orbit nodes are the agents. a pulsing ring means it's live and read from its source just now; a dashed outline means the console could not read it: not wired, or the read failed.
i
the instrument cards report live numbers. tap the i on any card for what it actually means and where it comes from.
▦
what needs you is the queue of calls only you can make: inbox instructions and open findings. rows glow gold when something's changed since your last visit.
⚄
the three gates (sc · pd · so) are the only three places you show up. everything between them is the fleet's problem.
shippers./ oracle
seen before it's asked
oracle.
an operating console, not a status report. every agent visible, every number sourced, you only show up where it matters.
01 · the agents, in orbit
every agent heartbeats or gets pulled. live nodes pulse and carry traffic to the core. a flat line isn't resting, it's not finished.
02 · the instruments
computed, never declared. silence on any one is the failure.
/ instrument 01
ci
4/4 green
/ instrument 02
fixed
100%
03 · the doctrine, moving
watch the job travel. it holds at amber, the operator clears it, it moves on. everything between gates is the fleet's problem.
04 · what needs you
not fleet state, the queue of things only you can call. empty is the goal, not the default.
inboxRe: Oracle - the console comes to you (today's changelog)you
routed: Message is informational/config note about cron and token rotation, not a direct engine command.
inboxRe: Oracle - the console comes to you (today's changelog)you
routed: Rotating a token and referencing a prior 'run the funnel' request, not a direct engine command now.
inboxRe: Oracle - the console comes to you (today's changelog)you
routed: Ambiguous 'run the funnel' request, not a recognized fleet action like sentinel/engine control.
an inbox instruction reads you when it needs a human call, not a guess. a finding reads high or critical when sentinel could not heal it alone. quiet equals not finished.
05 · the actions
what oracle actually did, not fleet state.
22h agourgent alert sent - checks failing on Driivasent
4d agoweekly digest sent (81 runs rolled up)sent
11d agoweekly digest sent (42 runs rolled up)sent
18d agoweekly digest sent (69 runs rolled up)sent
61d ago"test reply" — test reply, cleared in the 13 sep tt auditdone
61d ago"Re: test reply" — test reply, cleared in the 13 sep tt auditdone
100d ago"Yo" — Engine live sending is ON, last set by shippers-restore-false-positive-2026-07-07.done
115d ago"Re: Oracle - the console comes to you (today's changelog)" — routed: Message is informational/config note about cron and token rotation, not a direct engine command.needs you
waiting on you. close it here once it is dealt with.
115d ago"Re: Oracle - the console comes to you (today's changelog)" — routed: Rotating a token and referencing a prior 'run the funnel' request, not a direct engine command now.needs you
waiting on you. close it here once it is dealt with.
115d ago"Re: Oracle - the console comes to you (today's changelog)" — routed: Ambiguous 'run the funnel' request, not a recognized fleet action like sentinel/engine control.needs you
waiting on you. close it here once it is dealt with.
06 · the watch
every reading below is timestamped at source. silence is the failure state, so an instrument that cannot prove when it last spoke is not trusted, it is flagged.
ci
4/4 green
live
4/4 repos green on latest run
fixed
100%
live
329 of 329 findings resolved (all-time)
customs
beating · 6/6 agents
live
supabase heartbeats, live · checked 13/11/2026 15:21
all 6 agents beating. Nothing to do; the watch runs itself at 09:00, 14:00 and 20:00 if you want it sooner.
sentinel
watching · clean
live
last run 13/11/2026 12:00, ok=true
clean board. The deep sweep runs Mondays at 05:00; run it now if you have just shipped something structural.
outreach
sending · active
live
last send 13/11/2026 10:23 - no Mon-Fri send window missed since
ava
answering · last call 44d ago
live
+442045727044 -> agent, 10+ calls, last 13s done success · webhook portal.strydeos.com up · elevenlabs, live · checked 13/11/2026 15:22
the line answers, but nothing has called it in 43 days. That is a commercial question, not a technical one - the plumbing is proven.
ava 100% of gate 95% (47x1, 47 calls, 0 P0) 9d ago · trademind 93% of gate 95% (20x2, 40 calls, 2 P0) 18d ago · agent_activity, live · newest run 05/11/2026 00:09
trademind last failed its gate and nothing has run since. Re-run before the next real caller: doppler run --project trademind --config dev -- npm run warroom -- --limit 20
12/11/2026: 2 real calls (0 hang-ups) · 0 reviewed · 0 findings, 2 learnable · agent_activity, written 13/11/2026 15:19
2 real calls with caller speech got no review. The review beat runs inside the ElevenLabs webhook; check Sentry for ava_call_review.
2/2 instruments live · 5/5 orbit nodes livechecked 13/11/2026 15:22
07 · what the fleet has learned
every fleet keeps a ledger of the mistakes it has already paid for. a quiet ledger is not a clean fleet, it is usually a fleet that stopped writing things down.
ava hardening
19 rules · learned today
live
67 log entries, last 13/11/2026 · newest: live call review agent, PRs 1 to 5 (brief 09)
3 log entries, last 08/09/2026 · newest: scripts/sync-agent-prompt.ts absent from main, no delete commit
nothing written down for 66 days. Either this fleet has run clean all week, or its outcomes are not reaching the ledger - the second is far more common. Add the rule from the last thing that surprised you.