TS forward aliases lose interface element types

Forward alias lookup can freeze an unresolved placeholder inside an earlier interface element type. This blocks #863 and full Zod admission under #548.

Evidence from the exact-main candidate (source parent 65a0332, qualified patch 0e43dc4, native 754926f):

• Four paired declaration-order variants: TS6/native TS7 satisfy all 16 contracts; candidate satisfies 2/8 and fails 6/8. Positives gain false TS2339; negatives lose the required TS2322 and instead emit another TS2339.
• Moving Part after ModelDef reproduces the original Zod symptom. Moving only the schema interface does not. Forward LiteralPart also fails. Original library sources are unchanged.
• A diagnostic-only executable using production project admission, resolver, checker and rendering matches all 243 original Zod diagnostic locations/codes, admits 106/106 files, and shows part is Primitive.unknown before the original typeof guard on all six observed passes. It also matches all eight reduction diagnostic sets.
• Callback-body and local type/value-name controls all satisfy their candidate/reference contracts; these do not explain the defect.

Suspected mechanism: preRegisterTypeAlias uses a primitive unknown for nonrecursive aliases. Type-name lookup returns it before forward declaration recovery, and an earlier interface shape can retain it after the alias body is checked. Preserve declaration ownership and transparent alias identity rather than overwrite nested fields opportunistically.

Acceptance:

• All forward-order positives silently succeed and every negative retains exactly one TS2322 plus one TS2339 under the same strict reference/candidate configuration.
• Add direct checker and complete Program regressions, including forward alias-to-alias references, explicit unknown aliases, lexical/virtual-section ownership and productive/circular recursion contracts.
• No declaration-reordering workaround, added any erasure, new cutoff, skipped body, diagnostic suppression or weakened contract.
• Run unfiltered checker/Program/driver/CLI/entrypoint suites, unchanged extended/protocol 5 admission, and the full unchanged Zod graph. Classify every diagnostic delta before timing/adoption.
• Publish checksummed full raw evidence and document benchmark/qualification results, with a brief README table and link. Timing must use the exact parent and retain all samples.

Full reduction inputs/output are currently retained in home-ts-forward-part-controls.lTzq5k; original-project trace/image and copied production entrypoint in home-ts-project-surface-diagnostic.AFdbFb. A direct checker regression is running unfiltered. Evidence publication follows the terminal gate result; no implementation or speedup is claimed.

Links: https://github.com/home-lang/home/issues/863#issuecomment-6113421415 and https://github.com/home-lang/home/issues/863#issuecomment-6113347977

Published documentation-only fcbd36d to origin/main as Chris Breuer, conventional issue-linked commit with no coauthors (also refs #863/#548). README table now exposes forward-order reference 16/16 vs candidate 2/8; the full report separates the observed unknown-before-guard/type-order reproduction from the still-unverified implementation mechanism. Evidence archive bench/vs_tsgo/evidence/20261011-forward-alias-diagnostic-879.tar.gz has 158 independently verified payload checksums, SHA256 1d6f945d47b1cb4e5128516c99dd3f4124c71d13735068b773935367a62c01fa. It retains all 96 four-image records, eight matching diagnostic reductions, original 106-file/243-diagnostic trace, diagnostic native/copies/patches/restoration checks, failed attempts, and the new direct regression source with its unfiltered build log. That checker build reached 600-second guard exit 124 before any assertions ran: it is unqualified, not passing/failing test evidence; limits not raised and no identical retry. Production source remains uncommitted; installed 754926f qualified native unchanged and all temporary logging removed. Next is general declaration-owned forward-alias recovery, including alias use before declaration, true unknown aliases, lexical/virtual-section ownership and cyclic contracts, followed by full qualification. No timing/adoption; published ranking unchanged. Report: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#forward-alias-order-reproduces-lost-element-types

Implementation prototype now uses declaration-owned zero-argument references for non-generic aliases, completes each through the normal full alias checker on demand, and keeps active/checked state by HIR declaration identity. Contextual cached-type clearing invalidates corresponding alias-check entries; snapshots of already completed definitions remain immutable. Added a Program regression for alias use before declaration, forward alias/alias-to-alias array elements, explicit unknown (must retain TS2322), and a productive mutual cycle (must retain its string field and wrong-number TS2322). Existing direct checker regression remains unchanged. Source is frozen for an unfiltered Program run under the unchanged 600-second/3840-MB guard; no body/diagnostic filters, new limits, any fallback or library/benchmark predicates. This is uncommitted and unqualified; full checker/Program/CLI gates and unchanged Zod/reference/admission/performance qualification still required.

