Wetland full-sequence paint trace — November 3, 2026

Previous goal turn classification: progress. Repeat the exact executed original six-case order: current, original span1.25 and retained diagnostic span1.25 at desktop1440, then the same three controls at phone390. The earlier suggestion that phone contexts preceded the original failing desktop case was incorrect; both executed result files show desktop first. Phone warm-up is not a supported cause.

The30 new normal-client navigation frames use the sole disposable QA world, issue no world commands and preserve every artist byte, public source contract and public algorithm file. Observation wrappers invoke the original paint function and collect base-paint inputs/hashes before final terrain composition. In all six cases, base paint, final ground and foreground exactly match after reload. All12 complete screenshots match previously directly viewed witnesses byte-for-byte, so no new screenshots or raw buffers are retained.

This does not reproduce or explain the original32-pixel failure. Its original failed comparison and distinct reload hash remain preserved. No failing frame has yet been captured with the stage trace; neither the failing stage nor root cause is established. Deterministic reload, whole-ground nonrepetition, physical grain/palette/contact and full fidelity/performance/lifecycle remain open. Avoid further identical repeats without a new input-state or cache/lifecycle hypothesis.

Cumulative wetland evidence is80 frames:60 original-source and20 retained diagnostic-source, including50 private-contract frames. Forty of the80 frames use observation instrumentation. This stage adds30 such frames, with10 existing diagnostic-source and20 private-contract frames. Browser contexts and owned server are terminal. See exact validation validation-summary.json, executed frames world-results.json and resource closeout owned-resources-terminal.json.