Watch outcomes, not just uptime
Errors and latency were fine. The signal was in a product metric — abandoned searches — broken down by region and hour.
← The ADLC loop
Observe: production talks back.
Online chess is live. A monitor agent watches telemetry, traces and player feedback — not just for outages, but for signals. When it finds one, it doesn't file a vague ticket: it drafts a new intent with evidence, and the loop starts again.
Quick-match · ap-south · by hour of day Collecting
off-peak
0006121823 h
searches abandoned (%)
monitor-agent · investigation Watching
intents/INT-002-offpeak-matchmaking.md Not started
Player feedback · matched by monitor-agent
The loop
← From Release · 1.4.0 live
Not every signal is an outage — this one is a product insight.
Intent · INT-002 →
What good observation looks like
Explain, don't just alert
The monitor agent traced the signal back to a specific spec rule (FR-2) and the data that makes it fail off-peak, then matched it to real player feedback.
Hand back an intent, not a ticket
INT-002 arrives with a problem, evidence, a hypothesis, an experiment and a guardrail — ready for the PM and engineer to frame. The loop starts again.