Private wetland ground-scale calibration

The approved original WOG-A-003578 and existing separate diagnostic WOG-A-009265 remain byte-exact. Three controls compare original span2, original span1.25 and diagnostic span1.25 using only private in-memory source/scale contracts. No new painting is generated and no native image is downscaled or recompressed. The same legitimate empty wetland site and sole disposable QA world are reused without world commands, terrain edits or fabricated entities.

Thirty normal local-client frames cover desktop and phone DPR2. All twelve retained full Region-close/Hex scenes were directly reviewed. The finer original reduces apparent root/leaf size, but diagonal grouping, dry-looking floor and weak reed separation remain. The finer diagnostic retains recognizable leaf/stem chains and warm/dry colour. This does not justify replacing the original or accepting scale reduction as a complete solution.

Six Region return comparisons and four cross-mode Hex foreground pairs are exact. Five Hex reloads are exact. The desktop originalFine case has 32 changed ground pixels with maximum channel difference 245 despite equal context bounds; foreground remains exact. Keep that discrepancy for focused diagnosis. Navigation finishing without an exception does not make all pixel-fidelity comparisons pass. Region foreground, wetland shoreline/building contact, wider states, full cost/pressure/lifecycle and live use remain unqualified.

All six browser contexts and the QA server closed normally. Four complete control PNG witnesses are byte-identical to earlier retained controls and now share native storage; all paths and every PNG byte remain intact, reclaiming 17,229,069 duplicate physical bytes. No raw readback files, new world or rollback baseline is retained.

The two source-specific findings are registered in the town/volcanic batch and await its focused publication. The earlier town/wetland release does not contain this calibration. No private scale, painting replacement, source admission or public activation is accepted.

Evidence: visual review visual-review.json, validation validation-summary.json, world comparisons world-results.json, source and scale controls source-http.json, storage sharing storage-sharing.json, terminal resources owned-resources-terminal.json.

A separate focused desktop replay adds ten normal-client frames using only the unchanged original at spans 2 and 1.25. Both fresh Region returns and Hex reloads are exact. Four complete PNG witnesses match the previously viewed calibration scenes byte-for-byte and share their storage. The initial finer-original ground hash matches both replay captures; only the earlier reloaded frame has the distinct 32-pixel result. The original failure remains retained, its cause is not established and deterministic reload remains unaccepted. No player control, prerequisite, cost, staffing, timing or success check changes.