Isolated label fields: closed 11c605f build and failed normal gate
Source: 11c605f9170eae461913410f83f333e9314714a5. The experiment remains unwired to production. A local wrapper compiled the real module and its subtests directly against existing pinned Iced, unicode-segmentation and serde_json artifacts. The build recipe and expected membership are retained verbatim.
| Closed stage | Actual result | Evidence |
|---|---|---|
Isolated rustc --test build |
Exit 0; empty compiler log; 957 bound inputs unchanged | receipt, log, toolchain |
| Explicit v2 listing | Exit 0; all 25 expected test names, 0 benchmarks; selection passed | receipt, log |
| Explicit v2 normal suite | Exit 101; 21 passed, 2 failed, 2 ignored, 0 filtered; 25.45 s test time | receipt, log |
The normal target was 23 passing tests plus two intentionally ignored tests. Both failures were fixture preconditions in field_tests.rs: advanced_ligature_partial_field_uses_whole_shared_cluster_not_fraction at line 274 and auto_ascii_to_ellipsis_advanced_transition_uses_final_buffer at line 595. Each expected an ffi ligature cluster from the generic font selection; the actual shaped buffer did not satisfy that precondition. Both stopped before calling field_layout. This establishes a concrete evaluator/font-assumption failure, not a demonstrated field-geometry defect or a passing ligature evaluator. A font-fixture correction was subsequently requested; no correction, corrected build or corrected execution is adopted by this archive.
The normal receipt preserves summary: null because its parser stopped at the first unexpected FAILED status. The raw log contains the complete final failure summary and all 25 outcomes; archive verification independently checked that membership and the 21/2/2 counts. The two ignored checks—global font-loading invalidation and the bounded high-end lookup timing diagnostic—were not executed by this gate. There is no font or timing acceptance here.
Earlier runner diagnostics
An initial H alias collision failed wrapper preflight before a test process was invoked. After that helper was renamed, the original wrapper's intended list dispatch instead executed the default suite: 21 passed, two failed, two ignored. Its receipt records an intended command, not proven actual argv; it failed selection and is not an accepted listing or normal gate. The raw log, dispatch diagnostic, and subsequent literal direct-list output are preserved. The H preflight produced no test receipt; the archived run.ps1 is the helper-renamed original dispatcher, not a claimed copy of the earlier preflight-failing bytes.
run-v2.ps1 uses explicit native arguments for each stage. Its actual list and normal receipts above are the accepted execution records, including the normal failure. No earlier receipt or diagnostic was rewritten to manufacture a passing gate.
Identity and scope
The build ran from 10:50:46.9819788 to 10:58:37.3860306 UTC on 2026-09-13. V2 list ran from 11:20:17.9166806 to 11:20:22.5628349 UTC; v2 normal ran from 11:23:04.6338872 to 11:23:34.0337415 UTC. Build captures are identical; each run's bound before/after records are identical. V2 list binds 14 inputs and normal binds 16, including the prior successful list receipt/log. The build's existing dependency provenance comes from the closed 0409 desktop listing; this isolated build does not repeat or extend desktop acceptance.
The excluded test executable is 14,145,536 bytes, SHA-256 eb9c65aadc0e9fdb9ce23e42cc69055b842fac97a5e45207467cdf843eb3e475. The excluded libunwind.dll is 190,464 bytes, SHA-256 aa282fa47bb2c4e210cea5e470cdbdca58568f0be0bc73e404d86ab9cd95ef6a. Their identities are retained in the receipts and manifest; their bytes are not included. This is CPU Iced/Cosmic shaping and cache evidence only, with system-font dependence. It establishes no WGPU paint/ink containment, production field-navigation policy, native UI behavior or full application acceptance.
Archive integrity
All 21 original nonbinary files (556,771 bytes) are copied byte for byte under original/, including the original preparation notes, both runners, build captures, dependency information and all five logs. Historical absolute machine paths remain unmodified provenance; the links here and the pinned source commit make the evidence accessible without another machine's .local directory. Reproduction requires arranging a fresh source checkout and compatible dependency artifacts; the archived local-path recipes are not claimed to run unchanged from this archive location.
manifest.json records copy hashes, byte counts, Git blob hashes, binary exclusions and independent receipt/inventory checks. It excludes itself to avoid a circular hash. Local attributes preserve raw bytes through Git. All five original/attempt1/*.log files are Git-ignored and must be explicitly staged with git add -f; the empty compiler log is also required. Archive preparation performed no compilation, execution, staging, commit or publication and did not touch the separate live core probe build.