A good debug reply lists guesses in rank order, not one answer. Each guess ships with the evidence that would confirm it and the observation that would rule it out.
Confident wording is not evidence. Rank is an investigation order. Check quotations against the original log, revision, and observation conventions; existing evidence may settle a factual question without a fresh run.
The request names the shape of the answer. Vague asks return vague causes, while a shaped ask returns testable pairs.
Request element | Example wording | Why it is there |
|---|---|---|
Ranked list | List three hypotheses in rank order | Forces comparison instead of one premature answer. |
Evidence each | Cite relevant source lines and distinguish missing observations | Makes each claim inspectable. |
Rule-out each | State what observation would weaken or rule each out | Separates competing explanations. |
Next check | Name a source inspection or focused experiment | Bounds the next step before any edit. |
Use the same request ID, time window, and configuration when comparing explanations. A consumer stall for an unrelated transaction does not support a claim about this request. Ask which observation distinguishes a missing acceptance from a genuinely blocked consumer.
The block below is a schematic transcript, illustrative only and not a reply from any real system. It shows the shaped ask and the shaped answer side by side.
schematic transcript (illustrative; not a reply from any real system):
engineer: <slice: first TIMEOUT plus 20 lines before, 10 after> + "list ranked hypotheses; for each, state the evidence that would confirm it and what would rule it out"
assistant (schematic):
H1 consumer blocked this accepted request: inspect matching stall/completion
events and any documented shared-resource dependency.
H2 checker deadline conflicts with intended behavior: compare configured
deadline and units with the approved latency requirement.
H3 presentation was mistaken for acceptance: inspect the acceptance monitor.
A matching completion weakens H1; a matching acceptance weakens H3.
Missing lines remain missing evidence, not proof an event never occurred.SYNTHETIC follow-up evidence, same R42 artifact:
cycle 100 monitor: accepted req 4821
cycle 101 consumer: stalled req 9000
cycle 105 monitor: completed req 4821
cycle 110 checker: TIMEOUT req 4821
Exercise: does the stall line establish that req 4821 was blocked?
Expected: no; it names a different request. Completion of 4821 suggests
investigating checker matching, timestamp semantics, and observation paths.
It does not alone prove the checker is the root cause.Verify the quotes in the original artifact, matching run, request, units, and surrounding events. They may establish a fact or rule out a claim. Decide whether they support the causal explanation; a fresh experiment is needed only if relevant uncertainty remains.
Ask for matching events for 4821 and any documented shared-resource dependency. A stall on another request alone does not establish the timeout cause.