[n30nex]

Combine current upstream packet-hash search with Atlas observer views/hardening and Topology on dev1. The page retains all paths, packet glow and motion, Public chatter and trace-quality UI. No collected repeater telemetry is added to dev1.

Topology now reuses mesh/node layers when live trails change, camera coefficients and text measurements. It draws directly while the camera moves, preserves fractional-DPR backing pixels, and releases caches on disposal. Recent traffic expires incrementally, with a clock-regression fallback. Animation timing, the 512-flow cap and every existing node/link limit remain unchanged. No new runtime dependency or backend/RF change.

Deployed on dev1.meshcore.ca as 5345519 / v2.1.0-dev1.6, pinned to the verified Actions image, with backend #230 unchanged. The plain dev1 branch remains the upstream/Atlas baseline at add7acb; this combined candidate stays draft for operator review. The matching preview build is v2.3.3-n30nex.1 / 51679d0.

Validation: CI passed build/lint and 1,411 tests in 159 files. Preview CI passed 1,427 tests, and its exact source passed native Pi build/lint/28 Topology tests. All 27 dev1 / 34 preview assets and preview source match. Both public WebSocket subscriptions received real packet observations; dev1's 24h topology endpoint returned 7,966 links. Backends, configuration, history and unrelated services were preserved, with guarded frontend recovery available.

A 5,000-node / 10,200-link CPU/Skia benchmark submitted 78% fewer curve commands during trail refreshes, with the same particle draws and retained traffic counts. Six raster comparisons differed by at most 2/255 per channel; rotation was identical. Timings on the shared host are synthetic CPU measurements, not browser/GPU FPS. Two extra canvas buffers retain existing pixel-ratio limits and are released on closure. Browser automation was unavailable, so no fresh live-device visual/FPS pass is claimed. Production was not operated on.