SKANDER servertelegram v3.9.0 validation Date: 2026-08-29 BASELINE PASS built on servertelegram v3.8.2 cancel-then-replace source PASS v3.8.2 edit replacement behavior retained PASS v3.8.1 edit-cancel compatibility retained PASS v3.7.1 URL-safe direction parser retained PERSISTENCE PASS canonical data/settings.json added PASS data/settings.json.bak recovery added PASS settings saved durably and verified PASS stale Passenger-worker simulation cannot restore old settings PASS settings refresh across workers every 1 second PASS state.json still remains backwards-compatible runtime store PASS settingsRevision persists PASS recovery settings persist PASS fleetDefaults persist in canonical settings PERFORMANCE PASS runtime state writes use compact JSON PASS runtime writes no longer fsync/read/parse/copy backup on every save PASS hot writes are coalesced PASS Telegram polling persistence is coalesced instead of durable every cycle PASS offline cleanup saves only after an actual state change PASS dashboard HTTP state supports gzip level 1 PASS dashboard device payload uses a field whitelist PASS dashboard renders only the active view PASS device-card default render capped at 120; search still covers all devices PASS bulk control API responses no longer return thousands of full control objects PASS dashboard history payload reduced to 120 signals / 120 controls / 80 notifications TESTS PASS npm run check PASS npm test PASS dashboard JavaScript node --check PASS test_v390_persistence_speed.js SCALE FIXTURE 1000 devices + 1200 signals + 6000 controls v3.8.2 state save average in local fixture: ~29.61 ms v3.9.0 fast state save average in local fixture: ~5.95-7.87 ms Observed improvement: roughly 4x-5x in this local persistence fixture DASHBOARD FIXTURE 1000 devices, synthetic large raw device metadata Dashboard JSON response before HTTP compression: ~1,535,827 bytes Gzip wire payload observed: ~42,570 bytes Local endpoint response observed: ~0.05 seconds NOTE These are local validation measurements, not production latency guarantees. Production performance depends on cPanel/CloudLinux filesystem, CPU limits, Passenger process scheduling, network latency, and number of Telegram sources.