Physical Hex pixel density without changing native artwork

The isolated production renderer now has a narrow candidate that renders both Hex ground and foreground canvases at the selected physical pixel density. The current local renderer already has this behavior. Production still runs the verified 204f26b release; this candidate is not published.

The previous production-build body-clearance checks exposed a real fidelity gap: at devicePixelRatio 2, desktop Hex canvases were 958×884 and phone canvases were 374×522. The native source-only replay was exact, but composing into those logical-size surfaces softened roofs, trees and stonework on the final display. The correction increases the physical raster dimensions while retaining the same CSS/world positions and the player's existing resolution setting. No original artist image is resized, repainted or recompressed.

Ground and foreground share one physical layout. Material placement, moisture and working-land transforms use that layout once; the ground build keeps its logical viewport width/height and receives the selected DPR separately. Both foreground painting branches use the same projection. DPR participates in the Hex cache key, and loading/reset clears remove the previous transform before clearing the full canvas. The patch preserves the existing native-image Hex loader and introduces no paint readbacks or permanent image derivatives.

Thirty-nine final scoped tests pass, including the physical-layout and reset tests in the isolated production renderer, current local density checks and source-bound body-clearance checks. They exercise integer/fractional DPR, actual canvas setup, both foreground branches, bounded placement and native hash binding. These tests do not establish full browser, state-lifecycle or performance acceptance.

Forty normal Region/Hex desktop/phone views completed in a fresh disposable local world across ten naturally selected production sources: two forest cities, forest broken walls, a field village, grassland cottage, hills farm, tundra waystones, wetland strata, courtyard keep and market town. The first three repeat the previous body-clearance sites, permitting exact Region control and CSS-placement comparisons. No source, terrain or world feature is forced. Six real pointer-return/reconnect observations are repeated at those three sites.

A separate final browser phase compared the published production renderer with the combined body/density candidate at one natural forest city in desktop and phone viewports. Those pages have no source paint hooks or RGBA readbacks. They measure normal pointer movement and Hex switches; one workstation and scene cannot certify full performance across the library. Any slowdown must remain explicit and be addressed without reducing native artwork quality.

All earlier gates remain: phone control clearance, exact silhouette/final pixel attribution, terrain contact, physical scale, state lifecycle, mesh grain/nonrepetition and the historical terrain/Hex failures. The extra physical pixels do not approve a source or prove that a renderer-available asset appears in the production world. Every original, selected final, approved reference, both Citadel hamlets, canonical geography and persistent save is preserved.

Evidence: candidate hashes candidate.json, normal production checks normal-production/, preceding source-bound body candidate. Full-library scope remains 9,209 catalog files and 9,234 ledger records, with 5,637 pending qualifications.

Completed local verification

All40normal views completed with zero browser errors. All20Hex source-only replays are exact and both ground/foreground use the selected physical DPR2. The12paired body-site observations preserve CSS source placement within0.001pixel;all six preceding Region map PNG hashes are byte-exact while all six Hex maps change. All six real pointer-return/reconnect observations reproduce exact returned maps and transforms. Every pane was directly inspected.

Readback-free timing observations on one natural forest city measured Region movement p95 at19.4versus19.5ms on desktop and11.3versus12.1ms on phone for published production versus candidate. Desktop Hex switch observations were148/116/117versus148/116/99ms;phone165/100/100versus132/100/116ms. These three-switch samples and one workstation support the scoped correction;they do not certify whole-library performance. No image-quality reduction is used.

The disposable local world was retired only after server,browser and performance-context terminal checks. Compact hashes and ten decisive context sheets remain;no raw RGBA buffers or catalog mirrors were stored. Current per-asset findings retain coarse ground grain,shoreline/fence failures,phone UI,neighbor/scale and state/lifecycle gates.