Skip section navigation

Relay / fictional worked example

Assumptions, ambiguities and decisions

This record shows where the team must avoid silent interpretation. Decisions marked “provisional” would be reviewed when real product evidence arrives.

IDQuestion or conflictRisk if guessedDisposition / next step
Q1What is a duplicate request?Discard a valid new issue or create repeated recordsDecided for intake: identical (source, external_id) and payload = replay; same key with changed payload = explicit conflict; same text with a new key = distinct request
Q2Is urgency classification required?Build and maintain irrelevant featureUnresolved product hypothesis: optional experiment; not a foundation invariant
Q3Who may assign another user?Unauthorised accessRequires security design: identity and tenant permission before public release; demo avoids external access
Q4Can a resolved request be reopened?Hidden workflow complexityDeferred: not required for first slice; no assumption in code
Q5Is notifying an owner synchronous?Intake failure if external message service is downDesign decision: notification side effect should not be required to commit intake; detailed delivery semantics before release
Q6Is there a global request identifier?Collisions or cross-site exposureDecided for demo: opaque ID scoped to store; production requires tenant isolation and durable uniqueness
Q7What does “real time” mean in the queue?Unnecessary infrastructureNeeds user experiment: define acceptable freshness after workflow trial

A sample stop rule

If a provider capability appears inconsistent with the agreed contract, the agent must report:

  1. the exact expected contract;
  2. observed evidence contradicting it (source, version, failing check);
  3. the smallest available alternatives;
  4. the effect of each alternative on architecture and product scope.

The agent may run bounded checks to establish the facts. It may not invent a replacement subsystem or change a public contract without an explicit decision.

The principle in practice

Questions are not merely a form completed before implementation. New ambiguity will arise as the product evolves. It should have somewhere durable to go, a person responsible for the decision and a mechanism for influencing the next phase.

Consulting

Need help with the work?

We can review a design, help set up an AI-assisted workflow or work with your team on implementation.