Create a source matrix — GPT 6 Prompt Template

Complete prompt

Task
Organize research claims and the sources that support them.

Instructions
Build a source matrix for [QUESTION]. Use only the supplied sources: [SOURCES]. For each claim, record source, publication date, exact supporting passage in your own words, confidence, contradictions, and what remains unknown. Distinguish official fact, reported observation, inference, and anecdote.

Expected output
Claim matrix followed by a short evidence-bounded answer.

Checks
- Every claim has a source or is labeled inference.
- Dates are recorded.
- Conflicts are not silently resolved.

Expected output and checks

Expected output

Claim matrix followed by a short evidence-bounded answer.

Checks

Every claim has a source or is labeled inference.
Dates are recorded.
Conflicts are not silently resolved.

Worked example

Create a source matrix: recorded example

Codex · GPT-6 Astra · medium reasoning · 2026-09-07

Original input

Task
Organize research claims and the sources that support them.

Instructions
Build a source matrix for When did registration open, who can attend, and is evening demand established?. Use only the supplied sources: All sources below are explicitly fictional training materials, not claims about real organizations.
S1 simulated project announcement,2026-09-01: Paper Lantern registration opens12 September. The workshop accommodates24 people. Availability outside the local community has not been decided.
S2 organizer field observation,2026-09-03: I saw the registration form accepting a submission on11 September. I did not check whether this was a rehearsal.
S3 participant interview,2026-09-04: I prefer an evening session; one friend also asked about evening availability. I do not know the actual timetable.. For each claim, record source, publication date, exact supporting passage in your own words, confidence, contradictions, and what remains unknown. Distinguish official fact, reported observation, inference, and anecdote.

Expected output
Claim matrix followed by a short evidence-bounded answer.

Checks
- Every claim has a source or is labeled inference.
- Dates are recorded.
- Conflicts are not silently resolved.

What the checks found

Every claim is traced to S1, S2 or S3; sources are clearly fictional and their dates/types remain distinct.
The response catches an additional chronology problem: S2 reports a September 11 observation despite a September 3 publication date. It does not silently resolve the conflict.

Revisions after the first output

No factual correction. The synthetic date conflict remains part of the example.

Original output

All evidence below is fictional training material. Confidence describes what each source supports, not independent verification of a real event. Supporting passages are precise paraphrases, not quotations.

Claim	Type	Source and publication date	Supporting passage, paraphrased	Confidence	Contradictions or tension	Unknown
Registration was announced to open on 12 September.	Official announcement within the fictional scenario	S1, simulated project announcement, 2026-09-01	Paper Lantern's stated registration opening is 12 September.	High for the announced date; insufficient for actual opening	S2 reports a submission accepted on 11 September.	Whether the announced date was changed; actual public opening time; year/time zone are not explicit in the passage.
A form reportedly accepted a submission on 11 September.	Reported observation	S2, organizer field observation, 2026-09-03	The observer says they saw a submission accepted on 11 September and did not check if it was a rehearsal.	Limited: unverified observation with a chronology problem	The claimed observation date is later than S2's publication date if both refer to September 2026; also earlier than S1's announced opening.	Correct publication/event date, year, whether rehearsal, and whether acceptance meant public registration was open.
Workshop capacity is 24 people.	Official announced fact within the scenario	S1, 2026-09-01	The workshop can accommodate 24 participants.	High for announced capacity	No contradiction supplied	Remaining places and actual attendance; capacity does not establish eligibility.
Attendance availability beyond the local community was undecided.	Official status within the scenario	S1, 2026-09-01	Whether people outside the local community may participate has not been settled.	High for status on the announcement date	No later eligibility decision supplied	Final eligibility, definition of the local community, and other admission conditions; the passage does not guarantee admission to every local resident.
One interviewee prefers evenings and reports one friend asking about evening availability.	Anecdote; friend's interest is secondhand	S3, participant interview, 2026-09-04	The speaker would prefer an evening session, knows of a friend's inquiry, and does not know the timetable.	High that the preference was reported; low as evidence of broad demand	No conflicting preference supplied, but no corroboration or sample frame either	Friend's actual preference, wider demand, willingness to attend and actual timetable.
Evening demand is not established by these materials.	Evidence-bounded inference	S3, 2026-09-04; no demand study in S1 or S2	One preference and one reported inquiry cannot establish population-level demand.	High for insufficiency of the supplied evidence	Does not contradict the existence of some evening interest	A broader, defined sample and attendance intent by time slot are needed.

Registration was announced for 12 September, but its actual opening cannot be established: S2's 11 September report may involve rehearsal and has an unresolved publication/event chronology. Eligibility outside the local community was undecided, and no complete eligibility policy is supplied. Evening interest appears in one interview, but evening demand is not established. None of these gaps should be silently resolved.

Files from this run