The prototype unfiltered Program run stopped at disk-guard exit123 before assertions: only511MB free at the unchanged512MB critical floor. No test success/failure is inferred. Full command/log, exact49KB source patch and guard outcome are retained in home-ts-owned-alias-prototype.wVgnmk; limits were not raised. Per user-authorized task-only cleanup, removed only .zig-cache in this chat attached ts863-qualification-docs, ts-bench-snapshot-837 and ts-839-parent worktrees after verifying the original run processes were terminal and targets were real untracked generated cache directories. Source/raw evidence/retained images and shared/global caches remain outside all deletion targets; before/after protected checksum verification is finishing. Independent disk recovery plus cleanup now leaves about6.2GB; unrelated builds remain untouched. Prototype remains uncommitted/unqualified; resumption will be a separately retained run after verified space change, not a repeat-to-pass measurement.

Task-only cache cleanup is terminal and verified: deleted256412251 bytes from ts863-qualification-docs/.zig-cache,773269839 from ts-bench-snapshot-837/.zig-cache and186224174 from ts-839-parent/.zig-cache. All102404 protected source/raw/retained-image file checksums are unchanged; shared caches and unrelated jobs were not modified. Cleanup manifest retained in home-ts-owned-alias-prototype.wVgnmk/cleanup-outcome.json. After the verified disk-availability change, queued a project-only diagnostic build under the same guard; original session48512 is waiting on a newly live unrelated runtime guard34094, which remains untouched. No diagnostic compiler has started yet. Prototype patch SHA2563443bb87944a94deef7d1177812bf0a644cd4f88c80c0d33ec6164c37d90ad57 stays frozen, installed qualified native754926fd unchanged. Broader ownership/recursion controls prepared but not run. Review notes retain incomplete-alias cache deferral and watermark/OOM cleanup concerns for the next revision; source remains uncommitted and unqualified, with no timing/adoption.

Original queued session48512 acquired the machine lock after the unrelated runtime guard exited; diagnostic build is now live (compiler50409) on the exact frozen prototype patch3443bb87944a94deef7d1177812bf0a644cd4f88c80c0d33ec6164c37d90ad57. No restart or limit increase. Installed qualified native754926fd is unchanged. Cleanup terminal manifest independently confirms1215906264 generated cache bytes deleted and102404 protected file hashes unchanged. The project-only executable will retain full Zod output and all24 original forward-order/name/body diagnostic observations, not quiet CLI admission or timing. General ownership/nullable-recursion/circular-error reference controls are prepared separately; cache-deferral and rollback review follow-ups remain not implemented until this frozen experiment is terminal. No success/adoption claim.

Frozen prototype diagnostic image built successfully under unchanged guard (peak2857MB), SHA256074c84a5716737b31a460c400f503b5e1d795643c6a4172cff0f59791e93fb12. All24 original forward-order/name/body diagnostic observations now match exact reference code/exit contracts, with unchanged inputs. Full unchanged Zod locations show zero added diagnostics and exactly seven removedTS2339: all five original core/schemas.ts4585/4590/4592 template-loop false errors, plus classic/schemas.ts2833:47 and mini/schemas.ts1947:62. Remaining full project output is234TS +2HM9002 (236 total); it is still unadmitted/untimed. Complete stdout/stderr and records retained; installed production-qualified native754926fd and complete library inventory unchanged. The project-only image is NOT quietCLI admission or performance evidence. Broader scalar-before-declaration/unknown/shadowing/mutual/nullable-recursion/namespace/circular-error references are now running separately before hardening the incomplete-cache and rollback concerns. Full unfiltered source gates still required; no source adoption.

