Handle Upstream Failures in a Payment-Calculation Service

A payment-calculation service depends on upstream delivery events and a time-dependent peak-period configuration. What should it do when an upstream dependency fails or required information arrives late? ### Constraints & Assumptions The reported coding exercise involved delivery-worker payment and a peak-period follow-up, but its event schema, rate formula, and peak-window rules were not supplied. This question retains the explicit upstream-failure discussion. Treat the calculation formula as a versioned business rule supplied separately; do not invent a payment amount or silently substitute missing data with zero. ### Clarifying Questions Which inputs are authoritative? Is the unavailable operation a read or a possibly completed write? Can the service return pending status? What freshness, reconciliation, and correction requirements apply? ### What a Strong Answer Covers Distinguish incomplete data from a valid zero, use bounded retries and idempotency, preserve enough evidence for replay, and reconcile late inputs without duplicate financial effects. ### Follow-up Questions When is a cached peak-period configuration acceptable? What if a timeout occurs after an external service may already have accepted a write? How do you correct a previously finalized calculation after a late event?