TEMPLATE 1
Inspect [TARGET AND VERSION] only for [FEATURE].
Available inputs: [FILES, PACKAGE OR BROWSER OBSERVATIONS].
Describe the inputs, state changes and outputs for this feature.
List missing evidence and unresolved questions.
Do not implement yet. Exclude [OUT-OF-SCOPE FEATURES].
Finish when you can identify the relevant path and a test
that would confirm or reject the explanation.


TEMPLATE 2
For each material claim, return:
claim ID | claim | observed/inferred/unknown | exact reference
| what the evidence establishes | next verification action.

Check the referenced location and its surrounding conditions.
Do not convert a configured value into observed runtime behavior.
If evidence is missing, say what is missing rather than filling
it with a likely implementation.


TEMPLATE 3
Using the evidence ledger, write a build brief for [FEATURE].
Include the user flow, inputs, outputs, state changes, edge cases,
dependencies and acceptance criteria.
Attach claim IDs to requirements derived from evidence.
List new product decisions separately from recovered behavior.
Do not invent unavailable backend or platform integrations.
Do not write implementation code yet.


TEMPLATE 4
For each acceptance criterion, provide:
setup | action | expected result | failure condition.
Include the normal path and one relevant edge case.
Mark every test as proposed until it has been executed.
Do not claim a passing result from source inspection alone.


TEMPLATE 5
Criterion [ID] failed.
Expected: [EXPECTED RESULT].
Observed: [ACTUAL RESULT AND SAVED EVIDENCE].
Reinspect [REFERENCE] and identify the unsupported assumption.
Propose the smallest correction to the brief.
List affected tests to rerun; do not expand the feature scope.
