Forecaster
Negative-balance forecaster: reversals vs payout velocity
Every finding races two clocks: how fast you detected, and how fast the seller gets paid. This forecast bands findings as reachable, at-risk, or likely-dead from those two variables alone — explicitly modelled, since account-level reality varies — and shows why detection latency dominates every other lever.
Inputs and banding
Detection latency (days) · payout frequency · days since refund. Output bands: reachable (before next payout), at-risk (one cycle), likely-dead (multiple cycles without volume signals). Worked contrast: weekly payouts + quarterly detection → most findings dead; weekly payouts + event-driven detection → majority reachable.
Shortening clock one
Webhook-driven detection moves discovery from quarters to seconds — the only lever that improves every finding simultaneously. Everything else (netting, floors, write-offs) manages decay after it has happened.
Clock two is policy
Payout schedules are chosen, not weather. Slower schedules widen windows at seller-satisfaction cost; instant payouts collapse them entirely. Decide consciously per segment rather than inheriting defaults.
Common questions
Does this cover disputes?
Separate clock — disputes run their own lifecycle; use the dispute-exposure calculator for that shape.
Instant payouts?
Window collapses toward zero; pre-payout alerting becomes the operative strategy.