Reliability checklist
AI-built app reliability checklist: failure modes and exact terms.
Use this when a vibe-coded or AI-built app works in the demo path but fails under refresh, retry, two tabs, auth changes, partial saves, schema changes, background jobs, or real production usage.
Failure modes to name before asking for code
Lost progress after refresh
Use Draft persistence, Resumability, Checkpoint, and State reconciliation. Ask where unfinished work is stored, when it is saved, how stale drafts resolve, and what happens after a browser refresh or return visit.
Duplicate orders, records, emails, or charges
Use Idempotency, Idempotency key, Transaction boundary, Retry policy, Race condition, and Optimistic UI. Ask how repeated clicks, browser retries, network timeouts, and webhook retries produce one intended effect.
One failed step traps the whole flow
Use Recovery path, Fallback UI, Error taxonomy, Timeout budget, and Retry policy. Ask what the user sees, what data stays safe, what can be retried, and when support or manual repair is needed.
Back, next, edit, and submit create impossible states
Use Statechart, Transition table, Guard condition, Step state, and Invariant. Ask for allowed states, rejected transitions, side effects, and acceptance tests before component code.
Client, server, URL, and saved draft disagree
Use Source of truth, State reconciliation, Optimistic UI, Race condition, and Observability. Ask which system owns each state and how conflicts are detected, resolved, and logged.
AI agent fixes one bug and breaks another flow
Use Regression set, Eval harness, Golden tests, Rubric, and Acceptance criteria. Ask for tests that prove the important flows still work after every generated change.
Production failures AI-built apps often miss
Silent integration breakage
Use Integration contract, Health check, Observability, and Error taxonomy. Ask how each dependency is checked, versioned, monitored, and surfaced when it stops behaving as expected.
Database shape changes break old screens
Use Schema drift, Migration, Backward compatibility, Source of truth, and Regression set. Ask what old data looks like after the change and which screens prove compatibility.
Auth and permissions behave differently in production
Use Row-level security, Authorization boundary, Error taxonomy, Fallback UI, and Observability. Ask for role-specific tests and user-safe failure states.
Background jobs fail without anyone noticing
Use Health check, Observability, Retry policy, Timeout budget, Dead-letter queue, and Recovery path. Ask what gets retried, what gets parked, and what alert proves work is stuck.
Two tabs or stale sessions overwrite newer work
Use State reconciliation, Source of truth, Race condition, Optimistic UI, and Conflict resolution. Ask how stale writes are rejected or merged.
Generated code has no operational debugging trail
Use Observability, Event log, Error taxonomy, Trace, and Correlation id. Ask what evidence a developer sees when a user reports a failure.
Copyable reliability handoff
Audit request
Audit this AI-built app for statecharts, transition tables, guard conditions, invariants, draft persistence, checkpoints, idempotency, transaction boundaries, retry policies, fallback UI, recovery paths, state reconciliation, observability, health checks, integration contracts, schema drift, row-level security, regression sets, and golden tests.
Required output
Return a failure-mode table with symptom, exact term, likely cause, code area to inspect, user-safe behavior, observability signal, and acceptance test. Cover refresh, back/next, retry, double submit, two tabs, auth expiry, partial save, API timeout, integration failure, schema change, and background job failure.
Acceptance tests to demand
State and transition tests
Test every allowed transition plus rejected transitions for back, next, edit, cancel, submit, retry, resume, timeout, and expired auth.
Persistence and recovery tests
Test refresh, tab close, return visit, stale draft, failed save, partial success, retry, and support handoff.
Side-effect tests
Test double click, duplicated request, webhook retry, slow network, timeout, payment retry, email retry, and import retry.
Production contract tests
Test integration contract changes, schema drift, permissions, background job failures, and health check alerts.
Related guides
AI coding agent vocabulary for robust app workflows.
AI coding agent vocabulary for fixing vibe-coded apps: statecharts, step state, transition tables, guard conditions, draft persistence, idempotency, checkpoints, fallback UI, and recovery paths.
What you mean vs. what experts call it.
A translation map from vague AI-tool language to expert vocabulary for coding, images, UX, research, and buyer messaging.
Coding-agent terms cheat sheet.
A coding-agent vocabulary cheat sheet for state, workflow, retries, recovery, persistence, and failure handling.
State machine terms for multi-step forms and builders.
State machine and workflow reliability terms for multi-step forms: statecharts, transition tables, guard conditions, resumability, idempotency, and recovery paths.
Common questions
Is this a prompt template?
No. It is a reliability vocabulary checklist. The copyable handoff is only a way to carry the exact terms into a coding agent, spec, review, or issue.
Why does this help with AI-built apps?
AI agents often patch visible screens. Reliability terms force the work to cover state, side effects, recovery, production contracts, and regression proof.
When should I use it?
Use it before asking an agent to change a brittle app, before shipping a generated feature, or after a production failure that the original build did not anticipate.