Broader reference controls are valid: both TS6/native TS7 satisfy all32 ownership/unknown/recursion contracts. First prototype diagnostic matches13/16: scalar forward-use, lexical shadowing, sibling-local aliases, productive mutual recursion, namespace aliases and exact twoTS2456 circular errors are preserved; explicit-unknown negative is silently accepted and nullable recursive members addHM9002/falseTS2339. Full failures retained, not filtered. Source evidence: adjustAmbientUnresolvedBareTypeAnnotation maps a named alias resolving legitimately tounknown intoany even when its type declaration exists. New revision preservesunknown based on actual type-space declarations/imports/global facts, retains unresolved-name recovery otherwise; also marks unresolved/active local-alias references as deferrals before cache publication and participates in Program watermark/OOM cleanup. Unfiltered Program validation of this material new revision has started under unchanged bounds, session96983. No adoption/timing; previous prototype/binary/failed guard remain retained separately.

Hardened revision passed the complete unfiltered Program suite:265/265, terminal exit0, exact source patch58a7bac68211364b0c974ea644e26444d868cd467f17e34e2cf6172bd7e5d643. New direct regression verifies forward scalar use, forward alias/alias-to-alias array elements, genuine unknown retainingTS2322, and productive mutual-cycle field types retainingTS2322/TS2339; no filters or weaker assertions. Same guard peak3168MB. Full logs/outcome retained in home-ts-owned-alias-hardened.Y3mOcP. Now building its project-only diagnostic image for full unchanged Zod and all40 original/general ownership observations, with expectations taken from validated reference records. Still uncommitted/unqualified for adoption: full checker/driver/CLI/entrypoint, realCLI admission, remaining diagnostic classification and exact-parent performance risk qualification remain. No timings or speed claim.

Hardened diagnostic image5ece118293d76c52cd041d4f4108b7d33dc0d54953bf7738bd5d37381f101916 matched38/40 complete diagnostic code/exit contracts. True unknown now retainsTS2322; all original24 controls remain exact, full unchanged Zod still has no added/seven removedTS2339. Nullable recursive members remain two failures withHM9002/falseTS2339, retained unfiltered. Source shows the typeof narrowing classifier receives a raw recursive alias union from member lookup and classifies reference tags rather than completed object surfaces. New material revision resolves that same owned surface before typeof classification, without member/body/diagnostic filtering. Extended the direct Program regression with nullable recursive property reads: seven exact genuine diagnostics total, including wrong-numberTS2322 and missing-memberTS2339; no relaxed assertions. Unfiltered Program run of this revision now live under unchanged guard, session79463. Previous265/265 result remains attached only to patch58a7bac..., not relabeled new source. No adoption/timing.

Resolved-guard revision passed all265 unfiltered Program tests, including the seven-exact-diagnostic regression now covering nullable recursive member reads as well as forward/unknown/mutual-cycle cases. Terminal0, no filters, exact patchddf445924c7c8235fb3e99d7a02f8af876b0446f2d79fbd74f94dc946ce91f73, same guard peak3130MB. No expectations weakened. Full logs and source retained separately in home-ts-owned-alias-resolved-guards.Idqxcg. Now running its project-only diagnostic build for all40 reference contracts and unchanged full Zod; next full checker and remainingCLI/module gates remain required. Installed qualified native754926fd remains unchanged; source uncommitted and untimed. Prior source hashes/results remain historical rather than relabeled this revision.

Resolved-guard diagnostic image717098989ccdb76766e8157a10c8fc9e8ae3bb73a76277c943d024467adbe5d2 matches all40 complete reference code/exit contracts, including nullable recursion and trueunknown. All input inventories unchanged. Full unchanged Zod delta vs qualified candidate has zero added and eight removedTS2339 (five core template-loop, classic2833:47, mini1947:62, and classic/from-json-schema525:76); all other diagnostic codes/counts remain. Full project remains233TS +2HM9002, unadmitted/untimed. Complete raw output retained; project-only observations are not silentCLI admission. Current exact source patchddf445924c7c8235fb3e99d7a02f8af876b0446f2d79fbd74f94dc946ce91f73 already passed265 unfiltered Program tests; full unfiltered checker now started under same600-second/3840-MB guard, session88881. Source remains uncommitted and installed qualified native754926fd unchanged. Remaining full module/CLI/admission and exact-parent performance gates still required.

