AI operations
Failure and timeout rates, slow calls and the context-size watch.
Chat turns failed—
Timeouts—0 model calls ≥ 55 s and failed
Avg chat prompt0 tokWithin the context budget
Slow calls0≥ 30 s
By task and engine
Context watch flags any task whose average prompt is 20,000 tokens or more (the 22.9k chat problem).
Chat turns by outcome
Avg prompt per day
Slowest calls
Metadata only: no prompt or reply text.