quota exhaustion releases the claim + 60-min backoff instead of consuming the job as failed

self-drive: server-side tick loop (gr.Timer is a FRONTEND timer — with no browser open the worker sat RUNNING but idle; every past ZeroGPU solve coincided with someone polling it). Adds tick_age_s liveness to /tick state.

honor spec cold stamp: drop resident session, re-solve from anchor (restart re-clones fork v2 2e210d2b)

Basin domains: skip pre-basin specs, n_bbox capacity, huc12s/domain into manifest (restart re-clones fork v2)

V30: EVENT_ENGINE_IDENT for provenance stamps

P1.6: resident EventSessions (crestimap/session.py from fork v2) replace chunkrun — catch-up visits from the held junction state; session-priority scan; idle/LRU eviction

warm continuation: apply time-matched init_depth_<t>.tif

queue pressure > ZERO_PRESSURE_N(4) promotes backup to active drain (claim without deferral)

claim-less shadow: never write claims unless PUBLISH=1 (a shadow claim would block the HPC + stall on-mode waits)

P1.5 ZeroGPU backup worker: chunkrun.py

P1.5 ZeroGPU backup worker: app.py

P1.5 ZeroGPU backup worker: requirements.txt

P1.5 ZeroGPU backup worker: README.md

initial commit