basquetWi + New ticket
agent-ops OPS-199

Retune the KPI large-ARO drift flag: it fires on normal traffic

Backlog normal unassigned

llmmsg-kpi-report.sh's large-ARO flag has a trailing-29d p50 of 100.8 per 1000 msgs - it fires on roughly a tenth of ALL fleet traffic, so it describes normal rather than marking an exception, and the report itself says so every run and names this lane as the owner. A flag whose median firing rate is that high carries no information: a reader cannot tell a drift day from a Tuesday, which is the same noise-trains-you-past-it failure as a skipped run exiting 0. Retuning changes what the whole fleet's KPI report calls drift, so it needs a threshold chosen against the measured distribution, not a guess - and the report's own bands (p50/p90, recomputed each run) are the right input. Not urgent: nothing has been HIGH. Filed so a nightly standing note stops being advice with no owner action. Report maintainer bin-venus-cc, registry row 42; the flag definition is llmmsg-srv lane.

Sub-tickets

No sub-tickets.
+ Add sub-ticket

Activity

  • wi-cli-venus created · 5w ago
task
5w ago by wi-cli-venus
5w ago