Label-field evidence at 4c9bc87
All 24 functional tests passed: 21 normal tests and three separately isolated tests. The optional timing diagnostic also passed its exact geometry checks, but exposed severe debug-build long-field lookup cost. This is an unwired CPU Iced/Cosmic experiment, not application, WGPU paint, native responsiveness, operation-field navigation, or product font-policy acceptance.
The tested source is 4c9bc87e4e66bba573dd63b23df5ea039f85d0ca. The standalone wrapper imports that real module; it does not replace the implementation. Source bindings identify all 11 prepared source inputs, including the font fixture.
Closed observations
The strict standalone rustc --test -D warnings build exited 0 with an empty compile log. Its reference was the successful 8bf desktop LIST receipt, SHA-256 e2d9258afab904c7d0d9f3efbb3c3cc02511a33296c1698b3114b3d6b7a125a1. This reference binds dependencies and toolchain; it does not turn these tests into desktop integration tests.
| Stage | Actual result |
|---|---|
| Literal list | Exactly 25 names; no executed-test summary |
| Normal | 21 passed, 0 failed, 4 ignored, 0 filtered |
| Font mutation, isolated | 1 passed, 24 filtered; actual font faces 167 to 168 and version 0 to 1; retained label and old session rejected |
| Partial ligature field, isolated | 1 passed, 24 filtered |
| Auto shaping/ellipsis transition, isolated | 1 passed, 24 filtered |
| Timing, isolated | 1 passed, 24 filtered; coverage and exact geometry passed |
Each isolated stage used a fresh process and one exact ignored test. The normal run's four ignored tests were not silently counted as passed. Verified results preserve summaries and timing JSON; the unchanged original attempt records contain raw logs and receipts, including the exact commands and predecessor bindings.
Timing finding
On Windows x86_64 with debug assertions and warm fonts, the diagnostic timed one resolution followed by three separate whole-field field_layout calls. It did not time 4,096 independent lookup calls. An independently constructed actual Paragraph oracle was computed before timing; each lookup asserted full source coverage and exact rectangle equality.
| Input | Glyph-layout rectangles | Resolve | Three whole-field lookups |
|---|---|---|---|
| 4,096 plain scalars/graphemes | 4,096 | 0.242 s | 6.647 s / 3.877 s / 2.402 s |
| 4,096 combining-heavy scalars, 1,024 graphemes | 1,024 | 0.111 s | 1.504 s / 1.904 s / 1.268 s |
The actual timing font was Segoe UI. Advanced shaping, zoom 1, and a 1,000,000-unit width avoided truncation. These observations require investigation before a responsive interaction claim; they establish neither a native performance threshold nor optimized-build cost. No performance acceptance threshold was applied. The diagnostic's 19.21-second test duration also includes work outside the reported lookup samples.
Provenance and reproduction
Archive manifest records 29 byte-identical copied files totaling 633,185 bytes. Archival verification compared raw originals and copies, rehashed all 966 current build inputs and 32 unique current run inputs, checked before/after equality, exact list/execution membership, and every sequential receipt/log binding. No compiler or test was invoked during archival.
The executable and runtime DLL are omitted; their inspected identities are retained in the original receipts and verified results:
| Omitted artifact | Bytes | SHA-256 |
|---|---|---|
scenario-label-harness.exe |
14,755,328 | 6736938b4437383d5445c3ac35d09ed188b408e7c7167b31d1e207441a246016 |
libunwind.dll |
190,464 | aa282fa47bb2c4e210cea5e470cdbdca58568f0be0bc73e404d86ab9cd95ef6a |
NotoSans-Regular.ttf, available at the source pin |
556,216 | 2ec33f84606cbaa0a1a944488e14f97faf2f6a25ecdd8354f5358f06da13c7d9 |
The Noto fixture supports isolated font tests; it was not the timing font. Its unchanged provenance and SIL OFL license identify the pinned cosmic-text 0.15.0 package and original font bytes.
For reproduction, fetch the exact published source and follow the preserved recipe, supplying a matching successful desktop LIST receipt and its verified dependency artifacts to build.ps1. Use a fresh output directory, then run run.ps1 sequentially with stages list, normal, font, ffi, auto-transition, and optionally timing. The scripts validate local paths and hashes; archived absolute paths are provenance, not portable peer-machine paths. Global font mutation must remain isolated.
The original recipe's preparation-only wording is historical; the closed records above are the later execution evidence. Earlier failed 11c/list experiments are not copied or relabelled here. Production wiring and GUI/native acceptance remain outside this archive.