Published documentation-only a31df53 to origin/main as Chris Breuer, conventional issue-linked commit with no coauthors (refs #863/#548). README now gives a brief scoped table and links the full held qualification report. Raw archive 20261011-owned-alias-qualification-held-879.tar.gz has 94 verified payload checksums, SHA256 80f1679787584993499a5abb23e86e6ce61aef243bd8b3409584a4725c5cc8b0. It retains three exact revision patches/images, all completed Program logs, broader reference/parent and prototype observations, full unchanged Zod output, failed attempts and scoped task-cache cleanup verification. Full local preservation manifest remains retained; unrelated hash-only paths and currently live checker log are excluded. Current source ddf4459... passed 265 Program and 40 diagnostic contracts; full checker original session88881 remains live, no restart/filter. Compiler source stays uncommitted, installed qualified native754926fd unchanged; no realCLI admission, timings, source adoption or ranking win is claimed. Report: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#declaration-owned-alias-qualification-held

Full unfiltered checker is terminal, not pending:4565/4568 passed,0 skipped,3 failed, exit1 (same guard peak3318MB). The new forward-alias regression passes, but adoption is blocked by unchanged broader contracts: negative member guards subtracting projected union properties (expected0,found2), named import-type namespace cycles (expected twoTS2456,found0), and recursive-array alias signature diagnostic display (TestUnexpectedResult). Entire command/log/outcome retained in Idqxcg; no expectation changed or failing case filtered. Current published docs/archive explicitly exclude that then-live gate and do not claim complete semantic qualification. Next is to inspect all three actual failures and correct alias resolution/ownership/diagnostic behavior before any source commit or timing.

New material revision addresses the three full-checker failures without changing expectations: discriminant receiver surfaces now resolve the same owned aliases before object/union classification, normal on-demand alias checking participates in the existing exported-type declaration cycle stack, and signature diagnostic display resolves completed alias surfaces under its existing identity-active display guard. The earlier4565/4568 run and all three failures remain retained. Full unfiltered checker rerun now live under unchanged600-second/3840-MB guard, session57097, with a separately frozen source patch in home-ts-owned-alias-full-regressions.3zmqMo. This is not a repeat-to-pass measurement: implementation changed and no benchmark timings exist. No source commit/adoption. Published documentation-only a31df53 remains a snapshot of completed Program/diagnostic evidence and explicitly pending full gates.

The three-regression correction revision stopped before tests at a compile error in the new defer cleanup statement (invalid assignment in single-line conditional defer). Full source/command/compiler error/outcome retained in home-ts-owned-alias-full-regressions.3zmqMo; it is not assertion evidence. Replaced that statement with valid block-form defer, formatting/diff checks clean. New separately frozen syntax-corrected full checker run started under unchanged guard, session27286, evidence home-ts-owned-alias-regressions-syntax-fixed.3pPBmo. No expectations relaxed, no discarded failures, no limit changes. Source remains uncommitted; previous three semantic failures must still be resolved and verified by the full suite before adoption/timing.

Syntax-corrected full checker94e265d... finished4567/4568 passed,0 skipped,1 failed. Qualified namespace-cycleTS2456 and recursive-array signature-display regressions now pass. Only projected-union negative-member guards remain. All assertions unchanged. Exact-source external observation with the prior diagnostic image reports twoTS2339 on issue.code in the else-if branches; its initial minimal ambient omittedPropertyKey, so both references instead emittedTS2304 and that setup is explicitly invalid, retained rather than treated as oracle. Prepared a complete ambient follow-up, but its guard admission was refused at233MB free before any compiler started. Only task-owned regenerated .zig-cache was removed after the finished run,176532479 bytes; all140 protected source/evidence/retained-image hashes unchanged. Disk subsequently fell again below start floor; task cache already gone, no source/raw/retained/global/shared data or unrelated jobs touched. Continuing source review of incomplete recursive interface/alias surfaces; runtime probes require a meaningful disk availability change. No timing/adoption or relaxed contract.

Prepared a further material correction for the sole projected-union regression: an alias requested from a recursive interface member can encounter that interface while its shape is in progress. New source retains a declaration-owned edge for in-flight non-generic interfaces instead of capturing an incomplete HIR shape, defers cache publication, completes the edge only from the finished declaration, and invalidates matching declaration state in contextual cache clearing. This is a source-derived hypothesis/correction, not yet runtime-verified. Exact prepared patchd3d99aa127a89f9ec9d62caf470a792feb745357902fa6e8be6567baa2283532 retained in home-ts-owned-alias-inflight-interfaces.jHEAoG. Formatting/diff checks pass; all old assertions remain unchanged. Runtime validation not started: machine free space remains about282MB, below unchanged1024MB admission floor, task-generated cache already removed, source/raw/retained compiler/global caches protected. Installed qualified native754926fd unchanged. Next full checker plus complete reference probe requires meaningful disk recovery; no blind repeats, raised limits, source adoption or timing.

