus-delta --all (rc.21 frozen v2, 34 projects, old = main's release): violations 243 -> 247, by design, all four rows ADR 1166 r3's: kui/.../feature-consumers and shell, SC1001 createEffect apply callback at GroupRoute.tsx:80 (2 per project), uncertifiable -> violation; Chrome covers the shape (fixture reporting-run-apply-timer, STRICT_READ_UNTRACKED covered-violation). corpus-changes.json has no row kind for an added violation, so ADR 1166 records it. Needs a look: exactly those two projects. Uncertifiable rows -17 (471,595 -> 471,578): function-run-unplaced -8, closed-parameter-member-dispatch -7 (ADR 1173), pending-read-unresolved -3, callee-callback-timing -2, prop-pending-unproven -2, read-backing-unproven -1, value-to-unresolved-callee +6. The removed uncertifiable rows, checked (24 rows: app-game 20, kui 4): the 4 kui rows are the two SC1001 function-run-unplaced rows per project that became ADR 1166 r3's violations. Of app-game's 20, 15 are relabels (the rule keeps a row at an overlapping span with another blocker) and 5 SC1001 rows have no SC1001 row left at their site: popover.tsx :108 and :124 and navigation-popup.tsx :38, :44 and :50, each a read of a prop (props.navigationIsActive and the like) inside a callback of a component that is rendered through spread forwards, the target of ADR 1173 (its closed-parameter-member-dispatch rows go at the same spans). Every one of the five sites stays uncertifiable: SC2002 and SC9012 (value-to-unresolved-callee or leaf-callback-opaque) rows stand there, the shape the coordinator accepted for Batch 2026-11-22s: ADRs 1174, 1177, 1169 and 1155 r4 (refresh of a prop; a context read in a handler; a store link held in a binding; a forwarded leaf operation) #317's eleven. The proof log shows one new name at the five (local_access.discover_member.props-static, ADR 1173's static-prop proof); dropping it (SOLID_CHECKER_IR_ABLATE_SILENCE) adds 1,290 strict-read-unaccounted rows in app-game but none at these five sites, because each is already covered by an SC9012 census row, so no single switch brings the five rows back. Not a lost violation (corpus-guard: 0 lost), not a place that became clean.
  • corpus-guard (ADR 1012, both builds): 0 violations lost, 0 places newly uncertifiable, 4 violations added (ADR 1166 r3's), uncertifiable rows +1 where some stood and -18; nothing to name.

My commits

  • Merges of the six branches in the coordinator's order (no Rust conflict anywhere; content conflicts only in docs/decisions.md, docs/open-work.md, scripts/guard/counts.json, rises.json, rediscovery.json and scripts/fuzz/smoke/mx-census.json). counts.json: main plus each branch's own delta against its own base (a63e130, 960bfa1, b600297), additive, and the branches' own drops rows; count.mjs passes without --update.
  • Hand merges: docs/open-work.md for ADR 1179 (its accepted-trade line replaces the obsolete ADR 0200 hazard item; ADR 1169's rewritten alias lines stay); rises.json by a JSON-aware resolver (ADR 1179's two rows, ADR 1166's one row); scripts/fuzz/smoke/mx-census.json by a decoding merge (the manifest stores expected as indexes into the sorted behaviours): main's recorded 36 expectations kept, the two ids ADR 1166 r2 adds appended.
  • Both ADR 1166's earlier rounds (r1: bundle q, r2: bundle s) are held-out history on the same branch; r3 is the reduction of the finding that held r2 out.

Merge notes

  • No branch adds a symlink (checked against main after every merge). ADR 1179's open-work line is the only judgement call in a conflict; everything else was mechanical.

