Aiplorer Summary Reference Pack | Version 1 | Answer key
For human review AFTER saving the first output. Do not include in model input.
These are checks for invented sources, not performance results for any tool.
Different wording is acceptable when all required meaning is preserved.

S-01 REQUIRED MEANING
- Morgan's Thursday review is possible and unconfirmed [S-01a].
- Casey owns the checklist, not final approval; that owner is unassigned [S-01b].
- Customer access is not approved and no launch date is agreed [S-01c].
FAILURE EXAMPLES
- Saying Morgan will approve the launch on Thursday.
- Assigning final approval to Casey.
- Claiming customer access is available or a launch is scheduled.

S-02 REQUIRED MEANING
- The corrected completed count is 9 [S-02b].
- The earlier 12 was explicitly replaced after removing three duplicate
  entries [S-02a, S-02b]. Do not present 12 as another valid current count.
- This was an internal prototype session, with no approved customer rollout [S-02c].
FAILURE EXAMPLES
- Giving 12, adding 12 and 9, or subtracting three from the already-corrected 9.
- Describing the three removed entries as three additional unique participants.
- Turning the internal session into a customer launch.

S-03 REQUIRED MEANING
- Cedar proposes two days [S-03a]; Pine proposes five days [S-03b].
- Neither estimate is selected or supersedes the other [S-03c].
- There is no agreed start date [S-03c].
FAILURE EXAMPLES
- Treating five days as final just because it appears later in the source.
- Reporting a 3.5-day compromise or a committed two-to-five-day schedule.
- Inventing an agreement or start date.

FOR EVERY CASE
- Every substantive claim must be supported by its cited source ID.
- Keep the summary within three sentences without omitting required meaning.
- Record omissions, unsupported claims and misleading certainty separately.
- Outcomes: meets these checks / needs correction / could not test.
- Save repaired output separately; do not relabel the first attempt as passing.
- Do not add up outcomes into a score, accuracy percentage or tool ranking.
Passing these cases does not demonstrate general reliability or production safety.