Disk availability independently recovered from242MB to4.2GB, a meaningful external change; no filesystem redirection or guard override used. The previously unstarted complete projected-union reference probe now ran under the unchanged guard: TS6/nativeTS7 silently accept its exact source with properPropertyKey ambient, while retained diagnostic717098... reproduces two falseTS2339 on issue.code in the else-if branches. Prior invalid ambient and disk refusals remain retained. Full unfiltered checker for the prepared in-flight-interface-reference correction now started, original session65020, exact prepared patchd3d99aa127a89f9ec9d62caf470a792feb745357902fa6e8be6567baa2283532 in home-ts-owned-alias-inflight-interfaces.jHEAoG. No assertions changed, source still uncommitted/untimed and installed qualified native754926fd unchanged. Entire result will determine whether the remaining4567/4568 failure is corrected without creating others.

Prepared in-flight-interface-reference revision is now verified by the complete unfiltered checker:4568/4568, terminal0, no skipped tests, unchanged assertions, exact patchd3d99aa127a89f9ec9d62caf470a792feb745357902fa6e8be6567baa2283532. All three earlier full-suite regressions pass, including projected-union negative guards, named namespace-cycleTS2456 and recursive-array signature display. Same guard peak3347MB; entire source/log/outcome retained in jHEAoG. Complete external reference probe also valid/silent in TS6/nativeTS7. Now advancing this exact source through full Program and remaining modules, actual productionCLI build, extended/protocol5 admission, full unchanged Zod delta and exact-parent performance risk qualification. No source adoption/timing or ranking claim yet; all earlier failures and guard refusals remain retained.

Same in-flight-interface revision d3d99aa127a89f9ec9d62caf470a792feb745357902fa6e8be6567baa2283532 now passes both complete unfiltered suites: 4568/4568 Checker and 265/265 Program, terminal 0, no skipped/filtered/relaxed tests. Program guard peak 3096MB. Previous three-failure/one-failure/compile-error and disk-refused attempts remain retained. Advancing same source through full driver/CLI/entrypoint (original session79394), actual ReleaseFast stripped home-tsc build, 288 extended realCLI records and 180 six-image protocol 5 admission, with complete unchanged Zod output. No source adoption or timing; installed retained production-qualified native remains untouched until the new build succeeds and is frozen.

Same d3d99aa... source also passed complete driver201/201 and harness147/147. Sequential remaining-module runner original79394 is now onCLI, guarded process89916 confirmed live; no restart, filters or source changes. Full checker4568 and Program265 remain verified for this same source. ActualCLI build and288 extended/180 protocol5 admissions are prepared but not yet run; no timing/adoption. Baseline source65a0332fe vs currentmaina31df53df differs only inREADME/docs/evidence, and relevant build/compiler code search found no git-revision embedding; exact provenance/source-equivalence will be recorded before any A/B, not silently reuse an unrelated binary. All prior failed attempts remain retained.

Same exact d3d99aa... source now passed complete CLI78/78 as well as driver201/201, Checker4568/4568, Program265/265 and harness147/147. Original sequential session79394 remains live on entrypoint guard425; source and installed native hashes unchanged. No filtered/skipped/relaxed tests and no restart. Actual production native build,288 realCLI extended records,180 protocol5 records, full production Zod classification and exact-parent performance remain pending; compiler source still uncommitted and ranking unchanged. Raw gate logs remain associated only with this frozen source.

All required same-source unfiltered module gates now pass for d3d99aa127a89f9ec9d62caf470a792feb745357902fa6e8be6567baa2283532: Checker4568/4568, Program265/265, driver201/201, CLI78/78, entrypoint46/46, harness147/147. Original remaining runner79394 terminal0, last guard peak3020MB. No skipped/filtered/relaxed assertions. Actual strippedReleaseFast production home-tsc build now started under unchanged600-second/3840-MB guard; the builder freezes a candidate image and restores the installed retained754926fd image after a successful build. Next288 realCLI extended records,180 six-image protocol5, full unchanged production Zod delta, then frozen exact-parent performance risk protocol. No source adoption/timing or benchmark win yet. All earlier failed builds/tests/admissions remain retained.

