Deduplicate callbacks without accepting contradictory check results
A provider retries its callback and later sends a changed result for the same attempt ID.
- Focused work estimate
- 3h 30m + prerequisites
- Priority in the scenario
- High
- Engineering practice
- Deduplication · Integrity
Estimated field mix
- Distributed systems60%
- Integrations40%
Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence.
Review it, then add it to your workspace.
The board opens an editable draft; nothing is saved until you confirm it. Sign-in and workspace permissions apply, and Demo boards remain ephemeral.
Project context
A fictional vendor portal runs format, policy and duplicate checks in parallel. A slow checker blocks completion, while duplicate callbacks can produce contradictory final summaries.
Setup prerequisites
- Create synthetic document identities and three fake check providers.
- No real risk scoring, hiring decisions or candidate execution is included.
Preceding work
Complete these dependencies, or supply their agreed outputs before taking this ticket.
- ABATCH-101 · Freeze the required check set when a document review run starts
- ABATCH-102 · Define pass, fail, unavailable and not-applicable check outcomes
- ABATCH-103 · Bound parallel check fan-out per organization and run
- ABATCH-104 · Dispatch each check through a durable deterministic attempt identity
Acceptance criteria
- Identical callback retries resolve one recorded result.
- Different terminal content for the same attempt creates a conflict.
- Validate run, input and checker identities before recording.
Implementation constraints
- Use canonical result hashes and authenticated fixture callbacks.
Verification to include
- Deliver the same terminal callback twice.
- Alter a terminal result or input hash and keep the accepted result plus a visible conflict.
Deliverables
- Callback ingestion guard
Rollout and recovery
Canary callback handling; quarantine contradictory provider output without rewriting terminal facts.
Value of the work
For the engineer: Practice fan-out/fan-in, deadline policies and immutable aggregation.
For the team: Review trustworthy partial-result handling and controllable parallel work.
Evidence boundaries
Outcome Evidence: Tests, patches, and runbooks are requested deliverables. They become Outcome Evidence only through a qualified Mission and immutable Evidence IDs.
Ownership Evidence: Independent adaptation must be observed under a declared verification policy and cite immutable Evidence IDs. Completing a planning ticket establishes no Ownership Evidence.