perf(ts): map remaining JSDoc hotpaths after class-owner indexing

Evidence and scope

Follow-up to #416/#853. The landed source ac34883 (native SHA-256 fb77a18ed4b9b14a7c0ef29366963166754de8904602c2516ea2d223389cbb82) replaces both originally mapped repeated scans with complete syntax facts. Exact-parent JSDoc confirmation improves medians by 5.3% officially, 16.1% at 512 families and 34.0% at 8192, with all data retained. The full five-compiler matrix is being refreshed; all-benchmark leadership remains unproven. Large checked-JS projects still take roughly 12 seconds.

Profile the actual remaining work on this landed source, with a diagnostic-only image whose instruction bytes match the native build, complete positive/negative admission, source/input/image fingerprints, raw early and middle samples and precise source-level mapping. Never use diagnostic images for timing or call small samples whole-run CPU percentages. Choose subsequent optimizations from measured repeated work, not source/library/benchmark filenames, numeric semantic limits, approximations or skipped diagnostics/bodies.

Acceptance for subsequent changes

Complete ownership/invalidation and OOM behavior; representative full diagnostic/scalar-oracle controls; all checker/Program/driver/CLI/entrypoint and strict checked-JS admission gates; unchanged pinned Zod comparison; exact-parent official/scaling confirmation and full-family/adverse-risk rechecks retaining every observation; publish accepted or rejected results with raw checksum evidence, brief README tables and issue-linked Chris-only conventional main commits. This issue does not claim compatibility of Zod, an admitted C++ compiler, other-platform coverage or universal leadership.

The new landed-source diagnostic-only profile is now queued behind the fresh five-compiler matrix, under the same shared 600s/3840 MB guard. Its prepared runner requires a clean accepted-source tree, exact native fb77a18, original/renamed instruction-section byte identity, all 58 official binary/image admissions plus positive/negative 8192 controls (62 total), stable source/input/binary provenance and separate one-second samples at early and middle offsets. It will retain raw outcomes if the process exits before a declared sample rather than silently choosing another offset. No diagnostic image supplies timing data. No optimization or performance claim is made under this new issue yet.

Current landed-source profile and full matrix are published in4bbbd191d83df1e8c6f44c8b0bd96a9490f57e0c. Diagnostic image8c2ae47e passes62 admissions and preserves original/renamed instruction sections exactly (textSHAa03c8bf8); nativefb77a18e and inputbfb2fb1b remain unchanged. Early/middle separate1s samples have805/822 main-thread observations, with172/261 leaf observations injsDocTypeForPreviousIdentifierDecl,119/131 virtualSectionStartForNode,57/75 require-binding checks. These are NOT whole-run percentages or timing samples. Disassembly maps offsets400/416/432 fromsymbol0x1003a8278 to the inlined nearestPriorValueDeclInSameFunction full-HIR scan: var/let/const kinds3/4/5, earlier/nearest span, identifier/name match, function/section eligibility. Raw source/assembly/samples/images instructions/62 outcomes retained in24-checksum archive. Next action: inspect existing declaration indexes and reuse complete owned syntax facts while preserving those exact eligibility/tie rules, OOM fallback and mutable type resolution; no type-result/source-name shortcuts. Report: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#landed-source-jsdoc-profile-2026-10-09-utc .

Starting the measured nearest-prior value-declaration optimization from exact landed parent4bbbd191d83df1e8c6f44c8b0bd96a9490f57e0c/nativefb77a18e. Existing per-container local-value facts are NOT equivalent: they select first bindings (including functions/classes) and statement ordinals, while this query needs nearest earlier var/let/const, lowest NodeId ties, function eligibility and exact virtual-section checks. Plan a complete syntax-only name/position candidate index, not cached TypeIds or weakened visibility. Keep the original full scan on any construction failure, free every partial group, reset on source changes and rebuild if HIR node count changes. The existing outer-container JSDoc fallback and mutable type lowering remain unchanged. Full differential scope/position/tie/OOM/source-change controls and all module/admission/Zod gates precede frozen exact-parent timing; no source-name shortcuts, semantic caps or acceptance yet.

First draft now contains a lazy complete syntax candidate index by identifier name/source position, with binary search and exact reverse-order eligibility. It stores NodeIds only (no type results), frees every partial group on OOM, keeps the complete scalar fallback, resets on setSource and rebuilds after HIR growth. Six controls cover all identifier uses across scope/section/export/block/function/class shapes, all 3,087 synthetic ordering/position queries, lowest-NodeId ties, null-function recovery, source invalidation, real appended HIR declarations, every deterministic construction allocation failure and full checked-JS diagnostic slices. Original outer-container fallback/type lowering stays unchanged. Full checker/build validation started; complete initial draft/provenance retained separately. No source push, changed benchmark inputs or speed claim.

All full correctness gates now pass for the nearest-prior syntax index: checker4542, Program247, driver200, CLI74, entrypoint35, harness136. Native candidate69836df6bd81695ed3f78c2b147a7ec8ac908e5e406da5523b9da7ed3c4a895a against exact parent4bbbd191d/nativefb77a18e. Complete pinned Zod primary message/position/code multiset remains195 TS +3 HM9002 (zero added/removed), not production admission. Freezing the final diff before timing. Declaring all phases in advance: original official/512/8192 screen; independent30-pair official/512 confirmation; all20 families; independent30-pair checks for every >3%-slower or wholly-negative screen family plus carried predicates/reexports; separate class/interface risks; all five six-pair8192 confirmation batches with their own admissions/warmups. Each phase requires66 complete positive/negative admissions and unchanged fingerprints; fresh processes alternate order, all observations retained. No previous candidate samples mixed in, diagnostic images timed, guard relaxation, semantic cap, source-name shortcut or acceptance yet.

The complete screen passes all66 admissions with unchanged fingerprints: officialJS28.397 ->27.641 ms,16/20, saving CI[+0.069,+1.279]ms; 512 families128.661 ->118.902 ms,5/6,[+0.100,+17.005]ms; 8192 families13.895 ->9.468s,6/6,[+3.017,+7.355]s. All observations retained, including the adverse 512 pair and variable large parent. These are screens, not acceptance or a competitor ranking. Independent official/512 confirmation, complete family screens, all five large batches and risk checks remain required; source stays frozen/uncommitted.

Independent30-pair smaller-input confirmation passes all66 admissions and stable fingerprints: officialJS29.169 ->28.265ms,26/30 faster, saving CI[+0.568,+1.000]ms; 512 families147.003 ->132.509ms,27/30,[+6.169,+19.478]ms. About3.1%/9.9% lower medians in this exact-parent experiment, not a new compiler rank. All observations retained, including slower pairs. First large batch is running; complete20-family screen, separate class/interface risk checks and all five large batches plus carried/dynamic risk checks remain mandatory. No source commit or acceptance yet.

Full20-family screen and declaration risks finished. deep_types is selected by the predeclared threshold:25.375 ->27.308ms (~7.6% slower),4/10, saving CI[-5.151,+2.325]ms (inconclusive). It remains in the raw screen and is included in independent30-pair rechecks with carried predicates/reexports. The separate class/interface checks are inconclusive, not equivalence. First two large confirmation batches are13.425 ->9.610s (6/6, CI[+3.238,+4.356]s) and12.968 ->9.439s (6/6,[+2.538,+4.582]s); all variability remains. Large batches2/3 and the three-family recheck are now running/queued; final large batch4 is still required and will follow. No source changes, removed warnings, acceptance or source push.

Independent risk checks complete: deep_types24.603 ->24.799ms,15/30, saving CI[-3.308,+1.512]ms (slower median, inconclusive); predicates350.175 ->372.888ms,12/30,[-117.987,+159.554]ms (slower median, substantial variation, inconclusive); reexports17.467 ->16.048ms,22/30,[-2.992,+3.013]ms (inconclusive). Original deep_types warning and all raw pairs remain, not called equivalent or regression-free. Large batches2/3 both favor candidate6/6:16.974 ->12.130s CI[+1.852,+11.925]s and17.137 ->11.199s[+3.789,+7.860]s; all variable observations retained. Final large batch4 is now running under unchanged limits. Audit currently466 rounds/594 admissions, final target472 rounds/660 admissions includes the extra flagged-family recheck. No acceptance/source push yet.

