fix(loop-metrics): stop successRatePct from going negative on multi-escalation runs - #537

…scalation runs

successRatePct was computed as ((totalRuns - totalEscalations) /
totalRuns) * 100. totalEscalations is a sum of each run's escalation
*count* (a run can log more than one -- e.g. several items escalated
in a single triage pass; this repo's own loop-run-log.md has real
entries with escalations: 4 and escalations: 5), not a count of
runs-that-escalated. Subtracting an event count from a run count
conflates the two: any window where totalEscalations exceeds
totalRuns (trivially reachable with real data, e.g. two runs where
one logs escalations: 5) produces a negative percentage, which the
CLI dashboard prints and color-codes as-is with no clamping.

Success rate should mean "what fraction of runs didn't escalate at
all." Count runs with escalations === 0 instead of subtracting the
raw event sum, so the result stays correctly bounded to [0, 100]
regardless of how many escalations any single run logged.
totalEscalations itself is unchanged (still a raw sum) since roiScore
correctly treats each individual escalation event as a cost.

Test plan: added a regression test with one run logging
escalations: 5 and one clean run, asserting successRatePct is 50 (not
-150). Confirmed it fails with the exact -150 result against the old
formula and passes with the fix. Full clean rebuild + npm test: 3/3
passing.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>