Rejected native-alpha-mask perimeter path

The private native-alpha-mask path is rejected. In four actual native-image browser geometries, resampling that mask changed alpha on 516 previously opaque painted display pixels; the retained core guard stopped publication of those draws. The normal-client matrix recorded 6 baseline views, 9 baseline lifecycle observations with detailed exactness unavailable and 3 failed masked cohorts. No valid masked appearance comparison was obtained, so this is a technical rejection of the proposed renderer path, not a rejection or alteration of the native artwork. Keep the original source and smaller study pose unchanged. The separate display-partial-alpha-only proposal preserves all 1,111,595 opaque pixels in four isolated guarded geometries, but still needs normal-world, visual, lifecycle and performance checks. Private world views now exist; full artwork/contact, wetness, state and admission gates remain open.

The exact unchanged source is WOG-A-009138, 1584x1072 PNG, SHA-256 c3111d24e68ce1f76ae729ce24c4de0767e2b4ba84e59debcfc6c9bb703e9939. The generated original WOG-A-009137 and wetland floor WOG-A-003578 are also freshly hash-verified. The source image was drawn once at the previously selected smaller world placement; this trial changed neither source RGB nor the encoded native alpha file. No pattern, source crop or mirroring was introduced.

The failed method drew a separately sampled native-alpha canvas over the display image with destination-in. At display boundaries that second resampling can make a formerly255-alpha pixel translucent. Four isolated geometries from the prior actual world captures reproduce the guard failure. A separate diagnostic controller removes only the throw to count the defect; that diagnostic is not an accepted painter and is never used as a passing world candidate. The guarded normal-client matrix remains failed. Its final aggregate assertion prevented writing detailed baseline rows/comparisons;only counts survive,so baseline exactness is unproven and is not counted as a pass.

Normal framing still follows the admitted foreground reed rather than a contract for this unadmitted finite overlay. The planned desktop/landscape checks require complete finite-rectangle containment, but detailed rows from this failed matrix were not retained, so this package does not establish those bounds. Recheck them in the corrected normal-client study. The prior retained portrait-phone evidence still shows clipping of the unadmitted overlay; complete-edge integration remains open. No new candidate comparison sheet is retained because the guard prevented valid masked frames. Comparisons/readbacks remain in memory.

The corrected proposal changes display alpha only for pixels already in1..254 and retains the original opaque-core guard. Four isolated guarded browser cases preserve 1,111,595 opaque display pixels exactly. This does not certify its normal-client output, terrain contact, all lifecycle states or performance; those checks must remain separate. The native-mask trial is retained as compact failure evidence rather than silently removed from the history.

The Atlas approval flag remains false. Its stale no-world-frames explanation is corrected to acknowledge private context evidence while retaining all remaining qualification gates. The library scope, other approvals, original artwork, source permissions, public renderer, geography and saves remain unchanged.

Evidence: failed matrix summary normal/failure-summary.json, core failure diagnosis core-diagnosis.json, owned terminal proof normal/owned-resources-terminal.json, source-bound failure decision failure-review.json, original retained source/context evidence and separate corrected isolated proof ../mire-display-fringe-alpha-20261104/core-diagnosis.json.