Checks (this Mac)

On the final tree 8dda228:

  • cargo check --workspace --all-targets (0 warnings), make coverage, make proof-gate, the five CI guards, make textbook-check 21 of 21: pass.
  • CI's scripts-and-cli steps (sh -n scripts/*.sh, vitest scripts/*.test.mjs 548 tests, bun scripts/fuzz/test-lane.mjs 1,940 fuzz tests): pass.
  • Cover scan, oracle judge, the store-map cells, census and rest legs of the smoke, monotone, corpus-delta, corpus-guard: above.
  • make verify: passes in full, verify-performance included (enforced gate, powermode 0, load 11.0): best first IR 163,393 ns/source against the 175,000 ceiling. Same-sitting A/B against main's benchmark binary: not run: the enforced gate itself passed at powermode 0 (above), and the A/B only stood in for it while the machine was in powermode 1.
  • wasm owned tests with a FIFO in TMPDIR: 10 files, 80 tests, with and without a FIFO.
  • Deferred to CI: everything on the runners (rust-engine, rust-package, wasm, smoke, smoke-census, scripts-and-cli).

🤖 Generated with Claude Code

yumemi-thomas and others added 30 commits November 21, 2026 13:27
…ect call, forEach/map carried)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…U->V where Chrome warns), map carrier docs

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… the declarator and the run (a creation in an apply throws first)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…twin

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…d drops rows

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… cleans)

Bundle q's census smoke found two round-1 false cleans
(mx-{resolve,until}-in-reactive-scope-barm-carried-body-x-direct-none):
`const f1 = () => { resolve(() => 1); }; createEffect(f1, () => {});
[f1].forEach((g) => g());` lost SC2004/SC2005's function-run-unplaced row.
The reporting run sat in shared code: callee_callback_timing (read by
resolve_scope and every rule) took the untracked carried run as the
function's timing, and read_execution_role gave every caller the reporting
role. One untracked run is a witness only for SC1001's warning; a throwing
or tracked-run rule keeps every run's role. Now strict_read_role and
strict_read_callee_timing apply only where local_access builds a
ReactiveRead (five row sites and member_read); the generic paths are
main's; the member path hands its async row the unplaced role.

Checks: fixture twins TwinResolveComputeCarried/TwinUntilComputeCarried
(SC2004/SC2005 kept; r1 build silent), round-1 fixture rows unchanged;
Chrome case named-callback-reporting-run-tracked (both throws in the
console). Coverage: construct-only-class:149 and named-callback-escaped-
entry 81/89/97 become SC1001 violations (Chrome covered-violation),
accessor-provenance:16 back to U. Smoke, release, --cover all --cover-seed
1 --sample 0, the smoke's judge: census 44,471 and rest legs, nothing
needs a look (0 false-clean; census 3 known-open FPs). Round 1's 3,943
mx-recall agree: 3,567 kept, 376 back to recall-gap (flush-read-body*,
pos-class-*: named cost). Boundary gap for the arm, named rise of
foreach-carrier, rediscovery entries (both ids added to the census smoke
manifest). Ownership gate 96 passed, verify-delta passed; make verify
deferred.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…8bde5d merged into r2's entry)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…spatch-1173

# Conflicts:
#	docs/decisions.md
#	scripts/guard/counts.json
… proven (N5)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…k takes the layer call's role (N8); fixture reaction-closure-read; boundary gap

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… callback takes the layer call's role (N8); fixture reaction-closure-read; boundary gap"

This reverts commit aebaf2a.
…s in place is the computation's (R4 N5)

Probe vs main a63e130 (1,266 recall programs, every finding of every code): 14 programs, each one added SC5008
violation (12 lose the site's caught-read-unproven row); nothing else moved. Chrome rc.21 on caught-read-kept-closure:
CAUGHT_REACTIVE_READ at both violations. N8 (createReaction closure) tried and reverted: 4 V->U through the call-site
summary's timing proof; stated as not built. No new role arm.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ic props

N2 of the R4 catalog: a component reached only through `<Inner {...props} />`
kept SC9012 closed-parameter-member-dispatch on `props.setW1(7)` and SC1001
read-backing-unproven on the prop read. For a component whose only escapes
are such forwards, each one spread_of_own_props accepts (props root current
through its spreads; SpreadForwards::checked), passed_sources takes each
forwarder's call-site sources (witnesses; complete when every forwarder's is
and no forward sets the name), and prop_use/accessor_value_use prove a name
static through the forwards (forwarded_use). The reactive direction stays
ADR 0919's resolve_forwarded_props (a first cut that also carried reactive
witnesses skipped its getter checks: 7 spread-prop-backing reads, fixed).