Actual production native build original36502 remains confirmed live under unchanged guard25222/compiler26368; source d3d99aa... frozen and all source/module gates terminal passed. No candidate image or build-success record yet, so no actualCLI/admission claim. Prepared an additional exact projected-union positive/negative pair preserving the original guard chain and requiring two genuineTS2322 plus twoTS2339 in its negative; both references must validate it before extending admission. It is not yet run/claimed. Native builder will freeze the candidate and restore retained installed754926fd on success. Extended288 and protocol180 admission remain prepared; timing not started and source not adopted. Baseline provenance/source equivalence must be explicit in later performance protocol.

Actual strippedReleaseFast home-tsc build succeeded, frozen candidate SHA256 bb4c00dd2fdfba719098150082b696b4f9ba8ab65a7a47640b197540420cf771, exact source d3d99aa..., same guard peak3117MB. Installed retained qualified native754926fd restored and SHA verified. New projected-union pair validated in both references: positive silent, negative exact twoTS2322/twoTS2339; full old-image observations and inventories retained in ZVQ5Hb. Extended actualCLI admission now includes98 projects*three images=294 records (original56 +40 ownership/order/name/body +two projected union), with all inputs unchanged and no discarded outcomes; original session5453 running. Separate180 protocol5 admissions and complete production Zod remain next, then exact-parent performance if all semantics qualify. No source adoption or timing claim.

Actual production candidate bb4c00dd2fdfba719098150082b696b4f9ba8ab65a7a47640b197540420cf771 passed all294 extended realCLI admissions:98 TS6,98 nativeTS7,98 Home, zero failed records, complete input inventories unchanged. This includes the reference-validated projected-union pair; positives are silent and negatives retain exact contracts. Complete actualCLI Zod output remains233TS+2HM9002, exit1, unchanged full library inventory and unadmitted; no filtering or support claim. Now running separate six-image180-record protocol5 admission under sameguard, original90584. All source gates, candidate/build/output provenance and installed754926fd restoration retained. Timing and source adoption still pending.

All180 six-image protocol5 admissions passed with unchanged full executable/input provenance and independent compare.validate_admission verification. ActualCLI extended294 records also passed, all module/harness gates pass, source d3d99aa... and frozen candidate bb4c00d... verified. Full production Zod adds zero diagnostics:13 removed versus exact-source-equivalent baseline and eight removed versus prior candidate; remaining233TS+2HM9002 still unadmitted/untimed, no whole-library support claim. Before any timing, froze performance-protocol.json with all20 ten-pair screens, fixed12 independent30-pair risks and exactlyone independent confirmation for additional wholly adverse screens, paired bootstrap seed879/20000, all observations retained, no reruns/alignment tuning, zero-spanning not equivalence. Baseline-source-equivalence.json proves onlyREADME/docs/evidence differences between baseline65a0332fe and currentparenta31df53df; same build/compiler sources, both native SHA recorded. Correctness-feature adoption requires no wholly negative independent risk interval, not a claimed primary speedup. Timing not started yet; paired runner/statistics syntax verified.

Original complete200-pair screen finished and independently verified all order/exit/admission/provenance/raw files, zero discarded samples/reruns. Screen flags wholly negative paired mean-saving intervals in type_predicates_large, variadic_tuples and commonjs_graph (already fixed independent list), plus structural_objects (the one additional adverse workload). All20 row statistics retained; favorable screens are not adoption/speed claims. Started the full frozen independent risk sequence: fixed12 workloads plus exactlyone30-pair structural_objects confirmation, original session89232. Each phase remains guarded600/3840 and independently verified, all samples/admissions/warmup outputs kept. Any wholly negative independent risk interval blocks adoption; zero-spanning not equivalence. No source/runner edits or measurement reruns after observing screen; source still uncommitted, installed754926fd restored.