Landed nearest-prior syntax candidates in af5fb32 and complete docs/README/evidence in b9e6d25. Both are sole Chris author/committer conventional commits, no coauthors, pushed directly to origin/main. Full audit retains472 paired rounds/944 timing samples,660 admissions and1585 archive checksums. Independent official/512/8192 medians improve3.1%/9.9%/29.8%; all30 large pairs favor candidate, pooled saving CI[+3.908,+5.881]s and five-batch[+3.131,+6.658]s positive. The7.6%-slower deep_types screen and its extra30-pair recheck remain; deep/predicate adverse independent medians and all inconclusive class/interface/reexport results are explicit, not equivalence or universally regression-free. First-draft checker4542/Program247/driver200/CLI74/entrypoint35/harness136 all pass; full Zod195TS+3HM9002 unchanged, not production admission. NodeId-only facts reset on source changes/rebuild on HIR growth, exact tie/scope/section rules and all-construction-OOM/scalar fallback verified; mutable TypeIds/outer JSDoc fallback untouched. Report: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#nearest-prior-value-declaration-index-2026-10-09-utc . Latest whole-matrix rank is still1/20 and ahead of nativeTS7 11/20, predating this A/B; other remaining profile hotspots and the full #416 objective remain open.

Beginning the fresh full five-compiler matrix and new diagnostic-only profile of landed source af5fb32 (current docs b9e6d25), native69836df6. Clean worktree and exact accepted source/binary pins verified. Matrix keeps all145 schema4 admissions,30 rotating fresh processes for all20 families and all3,000 samples; no Home-only A/B samples mixed in. Diagnostic image is queued separately under the same shared600s/3840MB guard, requires instruction byte identity/all62 official-and-large rejection admissions and separate early/middle1s samples, and is never timed. Current full ranking remains1/20 and ahead of nativeTS7 11/20 until the new matrix is audited; all-benchmark leadership/C++/other-platform/Zod goals remain incomplete.

Refreshed landed nearest-prior matrix and diagnostic profile published in5ea87f0f81ce20d18782abcad2d0412797b9d3a0. Native69836df6 and sourceb9e6d2591 remain unchanged; diagnostic7a620df5 has exact original/renamed instruction-section identity (SHA1b07da70), all62 admissions and silent/stable8192 positive input. Separate1s early/middle samples record824/817 main-thread observations: virtualSectionStartForNode172/169, require-binding receiver100/80, previous-JSDoc15/79, function-signature scan44/37. These are exploratory leaf counts, NOT whole-run percentages or diagnostic-image timings. Source/disassembly retained in27-checksum profile archive: virtual-section source-prefix line scan (symbol0x100199aec; offsets40/52/1424/1432) and root-statement require-binding scans remain. Next step: complete source-owned section facts preserving clamped positions, marker-line semantics, exact spellings, source replacement/OOM behavior and full scalar parity; no benchmark/library-source exceptions or numeric semantic limits. Matrix still1/20 and nativeGo11/20, no overall completion claim. Profile report: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#landed-nearest-prior-profile-2026-10-09-utc .

Starting the next scoped candidate from 5ea87f0 / native 69836df6bd81695ed3f78c2b147a7ec8ac908e5e406da5523b9da7ed3c4a895a. The landed profile maps repeated virtual-section prefix scans. Candidate: complete source-owned ordered raw marker-line starts, binary search with exact containing-line/newline/clamped-position semantics, source replacement ownership, and complete scalar OOM fallback. No directive reinterpretation, filename shortcuts, semantic caps, or diagnostic changes. Full tests/admission, pinned Zod, exact-parent official/scaling confirmations, full-family and adverse-risk rechecks precede acceptance; all samples/failures will be retained.

Virtual-section candidate checkpoint: all 4,546 complete checker tests pass. Byte/EOF/newline/raw-marker scalar parity, clamped HIR starts, source replacement/null, every allocation failure with forced list growth, partial-fact cleanup, and full checked-JS diagnostics are covered. First complete run was 4,545 pass / 1 failure: a test asserted the failing allocator counted the rejected allocation; corrected only that assertion to successful allocations. Original draft and failure log preserved. Remaining full gates and the predeclared exact-parent timing/risk protocol are pending; candidate is not accepted or pushed.

Virtual-section candidate complete semantic gates pass: checker4546 / Program247 / driver200 / CLI74 / entrypoint35 / harness136; standalone ReleaseFast native build succeeds within the original 3840MB/600s guard. Pinned unmodified Zod4.5.2 retains the complete identical 198-diagnostic multiset (195 TS +3 HM9002; normalized SHA256 6bf2feb527d91c651631642e523da6d28cd2b0c7ae4898cac0888aa7b420874f), not production admission. Exact-parent A/B protocol now running with all samples and negative-admission outcomes retained. No acceptance or competitor-rank claim yet.

First virtual-section screen completed with all66 admissions and unchanged before/after fingerprints. 512 families: parent118.741ms ->candidate112.437ms,6/6 pairs; paired mean-saving95% CI +3.258 to +9.855ms. 8192:8.529s ->7.254s,6/6; CI +1.200 to +1.320s. Exploratory only: independent official/512 confirmations, all20 family screen, all5 predeclared large batches and carried/declaration risks remain pending. Every raw sample retained; no competitor ranking or acceptance claim.

Independent official confirmation is inconclusive:27.475ms ->27.284ms,23/30 faster but paired mean-saving CI -0.791 to +0.576ms. Independent512 confirmation supports a scale gain:119.253ms ->114.483ms,25/30; CI +2.752 to +5.372ms. Complete20-family screen selected deep_types and recursive_generics due negative intervals; both join carried type_predicates_large/reexport_graph in fresh30-pair risk checks. First of all5 predeclared large batches passes66 admissions:8.525s ->7.185s,6/6, CI +1.283 to +1.398s. No samples excluded and no acceptance yet.

Landed 30330c1 (Chris-only conventional main commit, no coauthors): complete raw source-owned virtual-section prefix facts, exact marker-line/newline/EOF semantics and complete scalar OOM fallback. Documentation/evidence is 71fa7db: all502 paired rounds/660 admissions retained,1617 primary checksums plus5 lint-context checksums verified. Independent512 median119.253->114.483ms (4.0% lower), saving CI +2.752 to +5.372ms. All5 large batches/30pairs:8571.067->7274.721ms (15.1% lower),30/30 faster, pooled saving CI +1275.955 to +1657.831ms and five-batch CI +1225.320 to +1708.466ms. Official CI spans zero. IMPORTANT: deep_types/recursive_generics negative screen intervals remain; independent rechecks span zero, but recursive_generics is1.5% slower,9/30 wins, CI -0.447 to +0.007ms: unresolved warning, not equivalence/regression-free. Complete4546/247/200/74/35 module +136 harness gates pass; pinned Zod unchanged195TS+3HM9002, not admitted. Repo-wide Pickier fails on unchanged11461 errors/10613 warnings; full log retained, no autofix. README table/full report: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#virtual-section-prefix-facts-2026-10-09-utc . Issue remains OPEN for remaining measured work and adverse-risk follow-up; this A/B does not update the existing1/20-leading matrix.