Checks: fixture prop-spread-forward-sources (Body SC2001 violation, Apply
silent, Override and Rest keep rows); Chrome case (covered-violation, main
covered-uncertifiable). Fuzz on a63e130, mx-src + lg-src x 6 codes (68,544
programs): mx-src 78 recall-gap -> agree, lg-src 273 U -> clean, no
violation lost, nothing needs a look. Coverage: spread-prop-backing
StaticOuter and competing-check-sync-memo lose an SC1001 U row (proven
static); guard drops recorded; rises 0. Boundary test 25 passed (no new
arm); proof-lint allowlist stated; ownership gate 96 passed; verify-delta
passed. make verify deferred. Not a defect fix: no rediscovery entry.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…d's inline callback is the store; fixture

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…parameter)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…nd's inline callback is the store, so x.v gets the store's read row

Exact callee (default-library Array.prototype member, no impostor or patch veto), a one-element spread-free array literal
naming a store, a plain never-written parameter of an inline non-async callback; anything else keeps its old answer.
Probe vs main 960bfa1 (SRU store-map-member/unread/apart, mx 1,441, lg 1,427), every finding of every code: mx 238
programs recall-gap -> agree, each one added SC1001 violation; no row removed anywhere; lg no violation added; the rest
added U rows. Chrome rc.21 on array-callback-store-read: STRICT_READ_UNTRACKED at the three violations, quiet in the memo.
Coverage: callback-runs-1109 MapParamMember gains its SC1001 violation (Chrome warns, README F2).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e every use is a Get a read row stands on

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…by Chrome

The corpus delta's 2 new SC1001 violations (kui GroupRoute.tsx:80) are
the apply-plus-timer shape: Chrome warns STRICT_READ_UNTRACKED
(covered-violation, nothing static-only). Kept; fixture
reactive-ir/reporting-run-apply-timer and Chrome case of the same name.
corpus-changes.json takes no added-violation kind, so the ADR records it.

Checks: make coverage (only the new fixture), ownership gate, boundary
vitest 25/25, guard count updated (new fixture rows).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ns exactly (SC5001/SC2001 false positives)

forwarded_compute_slot named its wrapper through function_for_symbol /
function_called_at: the first function of a `let` written later, or the
literal handed to a factory whose result the name holds. Where the function
the call runs never calls the literal, main claimed a violation Chrome never
reports: forwarded-wrapper-exact DropApp, FactoryDropApp, MemberDropApp
(SC5001), generator barm-fwdwrap-{let,factory} (SC2001, 6 legal cells).

Now the wrapper is project_callee_at_parameter's exact answer (ADR 0278): a
declaration, a const or unwritten let bound to the function, a member the
census resolves, no spread before the literal. The member census
(object_members::memo::literal_function) also follows a data slot naming a
function declaration (`{ hand: handDecl }`), as it follows the shorthand.

Fail-closed V->U, each Chrome-reported (accepted): loading-member-callee
WrapLet, ViaWrap (SC5003, ASYNC_OUTSIDE_LOADING_BOUNDARY; regress.json row);
forwarded-wrapper-exact SwapApp, FactoryApp, MemberSwapApp (SC5001,
PENDING_ASYNC_UNTRACKED_READ) and PassApp (SC5001, FORBIDDEN_SCOPE). Follow-up
in open-work: every value the binding can hold is an exact function that
calls the literal.

Fixture forwarded-wrapper-exact (Chrome runtime-oracle/forwarded-wrapper-exact):
Decl, Const, Let, Member (method), MemberData (data slot) V and Chrome-covered;
Drop twins U, Chrome silent; DataSwap, DataReplaced U (silent); DataArrow U
(Chrome reports: a slot naming a const arrow is not followed).
Generator: arm-shapes barm-fwdwrap-let, barm-fwdwrap-factory (legal,
REACTIVE_WRITE_IN_OWNED_SCOPE); rediscovery.json row (main 6 unexecuted
violations, this build 6 uncertifiable).