Landed source 4b4060e and separate documentation/evidence 8c1ec10 are confirmed on origin/main, both authored/committed Chris Breuer with no coauthors and conventional linked messages. Committed source diff exactly matches fully qualified d3d99aa... patch; installed adopted imagebb4c00dd... verified. All source gates,294 extended realCLI,180 protocol5 and complete frozen590-pair/104-phase-admission qualification verified. All13 independent risk intervals are not wholly negative; two show saving and11 spanzero (not equivalence/no slowdown proof). All four adverse screens and every sample/failed revision retained. Complete report/README table published, archive20261011-owned-alias-qualified-landed-879.tar.gz SHA2561852d6d51008a90bce47b0abf5142469a60d2968654bc3cea7131544107531d3 with947 verified payloads. Previous five-pin ranking is older source, fresh matrix remains next. WholeZod233TS+2HM9002 still unadmitted/untimed and universal goal incomplete. Report: https://github.com/home-lang/home/blob/main/docs/docs/TS_PERFORMANCE.md#declaration-identity-recovery-landed

Fresh complete five-pin matrix started after landed source/docs publication, original session63708, unchanged20 workloads,30 runs/threewarmups, existing pinned Darwinarm64 manifest. Supported HOME_TSC override selects exact frozen bb4c00d... executable; historical worktree binary/source are not overwritten. Standard runner re-admits all workload positive/negative contracts before timing and rotates compiler order; all raw rounds/provenance retained and guard600/3840 unchanged. This is a new landed-source ranking measurement, not a rerun of the prior paired qualification or a selective subset. No rank result claimed before terminal verification.

Landed image4b4060e05 (bb4c00d hash) now has a complete independently verified five-pin continuation published by bd0d82e:600rounds/3000samples/150admissions, zero repeated recorded samples. All1502 archivedpayloads verified; README/full20row report updated. Paused/sharedhost medians1/20 lowest,9/20 fasterTS7 are descriptive, not causal source-risk conclusions. Original exact-parent590pair correctnessfeature qualification remains separate and unchanged. FullZod233TS+2HM stillunadmitted; #548/#416 remainopen.

Current-main audit at ca881af (Darwin arm64, 2026-10-19): rebuilt zig-out/bin/home-tsc from this source under the 3,840 MB guard (SHA-256 aedafabe5be1023be138e500cfdc335dd36630f1a4c3128f40d14d1f9dc6992f; build 3/3, peak tree footprint 2,179 MB). Replayed all eight unchanged projects from the published 20261011-forward-alias-diagnostic-879 archive, one guarded CLI process at a time: all four positives exit 0 with no diagnostics; all four negatives exit 1 with exactly TS2322 at main.ts(15,76) and TS2339 at main.ts(15,110). The targeted existing checker forward-alias regression passes 2/2 (guard peak 1,758 MB); the unfiltered Program target passes 322/322 (peak 2,304 MB). The unfiltered checker target reached its 300-second guard wall bound (exit 124) before a result and receives no passing credit. These observations show the archived eight-project symptom is not presently reproduced by the fresh CLI, but do not prove the unchanged Zod graph, all reference compilers, or #879’s complete acceptance. No source change or benchmark timing is claimed; leaving the issue open.

Additional bounded audit on current main: the repository’s pin-verified Zod 4.5.2 ratchet was invoked with the freshly built home-tsc under the 3,840 MB ceiling and a 300-second wall bound. The process terminated with exit 124 before emitting a diagnostic count or writing its optional current/diagnostics outputs. This is an unqualified attempt, not a Zod admission result; the prior eight archived reductions and 322/322 Program result remain the only positive evidence from this host. No source change or benchmark timing followed this timeout.

Clarification after direct pinned-project diagnosis: the eight archived forward-alias order controls still pass on current main, but the unchanged 106-file Zod graph remains unadmitted for a separate expanding structural inference path sampled in Checker.inferFromPair. Focused #921 tracks that measured cycle. The provisional exact-pair guard was rejected after a 120-second bounded timeout and reverted; no #879 completion, Zod diagnostic count or speed claim follows.

Current-main recheck after the landed declaration-identity work: on c57880c, the retained forward-alias reduction archive 20261011-forward-alias-diagnostic-879.tar.gz was restored to a temporary directory unchanged. Rebuilt home-tsc and installed TypeScript 5.7.3 were run serially under the 3840 MB guard on all eight part-forward, both-forward, literal-forward and schema-forward positive/negative projects. All four positives were clean on both compilers. Every negative matched exactly in diagnostic code, message and location (one TS2322 and one TS2339); only the compilers' nonzero exit-code conventions differ. This verifies the reduction on the present checkout, not the issue's full Zod/TS6/TS7/performance acceptance. #879 stays open; no new source change is claimed.