Fresh landed-source matrix and profile published in 484ad69. Result 20261009T045701Z measures source 71fa7db / compiler 30330c1 / native SHA256 7ab04d806bc0557ff385a91cc993b95c3a727b7e82353ff3ea3e1b59301f822a: all 145 protocol-4 admissions, 600 original rotating rounds and 3,000 samples verify, with unchanged competitor pins/provenance. Home still leads 1/20 and beats native TS7 11/20; this is not an A/B or a causal cross-run regression verdict. Current-source diagnostic profile passes all 62 admissions, preserves original/renamed instruction bytes, and is never timed. Early/later main-thread samples 810/811 map require-binding root scans 130/108 leaves and previous-JSDoc fallback 8/77; offsets/source/full assembly are retained. Samples start at 0.1/3.0 seconds, not whole-run percentages or directly comparable windows with the old 5s profile. Matrix/context/profile archives verify 603/17/29 checksums. Next scoped target: complete immutable require-binding query facts with exact root/export/virtual-section/import semantics, ownership and scalar OOM fallback, all semantic/performance controls. The recursive-generics adverse warning remains unresolved and must stay in subsequent risk controls. Full report: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#landed-virtual-section-profile-2026-10-09-utc

Starting the next measured candidate from 484ad69 / native 7ab04d806bc0557ff385a91cc993b95c3a727b7e82353ff3ea3e1b59301f822a. Profile maps repeated root-statement scans in memberAccessReceiverIsRequireAssignmentBinding. Existing local-value root groups already contain require candidates, but use a conservative raw-source prefilter and lack HIR-growth invalidation. Candidate: complete root syntax candidates (same requireBindingForStatement predicate), revision-coherent index reuse, and same-section presence query without the separate specifier lookup position/fallback filters. Full original scanner remains OOM fallback. No name shortcuts, semantic caps, alignment hacks, type approximation or diagnostic omissions. The previous recursive-generics warning stays in subsequent independent risk controls. Resource checkpoint: inspected/closed task-owned unit-cache outputs only removed; manifest retained, 764MB rebuildable, no native/profile objects, source or evidence removed.

Require-presence candidate is implemented locally against 484ad69 and is NOT accepted/committed/pushed. Reuses complete root require groups, removes the raw-source prefilter only from candidate construction (specifier query keeps its original gate), preserves same-section presence with no earlier-position/cross-section-fallback filters, and adds coherent HIR-growth plus source/OOM revision handling. Six controls cover complete node/scalar parity, position/section/root cases, source null/replacement, actual root HIR growth, every forced-growth allocation failure/cleanup/retry, and full JS/typed diagnostic equality against scalar queries. Pinned Zig formatting and diff checks pass. Compilation did NOT execute: guard exit123 refused admission at 702MB free vs unchanged1024MB floor. 58 closed task-owned unit-cache outputs (763584512 bytes) were safely removed after inventory/hash/open-file checks, recoverable by rebuilding; no source, native/profile objects, benchmark binaries or evidence removed. Disk remains below floor despite cleanup. Current candidate/drafts/protocol/refusal/cleanup manifests are preserved in /Users/chris/.cache/home-ts-require-presence.VtpM5Y. No benchmark gain is claimed. Next step is full validation when several GB of disk headroom are available; no guard relaxation or unrelated cleanup. Previous deep_types/recursive_generics/predicates/reexports risks are all mandatory independent rechecks.

Validation remains blocked by disk headroom on the third consecutive goal-turn audit: 283MB free versus unchanged1024MB admission floor. No owned compiler/test/benchmark process is live; there is no job to resume or restart. Task-owned recoverable unit-cache cleanup is already complete, and remaining protected native/profile objects/evidence cannot provide the missing headroom. Candidate remains preserved, uncompiled/unadmitted/untimed/unaccepted/unpushed; current main remains484ad6904. The goal is being marked blocked pending several GB of free disk space, not completed. No guard relaxation, unrelated deletions, benchmark exclusions or unvalidated main push.

Resumed with restored disk headroom (~18GB), guards unchanged. Origin main advanced through three #846 correctness commits to f60a1fa. The original unvalidated require-presence patch is preserved byte-exact (SHA25695d72dc1df93cb9533790e24808dcff53e3f0fc47562e237d51fa774b102632b) in a named stash and the existing refusal/draft/protocol cache. Building a clean exact f60a1fa parent before restoring/reconciling that candidate, so A/B evidence includes all current upstream correctness work rather than attributing unrelated source changes to this optimization. No old timings will be reused as the refreshed-parent control. Full module, strict admission, pinned-Zod and carried-risk gates remain required.

Refreshed exact-parent build passes on clean f60a1fa; native SHA2560963397d935ae4ae70f9262c2e59c671a3e812c6374e25816a12846a753f997d. The preserved shared-require candidate reconciles without conflict. First complete checker run passes all 4,553 tests, including six new source/root/ordering/HIR-growth/OOM/full-diagnostic controls and the upstream #846 test. Remaining complete module/native-build/Zod/harness gates are running before timing. Both old blocked attempts and refreshed source/provenance are retained; no performance or compatibility claim yet.

Resumed checkpoint: exact refreshed f60a1fa parent was built cleanly (native0963397d935ae4ae70f9262c2e59c671a3e812c6374e25816a12846a753f997d), and the preserved candidate applied without conflict. Both complete checker runs pass all4,553 tests. The profiled root-loop source method remains byte-identical on this refreshed parent (source-method SHA2563c47f5a8b136f854448dc367766fe2b250e23f5142b6771c9b6c250bc55eb919); no old machine-address/percentage equivalence is claimed. First full Program attempt reached unchanged600s wall-clock limit before results; exit124 and complete command/log retained. Its exact full-gate retry is currently live, waiting through the shared machine lock while independent builds run in other projects. Those jobs are untouched. Remaining Program/driver/CLI/entrypoint/native/Zod/harness and all timing/admission phases remain pending; no candidate acceptance, benchmark gain or unvalidated main push.

The queued retry acquired the shared lock and completed the full Program gate: all 250 tests pass, no named filters. The prior 600s timeout remains archived. Source audit reconciles counts after upstream #846 renamed/replaced tests: Program net+3 (247->250), driver net0, entrypoint net+1; this does not disable or filter anything. Both complete checker runs remain4553/4553. Remaining gates are progressing in the same serialized workflow, guards unchanged; no timings or acceptance claim yet.

Additional full-gate progress on the preserved refreshed-parent candidate: the queued Program retry completed all 250 tests successfully, and the full driver suite now passes all 200 tests. Checker remains 4,553/4,553 in both full runs. The original Program timeout is retained; no guards changed and no tests filtered. The same live workflow is proceeding through CLI, entrypoint, standalone native build, pinned-Zod comparison and harness before the declared admission/timing phases. No candidate acceptance or performance claim yet.

Full module gates are now green on the refreshed-parent candidate: checker4553, Program250, driver200, CLI74 and entrypoint36. The standalone ReleaseFast native build is live and has acquired the shared lock; build/Zod/harness and benchmark admission/timing remain pending. Original Program timeout and all waits remain retained, guards unchanged, no test filters or unvalidated push. Read-only C++ upstream refresh under #838 found both previously audited source heads unchanged; no pins, admissions or timings were substituted.

All semantic/build gates pass on the refreshed exact f60a1fa parent/candidate: checker4553, Program250, driver200, CLI74, entrypoint36, standalone ReleaseFast build and136 harness tests. Pinned unmodified Zod4.5.2 retains the complete identical198-diagnostic multiset (195TS+3HM9002; SHA2566bf2feb527d91c651631642e523da6d28cd2b0c7ae4898cac0888aa7b420874f), not production admission. Original Program timeout and lock waits are retained. Measurement entry point now independently requires Zod/harness/module/build results before timing. The predeclared exact-parent protocol is starting: all66 admissions per phase, official/scaling confirmations, all20 family screen, all5 six-pair large batches, and carried deep_types/recursive_generics/predicates/reexports plus declaration risks. No timing result or acceptance claim yet.

