[leehack]

Summary

Clears these items on tracker #1565 (not the stable genkit item #1574):

• parseRendered Dartdoc (#1573 audit): engine equivalence needs parseToolCalls set like the engine, the engine's partial reparse after a PEG failure, copies (constructor, fromJson, isolate transfer) losing the marks, and reasoning chunks dropped with thinking off. Class-doc example fixed.
• Hermetic build_hook_stable_diffusion_integration_test.dart ("the stable_diffusion companion selects and links it in process"): seeds the LiteRT-LM macOS arm64 cache from fake libraries. Passes with outbound network denied (sandbox-exec); on main it fails there downloading litert-lm-native-runtime-macos-arm64-v0.18.1-4.tar.gz.
• Forced-open Seed-OSS (#1572 lane): a thought the prompt opened is now the reply up to the first </seed:think> as generated (keepsForcedThoughtVerbatim, as DeepSeek V3, Cohere2 MoE, Qwen3-Coder XML), so a repeated <seed:think> and text before it stay in the reasoning, as llama.cpp 620d2225 keeps them. Stream and parse agree; a thought closed for thinking off and later reached by a thinking tag streams like Cohere2 MoE.
• Reuse a cached native bundle only while it matches its archive; pin Seed-OSS ASCII trim #1572 audit items: the marker takes the new archive mtime after a successful rehash (no rehash every build); a cache with no marker logs an info migration message, a real mismatch still warns; tests kill mutants A1, A6, A9, A13, A14 and "no ASCII trim at all" in _endsUnopenedSeedOssThought (B12 too: a NBSP-led public parse now pins its reasoning).

Production-readiness scope

• User-facing scope: forced-open Seed-OSS reasoning matches llama.cpp; quieter hook logs on cache upgrade; Dartdoc.
• Supported platforms/paths: all targets (parser); native hook (all bundles).
• Unsupported or intentionally unavailable paths: llamadart trims the reasoning, so a leading NBSP that llama.cpp keeps in <seed:think>Plan is trimmed (llama.cpp keeps leading space too); content, tags and later text are identical.
• Out of scope / follow-ups: public ChatTemplateEngine.parse on the stock Seed-OSS template (#1565, 0.18 candidate).

Completeness checklist

• Declared scope is fully implemented, or explicitly reduced above.
• Unsupported platform/option combinations fail loudly or are documented: N/A, no new option.
• Changelog updated (both files); no API change.
• Regression coverage covers the original issue plus negative paths.
• Validation coverage registry: N/A, no public capability change.
• Security/privacy: no secrets, tokens or paths added to logs.
• Follow-ups tracked: None beyond the above.

PR type guidance

• Bugfix PR: forced-open Seed-OSS used the generic thinking extractor, which dropped a repeated start tag (and merged text before it into content); it now uses the verbatim forced-thought path. Tests: closed_thought_upstream_test.dart (values from the llama.cpp 620d2225 oracle), hook and marker tests. Platforms: all.
• Feature PR / Docs-only PR: N/A.

Test Plan

• dart run tool/prepare_workspace.dart
• dart format --output=none --set-exit-if-changed .
• dart analyze; dart run tool/testing/check_platform_boundaries.dart
• TMPDIR=$(mktemp -d) dart test -p vm -j 1 --exclude-tags local-only: +9799 ~8, all passed (re-run at the audit-fix head)
• dart test -p chrome --exclude-tags local-only test/unit/core/template test/unit/core/engine: +2139 ~1, all passed
• tool/testing/run_template_parity_suites.sh exit 0; test/unit/tooling +1240 ~5
• Hermetic check: sandbox-exec deny network-outbound, whole build_hook_stable_diffusion_integration_test.dart +34 passed (fails on main with the LiteRT-LM download)
• Failing-first: the new Seed-OSS test fails on main's parser; each new hook test fails against its mutant below

Matrix Evidence

Matrix row	Scope covered	Platform / model / backend	Result	Evidence / notes
static-format-analyze	format, analyzer, boundaries	macOS arm64	PASS	0 changed, no issues, OK
root-vm	VM suite	macOS arm64	PASS	+9679 ~126
root-chrome	template/engine subset	Chrome	PASS	+2139 ~1
oracle	llama.cpp 620d2225 vs engine path (whole, bytewise, every 2-split, final parse)	#1569 audit harness, base 176e12e4c vs head	PASS	main 185,160 rows: 0 changed; X 37,056: BAD->OK 234; Y 7,440: BAD->OK 336; whitespace 480: BAD->OK 120; OK->BAD 0 and new stream!=parse 0 everywhere; fixed stream!=parse 19/44/18 (X/Y/Z). Seed-OSS stock rows unchanged; later first-content than upstream: 9 rows of X ("Hello seed:thinkthink</seed:think> world", 1 char)
high-risk-exact-head-independent-qa	independent blocking review of exact head	fresh auditor session, exact head 0629111	PASS	CLEAR: CLEAR delta re-audit at 0629111; evidence /opt/UnitySrc/personal/llama/llamadart-evidence/pr-1575/evidence-1575-0629111.json

Mutants (all killed, incl. audit H3 wrong refresh digest and uncaught refresh failure): A1, A6, A9, A13, A14, A15 (no marker refresh), A16 (always warn); Seed-OSS: old forced-open parse, trait off, stream marker branch off, no ASCII trim, Unicode trimLeft() in _endsUnopenedSeedOssThought.

Review Notes

• Independent review status: CLEAR at exact head 0629111 by a fresh auditor session (0 P1, 0 P2).
• CI status / head SHA: fill after CI runs

High-risk regression review

• Classification: high-risk (artifactConsumer, backendRuntime, structuredOutput)
• Implementation task: this lane
• Independent blocking QA task: fresh auditor session, CLEAR, 0 known PR-caused P1, 0 unresolved threads
• Exact head / current base: 0629111 / 0bddcde (merge-tree clean)
• Production-branch deletion, bypass, or miswire proof: N/A (trait wired in chat_format_registry.dart; mutant with the trait off is killed)
• Affected-family real-model/artifact evidence: llama.cpp 620d2225 oracle rows above; no real Seed-OSS weights run
• Explicit unavailable-family or other N/A evidence: Seed-OSS weights unavailable locally
• Known PR-caused P1 regressions: 0
• Unresolved review threads: 0

[github-actions]

Chat app preview removed for leehack/llamadart-chat-pr-1575.