L4 live-adapter validation — Fusion 2704.1.53

Date: 2026-09-05 Fusion: 2704.1.53 · Python: 3.14.0 · Platform: macOS x86_64 Method: python3 tools/fusion_exec.py tools/probes/<probe>.py

Per CLAUDE.md section 11, L4 cannot run on shared CI (Fusion is GUI-only and licence-bound), so this record stands in for it. Every probe below is tracked and re-runnable verbatim, which is what makes [OPEN-6] (Windows) and [OPEN-7] (Apple Silicon) parity cheap to establish later — re-run these eight files and diff the output.

Probes

Probe	Covers	Result
p01_environment.py	version, interpreter, environment	recorded
p02_fixture_corpus.py	programmatic fixture corpus (PRD 13.2)	recorded
p03_api_facts.py	the 11 CLAUDE.md section 6 facts, C1–C11	22 verdicts, 0 REFUTED
p04_entity_token.py	[OPEN-1] token durability	RESOLVED POSITIVE
p04b_state.py	sandbox census and cleanup	clean
p05_curvature_seams.py	[OPEN-3] curvature at seams	RESOLVED POSITIVE
p06_texture_coordinates.py	[OPEN-8] textureCoordinates	RESOLVED POSITIVE (5-face corpus)
p08_cone_parameterization.py	R-F1 Finding 13, cone parameterization	tan form REFUTED

Outcome

• 0 refuted among the re-confirmed section 6 facts.

• 2 refined — the cylinder v window is per-face, not an invariant (PRD 7.14); the entityToken 316-character figure is not stable.

• 2 refuted — PRD 6.0 v1’s cone tan form and PRD 11.8 v1’s u = z_rel/radius.

• 3 new stub/runtime divergences (PRD 11.14, CLAUDE.md section 5).

• Timings are recorded as ranges, not constants: cloud settle 2.9–5.3 s, deleteMe() listing lag up to 95.4 s.

Full evidence and per-claim numbers: PRD sections 11.14 and 11.15, and the commit bodies on test/live-fusion-recon (merged at c27cdbf).

Live-Fusion state at close

0 documents open · Mighty Miter Test project intact with 0 data files · the stakeholder’s designs were never opened.