All full semantic/native/Zod/harness gates are green. Frozen A/B candidate native SHA256258e2d79a04ed0779be91552ddb7b7745cc6e73d3237f13a3e9a4e0ec770061f vs exact f60a1fa parent0963397d935ae4ae70f9262c2e59c671a3e812c6374e25816a12846a753f997d; patch SHA256f30ea0345f78ee928018543e3ee73fd7435473c0cdaabd7e010b65e208b03c39. First complete screen independently verifies32 paired rounds and66 admissions, all fingerprints unchanged. Official median28.215->27.629ms,15/20 faster, saving CI +0.061 to +0.675ms. 512 families119.435->117.445ms,4/6, CI -6.397 to +16.055ms; 8192 families8.935->7.775s,5/6, CI -0.146 to +2.224s. Scale observations are variable and retained; this screen is exploratory, not acceptance or equivalence. Independent confirmation is live/queued under the shared lock; all20 family screen, all5 declared large batches and carried/declaration risks remain pending. No new main push, competitor ranking or claimed universal gain.

Partial protocol checkpoint independently verifies298 paired rounds and264 admissions, with all source/binary/input/harness fingerprints unchanged. Independent512-family confirmation:115.434->110.414ms (4.3% lower median),25/30 faster, paired mean-saving CI +0.588 to +5.410ms. Official confirmation26.802->26.968ms,18/30, CI -0.094 to +0.498ms: inconclusive. Full20-family screen retains recursive-generics/re-export warnings; all four carried risks remain required regardless of current screen. First of all5 declared large batches:8.945->7.190s,6/6 faster, but mean-saving CI -0.028 to +4.807s due retained variability, not a confirmed standalone batch claim. Remaining large batches and independent risk phases are live/pending; no acceptance or source push. Broader #838 discovery records another C++/LLVM candidate without runtime admission, and an attempted existing Linux workflow was refused by GitHub because repository Actions are disabled. No setting/pin/policy changed; asked user for Actions or runner access while local work continues.

Resumed measurement audit: old tool handle94278 and process60129 are absent; interrupted batch2 left only an empty directory, no persisted admission/timing/exit records. Cause/exit is unknown, not reconstructed. Two large batches and all earlier phases remain immutable; independent verifier confirms304 completed rounds/330 admissions. Empty attempt preserved by recoverable rename, and exact source/native/input/harness checks pass. Recovery executes only missing declared batches2/3/4 and risk phases under unchanged guard; no completed observations rerun/replaced or outliers discarded. Origin main has advanced to3f967d367 with further #846 correctness work; frozen A/B remains against f60a1fa, and any accepted integration must reconcile/revalidate before a non-forced Chris-only main push.

Complete frozen experiment verifies all502 paired rounds and660 admissions. Candidate REJECTED: independent class_hierarchy30-pair confirmation22.280->22.405ms,9/30 wins, saving CI -0.429 to -0.041ms. This confirmed regression overrides positive512-family and pooled/five-batch8192 gains (all30 large pairs faster). No repeat-until-pass, layout workaround, sample/batch exclusion or semantic shortcut. Entire candidate source/tests were restored byte-exact to tested f60a1fa; installed native restored exactly to0963397d. Worktree then advanced cleanly to latest origin/main3f967d367, preserving all newer #846 work; clean current-main native rebuild is underway and is not mislabeled as the older tested parent. Candidate was never committed/pushed to production. Full rejection tables/README and checksum archives, including timeouts/refusals/interruption/recovery and broader platform audit, will be published next. Issue remains open for a different logical design and remaining performance work.

Rejection documentation/evidence published on current origin/main in Chris-only conventional commit68ea38d7a4654b45e91e5386f3d9ca8e8562e5f8, no coauthors and no compiler changes. Complete502 paired rounds/660 admissions and all1004 times retained; primary archive1665 checksums verify (SHA256dbae49f0a1008d64625ada5a2846211e53d13d63f7521a066d15d25b69cf7149), plus5-checksum lint context. Shared root reuse candidate rejected for independent class-hierarchy CI -0.429 to -0.041ms,9/30 wins, despite positive512/8192 gains (4.3%/14.4% lower medians; all30 large pairs faster). Source restored exactly, current-main restoration checkpoint3f967d367 rebuilt to native34118002; subsequent runtime commits preserved by clean documentation rebase. Candidate never committed/pushed to production. Full semantic gates and unchanged198 Zod diagnostics remain documented; Zod not admitted. Existing repository lint11461 errors/10613 warnings retained, no autofix. README summary and full report: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#shared-require-binding-presence-rejected-2026-10-09-utc . Issue stays OPEN for a different logical query-fact design and remaining measured work; no alignment tweak, retry-until-pass, omitted body/diagnostic/sample or universal claim.

Starting a distinct presence-specific design from clean main68ea38d7a. Exact clean control build passes; native34118002e25d69a8351a60ee493c504ca734b477ed31efc79fb03ee63a9ad331. The rejected broad-index coupling remains absent. New scope: complete root/name/virtual-section presence facts only, natural absence, source/HIR invalidation and complete scalar OOM fallback, with general local/specifier/ordinal indexes unchanged. Full semantic/admission/performance and carried/class risks remain required; no name shortcuts, semantic caps, layout tweaks, omitted work or repeat-until-pass. Earlier gh issue helper returned HTTP401; authenticated REST account/issue reads now succeed on the existing chrisbbreuer account, and this standard REST update changes no credentials or permissions.

Presence-specific candidate checkpoint against clean current-main68ea38d7a/native34118002: first full checker run passes all4554 tests. Seven controls cover exact node/scalar/root/section/order behavior, complete absence/source replacement/null/detached uses, multiple independent HIR roots, actual root growth, every forced-growth allocation failure/ownership/retry, and complete JS/typed diagnostic parity. The general local declaration/ordinal/specifier indexes are untouched. Remaining complete module/native-build/Zod/harness gates are running before any timing. Distinct immutable root/name/section facts replace broad cache coupling; no scope/file shortcuts, semantic caps, alignment changes or omitted diagnostics/bodies. Prior class-hierarchy rejection remains mandatory risk evidence, not retried until pass.

Presence-only candidate full-gate progress: both checker runs pass4554 tests, and the complete Program suite now passes251 tests. Exact control remains clean main68ea38d7a/native34118002; frozen first patch23c7c20dd5f2b3a871240b32fb8018fed736150a45eac10516e41a1896aeacbd. General local/ordinal/specifier indexes are unchanged; only complete immutable presence facts are added. The same live workflow proceeds through driver/CLI/entrypoint/native/Zod/harness before timing, with unchanged guards and all class/four carried risks mandatory. No acceptance, benchmark claim, filtered tests or unvalidated main push.

Presence-only candidate gate progress: full checker4554 (both runs), Program251 and driver200 all pass. The same workflow is live on subsequent CLI/entrypoint/native/Zod/harness gates, with no timing or acceptance claim. Root/name/section facts remain the only production addition; general local/ordinal/specifier indexing is untouched. Prior class-hierarchy regression and four carried risks remain mandatory independent checks before any accepted Chris-only main commit.

Presence-only design: the complete CLI suite passes all76 tests, alongside checker4554 (both full runs), Program251 and driver200. Same live workflow proceeds to entrypoint/native/Zod/harness, without changed guards, filtering or timing/acceptance claims. Candidate remains uncommitted/unpushed. Full class-hierarchy and four carried adverse-risk checks stay mandatory in the declared exact-parent protocol.

Presence-only full module gates are green: checker4554 (both runs), Program251, driver200, CLI76 and entrypoint38. The same bounded workflow is now on standalone native build, then unchanged pinned-Zod diagnostics and136 harness tests before timing. Candidate remains uncommitted/unpushed. No guard relaxation, source/payload/name shortcuts, semantic caps, filtered diagnostics/bodies or acceptance claim; class-hierarchy and carried risks remain mandatory independent confirmations.

Presence-specific candidate passes every full semantic/build gate: checker4554 (first/cached), Program251, driver200, CLI76, entrypoint38, standalone ReleaseFast build and136 harness tests. Unmodified pinned Zod4.5.2 preserves the complete identical198 primary-message/position/code diagnostic multiset (195TS+3HM9002, SHA2566bf2feb527d91c651631642e523da6d28cd2b0c7ae4898cac0888aa7b420874f), not production admission. Frozen exact-parent measurement protocol now starts only after all gates, with all66 admissions per phase, five declared large batches, full20-family screen and mandatory class/four carried adverse risks. Distinct presence-only facts do not alter general declaration/ordinal/specifier indexes. No performance or acceptance claim yet; no unvalidated main push.

Presence-specific protocol checkpoint: every semantic/native/Zod/harness gate passes; frozen candidate native de7d12e8370340759031a22b35a68b4f294fe2dc308c2842dbc4a5dc2873219b vs clean main68ea38d7a control34118002e25d69a8351a60ee493c504ca734b477ed31efc79fb03ee63a9ad331, patch23c7c20dd5f2b3a871240b32fb8018fed736150a45eac10516e41a1896aeacbd. Independent verifier now confirms92 paired rounds/132 admissions with unchanged fingerprints. Screen scales were highly variable and retained, intervals span zero. Independent512 confirmation121.349->115.059ms (5.2% lower median),26/30 faster, mean-saving CI +4.278 to +9.173ms. Official confirmation29.938->29.991ms,18/30, CI -4.646 to +2.181ms is inconclusive. Full20-family screen, all5 large batches and mandatory class/four carried risk checks remain live/pending. No acceptance, discarded observations, relaxed guard, compiler pin change or unvalidated push.

Presence-specific partial checkpoint: all completed phases retain original arrays/fingerprints. Two declared large batches are complete. Batch1 median8.342->6.979s,6/6 faster, saving CI +0.313 to +3.425s. Batch2 median7.909->6.557s,5/6 faster, but CI -1.316 to +2.484s; the slow candidate observation12.371s is retained and the batch is not credited as confirmed gain. All3 remaining large batches, five selected adverse rechecks and class/interface risks remain pending/live. Independent512 confirmation remains supported (5.2% lower median); official checking is inconclusive. No candidate acceptance, outlier filtering, pin changes, guard relaxation or unvalidated main push.

Complete presence-specific experiment verifies532 paired rounds/660 admissions. Candidate REJECTED: independent interface_composition30-pair confirmation44.753->45.817ms,10/30 wins, saving CI -6.423 to -1.076ms. Positive512/large gains do not override this regression; all observations/batches/adverse medians retained. No repeat-until-pass, alignment tweak, semantic shortcut or discarded data. Entire candidate source/tests restored exactly to parent68ea38d7a, native restored byte-exact34118002, then cleanly advanced to latest origin/main37b5225c2 preserving runtime changes; matching clean native build now verifies restoration. Candidate never committed/pushed. Rejection README/full tables/raw checksum evidence will be published next; issue remains open for remaining performance work and a different evidence-led approach.

Completed presence-only rejection report/evidence published on current origin/main in Chris-only conventional commit624916ec4210bb13ebbc8e9368f9f28d9152cb89, no coauthors. Entire candidate/controls removed, native restored exactly; current-main checkpoint37b5225c2 clean build matches34118002. All532 paired rounds/1064 times/660 admissions retained, primary archive1656 checksums verify (SHA2565dc948934a065562ddcf9c8528f418657e4f6f946e4127fee0ed63faf5b5da01), plus5-checksum lint context. Candidate rejected for independent interface-composition CI -6.423 to -1.076ms,10/30 wins, despite supported512/large scale gains5.2%/17.5%. All adverse/variable pairs, five large batches, added checked-JS recheck and nearly negative re-export warning retained; no equivalence/global gain implied. Full semantic gates/Zod diagnostics unchanged, Zod not admitted. Existing repo lint11461 errors/10613 warnings logged without autofix. README4-row summary links complete report: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#presence-specific-require-facts-rejected-2026-10-09-utc . Issue stays OPEN; no retry-until-pass/layout workaround or unvalidated runtime push.

After both completed require-presence candidates were rejected and restored, refreshing clean current-main624916ec4 / compiler-input checkpoint37b5225c2 / native34118002. New full five-compiler matrix uses unchanged pins, full protocol4 negative controls,30 fresh rounds after3 warmups, all samples retained; it is not a patch A/B. Separate diagnostic-only current-source profile requires62 native/diagnostic admissions and identical original/renamed instruction sections; no diagnostic timing. Source/symbol/assembly mapping includes previous-JSDoc and value-declaration resolution alongside remaining require/section paths to choose actual next repeated work, not another cache/layout retest. Both jobs serialize via unchanged machine-lock/time/memory/disk guard. No result or new rank claimed yet.

Restored current-main matrix/profile published in Chris-only main commitabe538a8638ed4542abfa7b256f3a2711a1f59c6, no coauthors or compiler changes. Result20261009T194221Z measures source624916ec4/compiler37b5225c2/native34118002: all145 admissions,600 original rotating rounds/3000 samples verify with unchanged pins/fingerprints. Home leads1/20 and beats nativeTS7 on13/20 unrounded medians; no causal cross-run gain/equivalence claim. Diagnostic-only profile passes62 admissions and identical original/renamed instruction sections, never timed. Early/later822/814 main observations map previous-JSDoc11/66 leaves and value-resolution14/106; exact source/symbol/assembly included. Previous-JSDoc offsets4344/4356/4432 map the ancestor original-order prefix span/kind/name loop after nearest-prior lowering fails. Next investigate complete source-order candidates preserving first-match/null-return/prefix-break/recovery/OOM semantics, with no invented virtual-section filter or mutable-type caching. Both rejected require caches remain absent, not retested/tuned for layout. Matrix/context/profile/publication archives verify603/17/33/7 checksums. README/full report: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#restored-current-main-profile-2026-10-09-utc

Starting the mapped previous-JSDoc ancestor-prefix fallback from clean current mainabe538a86/native34118002. Original loop searches ancestor block/namespace statement lists in original order, stops at the first statement.start>=use.start (even unrelated/recovery statements), and returns first matching simple var/let/const leading-JSDoc result including null. Nearest-prior fast path/live lowering remain unchanged and no absent virtual-section filter is added. Candidate syntax facts: first declaration per name plus maximum statement start through its original prefix, preserving exact stopping/order semantics without sorting statements or caching mutable TypeIds. Complete source/HIR ownership/invalidation and full scalar OOM fallback, exhaustive position/order/null/scoping/full-diagnostic controls, all semantic/admission/fair timing and carried/class/interface risks required. This is different actual repeated work, not either rejected require-cache variant or an alignment tweak. Clean exact parent build precedes source edits; no result or push claimed.

JSDoc ancestor-prefix candidate update: full unfiltered checker suite passed twice (4,553 tests); Program/driver/CLI/entrypoint/native-build/Zod/harness gates are running before timing. Exact parent abe538a, native control SHA256 34118002e25d69a8351a60ee493c504ca734b477ed31efc79fb03ee63a9ad331. Candidate patch SHA256 0719bd67d2b1310753cc8097b099ccef43b18bd326e07ebbff5122443c7cf951. Frozen protocol retains every original-order statement prefix maximum, scalar OOM fallback, source/HIR invalidation, live JSDoc lowering, and nearest-prior lookup. Timing remains the full predeclared A/B protocol: 20-family screen, official and 512-family independent confirmations, five 8192-family batches, carried deep/recursive/predicate/reexport/official-JS risks, and 30-pair class/interface controls. Complete original negative diagnostics required before every phase; no sample exclusions or repeat-to-pass. Either independent adverse interval rejects the candidate. No production commit or speed claim yet.

Validation progress for the distinct JSDoc ancestor-prefix candidate: checker4553 (full initial and cached confirmation), Program251, and driver200 all pass without named filters or skips. CLI/entrypoint/native-build/complete pinned-Zod comparison/harness gates remain live in the same serialized runner. Frozen protocol SHA256 d0cc1b2f8998136289cde70a86f979a37dfabe706ccb67eece5b4bc873bc6664; source parent abe538a remains origin/main. No timing, acceptance, candidate production commit, or benchmark leadership claim yet. Every independent adverse-risk result will be retained.

All pre-timing gates passed for JSDoc ancestor-prefix facts: checker4553 (twice), Program251, driver200, CLI76, entrypoint38, stripped ReleaseFast native build, and136 benchmark-harness tests. Complete pinned Zod 4.5.2 primary diagnostics remain identical:195 TS plus3 HM9002,198 total; Zod remains unadmitted. Native candidate SHA256 68c6ed517f9125ae1b570ff2e585c5b368745389af828258f6621fe5e6e0311c; parent34118002e25d69a8351a60ee493c504ca734b477ed31efc79fb03ee63a9ad331 on exact source abe538a. Frozen patch0719bd67d2b1310753cc8097b099ccef43b18bd326e07ebbff5122443c7cf951. Full predeclared native A/B has started only after all gates passed. Complete admission precedes every phase; all samples and independent adverse-risk controls retained. No acceptance, candidate source commit, or speed claim until the entire protocol is verified.

Initial frozen JSDoc ancestor-prefix screen completed with stable fingerprints and66/66 complete admissions. All original pairs retained: official checked JS parent25.945500ms/candidate25.879688ms (14/20 faster; paired mean-saving95% interval -0.041630..+0.256559ms),512families113.006771/112.363146ms (5/6; -0.054523..+2.795107ms),8192families7.758223396/7.024730646s (6/6; +0.598092184..+0.750393010s). Large median9.5% lower in this screen; official and512 intervals span zero. Independent confirmation is live. These are screening observations, not acceptance, significance across the complete protocol, or competitor/universal leadership claims. All five large confirmation batches and independent adverse-risk controls remain required.

JSDoc ancestor-prefix protocol progress: verifier confirms292 retained original rounds and198 passed admissions across completed screen, independent official/512 confirmation, and all20-family screen. Independent official median28.624500/28.647854ms,20/30 candidate faster, paired saving95% interval -0.053562..+0.350749ms;512median114.293896/111.934209ms,23/30, interval -1.460720..+2.965023ms. Both remain inconclusive by the predeclared interval criterion; all slower observations retained. Re-export family screen median13.527500/14.111855ms (~4.3% slower), interval spans zero; retain and independently recheck, not dismiss. Selected rechecks are checkjs_jsdoc, deep_types, recursive_generics, reexport_graph, type_predicates_large. Five predeclared8192-family batches are now live, then all selected rechecks and independent class/interface controls. Partial verified evidence only; no acceptance or production candidate commit.

First predeclared8192-family independent batch completed,6/6 original pairs retained: parent median7.582381438s, candidate6.914781500s (~8.8% lower), paired mean-saving95% interval +0.413052460..+0.818854193s. Partial verifier now confirms298 original rounds/264 admissions with stable native/input/runner fingerprints. Four further batches, all five carried-risk rechecks, and30-pair class/interface controls remain mandatory; next batch is live in the same measurement job. No exclusions, repeat-to-pass, alignment tweaks, candidate source commit, or acceptance claim.

Large confirmation batches2 and3 completed with every original sample retained. Batch2 parent7.909771854s/candidate7.303896501s,6/6 faster, paired saving95% +0.417625592..+0.826191866s. Batch3 parent7.674694313s/candidate6.890669021s,6/6 faster, +0.580863622..+0.888742573s. All18 pairs across first3 batches favor the candidate, but the full protocol remains incomplete: two batches, all selected carried-risk rechecks, and independent class/interface controls still required. Partial verifier confirms310 rounds/396 admissions; stable fingerprints, no dropped observations, no new source edits or production candidate commit. Prepared documentation checks will compare all detailed tables and brief README medians/links to complete verified evidence; no universal, competitor-ranking, equivalence or Zod-admission claim.

Accepted distinct JSDoc ancestor-prefix facts pushed to origin/main as Chris-only conventional commit11055ba4746b224735c8ec2cba2f5e449bfff980 (https://github.com/home-lang/home/commit/11055ba4746b224735c8ec2cba2f5e449bfff980), no coauthors. Exact-parent source/test patch matches frozen0719bd67 SHA256. Complete protocol532 rounds/660 admissions; all semantic gates pass, identical198 Zod diagnostics (still unadmitted). Large pooled median7.712273542→7.007292771s,9.1% lower,30/30 faster; paired saving95% +0.622100622..+0.730809487s and five-batch interval +0.606433279..+0.746476829s. All independent adverse controls span zero; recursive generics retains1.0%-slower median,11/30 faster, nearly negative -0.670..+0.041ms interval. No equivalence/regression-free/all-benchmark claim. Native SHA25668c6ed517f9125ae1b570ff2e585c5b368745389af828258f6621fe5e6e0311c. Documentation/brief README table and complete1659-checksum archive are locally verified and being linted for a separate publication commit. Full competitor ranking still predates this change; #416/#838/#548 remain incomplete.

Publication complete: Chris-only conventional docs commit0d2b0097abf6dbb37630505b69e8a261c7d53c9a is pushed to origin/main after source11055ba4746b224735c8ec2cba2f5e449bfff980, no coauthors. Brief README table and full verified report: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#jsdoc-ancestor-prefix-facts-2026-10-09-utc . Complete532-round/660-admission archive (1659 payload checksums): https://github.com/home-lang/home/raw/main/bench/vs_tsgo/evidence/20261009-jsdoc-ancestor-prefix-accepted-854.tar.gz , SHA256336e6b45731608f04309aa999b5ead588c847968709d7b7f81347bf7adca8ff4. Publication lint context (9 checksums): https://github.com/home-lang/home/raw/main/bench/vs_tsgo/evidence/20261009-jsdoc-ancestor-prefix-lint-context-854.tar.gz , SHA256e2ea8a453b9113dd58487eba17071c79c22d3bdb5e308d0a9890232c6bdce018. Initial documentation lint found one new malformed heading from wrapped #846; proper issue link repaired it, failed log/source snapshot retained, complete rerun returns prior11461 errors/10613 warnings. No auto-fixes or fabricated lint pass. All benchmark/test/build jobs terminal, no observations excluded. Recursive-generics adverse warning and all inconclusive results remain. Issue stays open for actual remaining mapped work; no current-source five-compiler ranking or universal/Zod/C++/platform completion claim.

Landed ancestor-prefix compiler now has its own complete competitor snapshot, published as Chris-only conventional docs commitb6608b14c631db91be6099e78c957ab67a3b5266. Result20261009T205353Z:145 admissions/600 rounds/3000 samples, all original competitor pins and protocol4 unchanged, source0d2b0097a/compiler11055ba47/native68c6ed517f9125ae1b570ff2e585c5b368745389af828258f6621fe5e6e0311c. Home1/20 leads and12/20 vsnative TS7; official checked JS still2.03x behind fastest competitor. The separate9.1% large exact-parent gain and recursive-generics nearly-negative warning remain fully documented, not superseded by cross-run medians. https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#landed-jsdoc-ancestor-prefix-matrix-2026-10-09-utc . Allraw603checksums, context28 and publication6 preserved; the two supplemental verifier orchestration failures are transparent and did not trigger a measurement rerun or exclusion. Publication context SHA256013015a0e7b0c93a762a0b48f3c37ae8bfbbc66001be03ee16a5701f9c5a0956. Next work must use fresh current-source samples/assembly for remaining loops; no revival of either rejected require-presence binary or layout tuning.

Fresh landed-source profiling audit on clean b6608b1/native68c6ed51 caught a stronger-proof issue before any admissions/sampling: local-symbol renaming preserves original/renamed object instruction bytes, but relinking through clang does not preserve the full final executable instruction section. Native9675688 text bytes SHA25677288d82ae38d7d741276d9f2734dce2f2d6916af69a11d4c14ed953e0f25e5c vsdiagnostic9674440 SHA256732298add1119f7332c21be55ad844656cdef69524cba540045f743b9063473c. The guard stopped this attempt; all logs/images retained and installed native unchanged. Earlier object-byte checks prove only their explicit pre-link scope, not final-native executable equality; their diagnostic leaf samples must not be described as exact native instruction profiling. No benchmark timing ever used those diagnostic images, and accepted exact-parent timing/semantic evidence is unchanged. Next safe method is the canonical pinned Zig ReleaseFast build with symbol stripping disabled, requiring final executable text equality before profiling; native preserved/restored, no optimizer/source/alignment changes or weakened guard.

Exact-native sampling now succeeds without a replacement image or weakened guard. Canonical unstripped build was stopped at3885MB by the unchanged3840MB ceiling (bounded exit125); native68c6ed51 restored byte-for-byte. Read-only retained-object/native mapping validates all9674440 object-text bytes except159369 explicitly declared linker relocation entries (633752 operand bytes), with zero undeclared differences; native text9675688 bytes includes appended thunk code. Cached native image SHA matches68c6ed51, objectSHA36aba5ab5e5bf1c982bac3265552fa50c992938ec7cf67bf6b2e46f685a00f92. All31 native admissions pass, actual native process exits0/silent, source b6608b1/input/binary fingerprints stable, early0.1s and middle3.0s one-second samples complete. Main observations826/823. Raw native sample collapses local funcs under main, so only explicit PC locations are mapped; ellipses/group counts retained and never invented per-function CPU counts. Namespace-import statement scan offsets188/200 appear in both windows; signature collection offsets36/56 early and36/96 middle. Native disassembly/source snippets verify those positions. No diagnostic image sampled or timed. Archive/documentation in progress; source unchanged and full benchmark standing remains1/20.

Exact-native profile and byte-scope audit now published in Chris-only conventional main commit88a8c2d39c5608be367e75f8379d572b4f571c29, no coauthors: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#exact-native-profile-and-byte-scope-audit-2026-10-09-utc . README includes a brief displayed-PC table/link, not invented CPU shares. Full73-checksum native-profile archive SHA2560f0fb7ebcbf75e736903fd8cc8952cae9414af218289fea157f4523cdd051d60; separate9-checksum publication context SHA256a1d19b5be116c2c04e53ea38ab16fe1e894fd3493fe2469bfefbc86beb6598ce. Clarified that declared relocation-site comparison excludes whole relocation-sized words, not only operand bits/opcode identity. Both failed approaches/guard125/unsupported metadata probe and raw collapsed counts/ellipses remain. Direct unchanged-native31 admissions and source/assembly positions verify. Four concurrent runtime commits preserved; clean matching standalone build reproduces native68c6ed51. Final changed-doc Pickier passes; full repo still11460 errors10613warnings, no auto-fix. Next implementation is tracked separately in #857; no new gain/universal leadership claim.

Next observed scan now tracked in #859: coherent reuse of ordered bound function declarations for collectFunctionSignaturesFromStatements (native displayed scan offsets36/56 early,36/96 middle; no function CPU shares invented). Read-only standalone2-test characterization passes representative exact-container filtering/order, but proves real HIR-growth incompleteness of stale binder list (scalar2,binder1). Existing first-declaration local-value index cannot supply all overloads. Module ownership/root/revision/completeness/augmentation boundaries and full scalar fallback must be proven; no unguarded first-declaration substitution or independent cache/layout retune. No compiler source changes or performance result yet. Carry #857 destructuring/control-flow risks and every earlier adverse control. #857 rejection remains fully documented/pushed, original native68c6ed51 current main intact; universal goal incomplete.

Continuing from clean adopted source d3cfdac, compiler0b8d1f049/native ff78eeff3f36f622db610d23881be7165dce31b424481318e82360f1da5935f8. Complete five-pin protocol5 matrix20261010T081352Z now identifies checked JS as the largest remaining descriptive fastest-competitor gap: Home27.666916ms vs Bun14.278250ms (1.937697x), nativeTS7 23.608562ms. All20rows/150admissions/600rounds/3000samples retained. Plan: fresh read-only exact cached link/object mapping and one actual unchanged native checked-JS profile, with complete reference/native positive and first/middle/last negative admissions before sampling. No old PCs, diagnostic relink, benchmark-specific shortcuts, omitted bodies/diagnostics, work caps, filtered samples or repeat-to-sample. Existing rejected presence/name/group candidates will not be revived. Scope remains parent416 and all original compatibility/platform requirements.

Fresh current-native profile published in 91c8756 as Chris/no coauthors on origin/main. Native ff78eef, source d3cfdac, exact cached object048ca33c; independent map verifies5,891 ranges and zero changed bytes outside633,796 declared whole relocation-word bytes (no opcode equivalence within masks claimed). One unchanged native8192-family checked-JS process, fixed0.1/3.0s one-second windows, compiler0silent; all36 admissions pass including15 required negative errors. Explicit PCs identify memberAccessReceiverIsRequireAssignmentBinding at0x1006893ec (early244/416, middle376/476), whose exact source/assembly scans root statements for require bindings. Local JSDoc attachment/prior-declaration paths also retained. This is a new investigation lead, not CPU-share/speedup or authority to revive rejected presence/name/group guards. Full report/readme link: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#current-native-checked-javascript-profile-2026-10-10-utc . RawSHA256 2d5f88bb08f0446bac4307765f0d9cf9813983e53d1aefdb7e41be29d12673b3,47 checksums verify. Next inspect existing syntax ownership to remove exact repeated traversal without semantic caches/shortcuts.

Current-native ownership audit found HIR fields can mutate without node-count changes, so no incomplete root-query cache is being added. Distinct current-profile candidate now under full checker gate: leadingJsDocBodyWithStart combines its backward line-boundary traversal and forward // search into one stateless backward traversal, retaining the leftmost // exactly. No cache/allocation/source-word/name shortcut or changed attachment rule. Independent original-algorithm oracle covers every cursor over7,776 five-byte trivia strings plus malformed/export/multiple-comment/CRLF/virtual cases. Frozen pre-timing acceptance on parent91c8756ac/nativeff78: all20rows10pairs, separate official checked-JS30pairs, ten fixed independent risks30each plus exactlyone confirmation of other adverse screens, seed854/20000pairedbootstrap. Positive independent primary required, any wholly-negative independent risk rejects. Complete4559/255/200/76/38/147 suites,180imageadmissions and unchangedZod required before timing. No timing/sourcepush/gain claim yet.

Single-pass JSDoc line candidate full checker passes4,559/4,559 on first actual run, including exhaustive short-trivia/cursor oracle and malformed/export/CRLF/leftmost-comment controls. Guard exit0 peak3,304MB; harness147passes. Independent source audit confirms only production delta is the single backward line traversal; existing source facts/export/adjacent-trivia/comment-body logic unchanged. PatchSHA2560bac1b9c2b3f827256a7176a0dc908fe0919f6b77f611233bfc46837c81741f7. Nativebuild nowlive; fullProgram/driver/CLI/entrypoint original queued gates next. Frozen acceptance unchanged, no timings/candidatecommit/gain yet.

Candidate native ReleaseFast build passes first run under unchanged guard, peak3,058MB. NativeSHA2560429057e7b3c19084b4f1b26298aa604c2b2d2bce8e8658f5317b2e20ef83540; retained parentff78. Fullchecker4559andharness147alreadypass. Program fullgate nowlive, remainingdriver/CLI/entrypoint sequential; exactcandidate/parent plus unchangedfivepins admission180 queued once with retained wrapperlog/accounting. No timings/adoption yet; frozen rule unchanged.

Candidate qualification advances: Program255passes under guardpeak3226MB; complete protocol5 six-image180admissions pass with unchanged provenance, retained wrapperexit0/peak690MB. Both Home images retain241TS+2HM9002 on unchanged pinnedZod, stillunadmitted/untimed. Checker4559/harness147/nativebuild alreadypass. Driver nowlive, CLI/entrypoint next; sourcepatch unchanged and no timing started. Frozen officialcheckedJSprimary and independent risk rule unchanged.

All pre-timing gates pass for exact stateless single-line-scan candidate:4559checker/255Program/200driver/76CLI/38entrypoint/147harness and180protocol5imageadmissions. Wrapperaccounting fullyretained; entrypointpeak3140MB, unchanged600s3840MB limits. Source and parentff78/candidate042905 identitiesverified; Zod241TS+2HMunchanged, unadmitted. Starting frozencomparison once:20officialrows10pairs, independentofficialcheckedJS30pairs,10fixedriskrows30pairsplus deterministicotheradverses. Seed854/20000pairedbootstrap; positiveindependentprimaryrequired, confirmedrisklossrejects. No repetitions/filtering/alignmenttuning/sourceadoptionclaim.

Frozen single-pass line-scan candidate REJECTED. All590originalpairs complete (14phases, including deterministicone-time startup/overload confirmations),180initial+104phaseadmissions. OfficialindependentcheckedJS median28.101771→27.8168125ms,25/30faster, savingsCI[+0.039431792,+0.478554092]ms positive. But startup independent30pairs savingCI[-0.156801102,-0.003057375]ms whollynegative,11/30faster, so frozenriskrule rejects. Overload confirmationCI[-0.482327288,+0.457383249]ms spanszero. Every original/adversevalue retained; no repeats/filtering/tuning. Independent raw/statistical recomputation passes. Candidate+2tests removed via scopedinverse of exact retainedpatch, nevercommitted; freshparentnative restoration running. Fullrejectionevidence/docs/README follows.

Rejection fully published in0b133c8c0 onorigin/main asChrisauthor+committer,no coauthors; concurrentruntime docs7b91c2743preserved. Candidate nevercommitted, exactpatch+2testsremoved, freshcleanparentbuildexit0restoresff78native byte-for-byte. Complete590originalpairs/14phases/284admissions independentlyverified; officialcheckedJSprimary savingpositive but startupindependentCI[-0.156801,-0.003057]ms triggersfrozenrejection. No filtering/retiming/alignmenttuning. Full33rowreport plusbriefREADMEtable: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#jsdoc-single-line-scan-rejected-2026-10-10-utc . RawSHA256b8d5ea3fc173e1ccab6147a8435ef90442829662cb814e6178fe46ac81abb2ee,1342checksumsverify; scopedPickierpasses. Broaderinvestigation remainsopen; donotreviveortune thiscandidate. Rankingunchanged1/20leads,13/20vsnativeTS7; goalrequirementsremainincomplete.

Starting current-image profiling after bd0d82e: the complete paused matrix retains all3000 samples and identifies checked JS as1.88x slower than the fastest admitted pin. No causal inference from shared-host medians. The measured native is bb4c00d from qualified4b4060e05; compiler/build module sources are unchanged. Reconstructing a cached link/object pair with the same Zig2163 ReleaseFast/stripped/no-JSC flags into a separate prefix, preserving both installed and retained images. Mapping/profiling must stop if the reconstructed native does not match the measured SHA256 exactly. No compiler edit, relinked diagnostic image, timing rerun, new cap, skipped body or weaker negative. Plan uses existing8192-family checked-JS generator, complete TS6/TS7/Home positive+negative controls, full protocol5 native admission and one native process with fixed exploratory windows; all failures and raw provenance will be retained.

Guarded reconstruction succeeded: the fresh native SHA256 is exactly bb4c00dd2fdfba719098150082b696b4f9ba8ab65a7a47640b197540420cf771. Installed and retained binaries are unchanged. Retained matching object SHA256 is de747c0cb2eecc26006b60aa83e99b8db6aa56c17a50a2c06559223849d328da. Peak tree footprint was 2986 MB, host low-water 30%, guard exit 0. Read-only mapping and independent relocation/range verification are now starting. The fixed diagnostic protocol reuses published algorithms, the existing 8192-family generator and all 15 planted errors, with one native process and windows at 0.1 and 3.0 seconds. No compiler changes or benchmark timing samples; failures and missed windows will be retained, not repeated.

Read-only address mapping and independent verification passed: 5896 ranges, zero undeclared changed bytes, 634788 whole declared relocation-word bytes masked, 1248 native tail bytes explicitly unmapped. No opcode equivalence claimed inside masks. Matching object and link are retained outside disposable Zig cache. The unchanged 8192-family positive/negative inventories exactly match the prior published hashes; negatives still require nine TS2322, three TS2339 and three TS2345. Admission/profile session48214 is queued under the unchanged machine guard; sampling has not started. The job fingerprints reference/native/harness/input artifacts before and after, stops any failed admission, and permits only one native profiling process with fixed windows. No benchmark timing samples or compiler changes.

Published ea59fe0 to origin/main as Chris Breuer, no coauthors. The measured bb4c00d native remains unchanged. Independent mapping verifies5896 ranges, zero undeclared changes and634788 whole relocation-word bytes (no opcode/bit equivalence claim inside masks). All6 complete scaled TS6/TS7/Home admissions plus30 full native controls pass. One native process; both fixed windows succeed;801 observations each and275 explicitly mapped symbol groups. Full66-payload archive and all documented offsets/addresses verified; README links the detailed diagnostic report. Root require scanning remains an observed lead, not a speedup or CPU-share claim. Follow-up #881 requires complete owned declaration/section facts: existing earliest-variable prefix data excludes imports and duplicates and is insufficient as-is. No compiler edit or easier benchmark; latest full paused ranking remains1/20 and9/20 vs TS7.

Current-image profile led to #881 coherent-root-receiver candidate. It passed allsource/actualCLI gates but failed the predeclared official-target performance requirement; its final8192 independent phase also stopped at guard124. Exactparent/source/images and all877 complete pairs, empty partialfile,264warmups are retained in e7ed4c0; source restored, no adoption or layout/source/benchmark tuning. Large-scale positive observations remain separately labeled and do not substitute for officialgain. Root-scanning investigation remains open for an algorithmically different design.

Follow-up #882 fully rejected, restored and documented in 175c6ca: all 880 fixed pairs retained, official checked-JS interval inconclusive, independent control-flow/overloads/recursive-generics losses. Completed attachment-result memoization is not adopted. Full report and README include adverse intervals and independent scaling savings: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#owned-jsdoc-attachment-memo-rejected-2026-10-12-utc . Concurrent main recursive-interface changes are preserved; next profiling must build that current source and not relabel the previous measured compiler as current.

Resumed current-source investigation after publishing #882 rejection. Building exact main 175c6ca in a separate prefix with pinned Zig2163, stripped ReleaseFast, unchanged 600s/3840MB guard. Existing installed/retained images are hash-protected. Build context home-ts-current-source-profile.kLcSvJ; no profiling or benchmark timing yet. Initial provenance preparation stopped before compiler launch on a tracked gitlink directory; corrected by hashing regular files and retaining gitlink entries separately, failure record preserved. No source change or performance claim.

Latest-source diagnostic profile completed and pushed in d82db94. Exact native 2c6666a includes compiler36a2d95e; it is distinct from the older bb4 competitor-matrix image. All36 full admissions pass, one process/two fixed windows succeed, 293 explicit symbol groups independently resolved, 5894 object ranges verified with zero unexplained changed bytes outside relocation masks. Root require-binding statement walks and leading JSDoc work are still explicitly displayed; previous-identifier helper is not displayed. No CPU-share or speedup claim, no repeat-to-sample. All68 payload checksums and original bytes verified; docs and README table link raw evidence: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#latest-source-native-checked-javascript-profile-2026-10-12-utc . Next investigation is whether existing complete binder declaration lists can replace repeated root walks while preserving scalar answers, section ownership, duplicates and unbound/mutated-HIR fallback.

Cleanup reconciliation after temporary authentication refusal: only the completed profile task .zig-cache was deleted (100080475 bytes). All83518 protected source/raw/retained-image hashes remained unchanged; custody manifests retained. Published native/link/object evidence is preserved. Shared caches and unrelated jobs untouched. Follow-up #883 now has a local prototype and unchanged full source qualification in progress; no timing or adoption claim.