Arctic source-cache qualification — November 5, 2026

The original limestone, finite ground and arctic glyph paintings are preserved. The native-owner approach remains under qualification: the actual 128 MiB client cache reveals a desktop repaint discrepancy that the earlier 256 MiB scope did not expose. Source approval, renderer availability and game-ready qualification remain separate.

The completed investigations observe 36 normal Region/Hex views and 23 exact regular lifecycle maps. The first eight-view attempt stops before two desktop Hex records because the actual LRU assertion fails; those missing observations are not counted as passed. Later focused captures measure a 25,434-pixel raw foreground difference. Four exact PNG/native-decoder pairs, all six image draws, and the original full-number transform, save/restore and paint-state arguments remain identical. Independent replays reproduce the later reference but do not recreate the initial failing foreground; the underlying raster cause is unproved.

The visible-world comparison covers all three complete projected source rectangles over the real terrain at native output density. Actual LRU/reload changes 25,098 visible pixels, with maximum channel difference 53, at the same camera and bounds. The complete map changes in the same subject area. Taking the screen captures does not change the sampled raw foreground. This is a visible rendering failure, not merely a hidden-alpha serialization observation. Native before/after and difference ../arctic-visible-cache-pressure-20261105/normal/visible-complete-subject-before-after.png.

The earlier exact-alpha-255 core count is insufficient to describe interior fidelity: the limestone has only 51 fully opaque source pixels and 252,370 partially transparent pixels; the glyph painting has no fully opaque pixels and 416,095 partially transparent pixels. Most painted surfaces use partial alpha. A zero opaque-core difference therefore does not establish that only outlines changed. The original source pixels remain unchanged.

Two alternatives are rejected for selection: native source ImageData materialization and an explicit draw-oriented foreground context each complete eight views and six regular lifecycle maps but reproduce both desktop LRU failures. A frequent-read foreground context completes eight views, six lifecycle maps and twelve exact LRU/revocation/fingerprint restorations; however, it changes 185,790 desktop and 43,584 phone foreground pixels from the earlier reference. Independent complete replays bridge both actual candidate foregrounds, and native comparison witnesses retain the painted detail, but this alternative is not selected. Desktop native comparison ../arctic-owner-consistent-context-20261105/context-representation-1440.png, phone native comparison ../arctic-owner-consistent-context-20261105/context-representation-390.png.

Chromium's implementation distinguishes an omitted willReadFrequently setting from explicit false in automatic readback fallback. That observation motivated the explicit-false experiment; its failure here means the source inspection does not establish the cause in this Edge 156 binary. Chromium implementation.

The full-size ground's boundary, coarse grain and snow-contact concerns remain unresolved. The preferred original-RGB alpha-domain study WOG-A-009281 is not combined with this owner yet. Other memory tiers, full renderer cost, clean production integration and relevant role/state scope remain open. No artist PNG, source contract, public renderer, geography, saved world or game-ready approval changes.

Eighteen verified redundant or unselected QA files were retired, reclaiming 7,793,128 bytes. All observations, per-pane hashes, complete comparisons and unresolved gates remain; three identical counterexample PNGs now reference the original exact witness. No raw RGBA files or catalog/ledger mirrors were created. Both Citadel hamlets and every original/final artist file are retained. Retirement plan temporary-retirement-plan.json, completed cleanup temporary-retirement.json.

See the compact complete review review.json, three individual findings visual-review.json, source registration registration.json, and the verified previous automatic release ../arctic-owner-client-integration-20261105/closeout.json. The full library remains 9,212 catalog files and 9,237 ledger rows, with 5,640 pending qualifications and 3,597 explicit dispositions.