Prove the outcome, not the conversation.

Three layers of evidence

Golden financial scenarios

Independent expected amounts and effects for partial payments, paid invoices, currency mismatch, duplicate allocation, stale approval and revoked roles.

Real application services

Exercise PostgreSQL, permissions, financial transactions and audit records. Verify denials and concurrency, not only a successful HTTP response.

Provider test environments

Create namespaced Tallylane QA resources in authorised Stripe test mode and a Xero demo company. Keep a manifest and reuse resources; never seed live accounts.

Planned event delivery contract

The implemented Stripe TEST inbox verifies exact raw bytes and durably retains verified event references before acknowledgement. Its finite processor independently reads checkout and payment intent, deduplicates financial effects and retains ambiguous outcomes for recovery or manual reconciliation. No real provider delivery is implied by local fixtures.

Outbound partner webhooks are planned. Event schemas, signature headers, retry schedules and replay controls are not published contracts yet. Their release requires durable outbox delivery, scoped subscriptions and recovery tests.

Recover without losing the record

Respect provider Retry-After responses, pace test data creation and retain checkpoints. A timeout cannot imply failure or justify a duplicate write. Payment confirmation and Xero accounting-sync status must remain separate.

A feature is ready when its evidence is ready

Each API and web capability is mapped to a persona, positive scenario, negative scenario and executable test in the repository QA registry. Release also requires mobile reflow, readable text, keyboard operation, contrast, CI and live deployment checks.

For agents, assert the tools selected, records accessed, permission decision and durable effects. Plausible prose and a passing build do not prove financial correctness.