Probe (release, Chrome cache), main a63e130 vs this: fwd contexts (lg 9,
mx 6, census 210) only the 6 new cells move; all of lg-recall (97,176 legal)
only those 6 move. Coverage: loading-member-callee WrapLet/ViaWrap (+ a byte
shift from the corrected comments); nothing else, the data-slot route included.

Checks: make coverage + update, ownership gate (96), guard count/rises/regress,
make verify-delta (passed), boundary.test.mjs via vitest (25 passed).
Not run: make verify, corpus-delta, fuzz-smoke. mx-recall whole-grid sample
timed out (0 programs); its fwd cells are in the focused probe.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… parameter's every use is a Get a read row stands on

Probe vs main 960bfa1, every finding of every code: lg-recall 227 programs uncertifiable -> clean (1,434 value-in-carrier
rows gone), mx-recall 3,503 such rows gone with no outcome moved (no misuse program clean). Coverage: only
array-callback-store-read and callback-runs-1109 lose their value-in-carrier rows at the clauses whose reads have rows;
guard drops rows; ADR entry and rule page.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… in the leaf walk

A direct call with no spread that passes fewer arguments than a
parameter's index runs its default on its own stack. A default that is a
whole call (flush, a creation, onCleanup, a helper) is walked as a body
call, guaranteed unless an earlier default calls. Its span is exempt from
ADR 0454's parameter-list demotion. A passed argument proves nothing.

Fixture reactive-ir/param-default-leaf, Chrome case of the same name
(creation and cleanup covered-violation; flush throws carry no frames,
ledgered as attribution). Fuzz on the 7 leaf/PD codes, release vs main
a63e130: mx-recall 450 recall-gap->agree, 0 V->U, 0 clean->U;
lg-recall 924 unchanged.

Checks: make coverage (only the new fixture), ownership gate, boundary
vitest 25/25, guard counts updated.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…' into claude/batch-2026-11-22t

# Conflicts:
#	scripts/guard/counts.json
yumemi-thomas and others added 7 commits November 22, 2026 09:11
…' into claude/batch-2026-11-22t

# Conflicts:
#	docs/decisions.md
…179' into claude/batch-2026-11-22t

# Conflicts:
#	docs/decisions.md
#	docs/open-work.md
#	scripts/fuzz/rediscovery.json
#	scripts/guard/counts.json
#	scripts/guard/rises.json
…-1181' into claude/batch-2026-11-22t

# Conflicts:
#	docs/decisions.md
#	scripts/guard/counts.json
scripts/guard/counts.json as exactly main 01d5fdc plus, per branch, its own delta against its own base (a63e130 for 1173, 1176 and 1179; 960bfa1 for 1181),
with the branches' own drops rows: node scripts/guard/count.mjs passes without --update (309 keys, 37,787 uncertifiable rows, 492 drops recorded).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… claude/batch-2026-11-22t

# Conflicts:
#	docs/decisions.md
#	scripts/fuzz/rediscovery.json
#	scripts/fuzz/smoke/mx-census.json
#	scripts/guard/counts.json
#	scripts/guard/rises.json
…claude/batch-2026-11-22t

# Conflicts:
#	docs/decisions.md
#	scripts/guard/counts.json
…r3 and 1180

scripts/guard/counts.json as exactly main 01d5fdc plus, per branch, its own delta against its own base (a63e130 for 1173, 1176, 1179 and 1180; 960bfa1 for 1181;
b600297 for 1166), with the branches' own drops rows: node scripts/guard/count.mjs passes without --update (310 keys, 37,889 uncertifiable rows, 500 drops recorded).
Replaces the four-branch counts of the previous commit.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@yumemi-thomas
yumemi-thomas merged commit 0751069 into main Nov 22, 2026
14 of 15 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant