The protocol enforces strict rules on what counts as a valid finding and what can be reported. These rules exist to keep the output structurally useful and prevent inflation, padding, and noise. A finding that does not meet all criteria below is silently dropped — it is not included, noted, or flagged as excluded.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/XxYouDeaDPunKxX/PA-PVP-mini/llms.txt
Use this file to discover all available pages before exploring further.
Finding Validity Criteria
Every finding must satisfy all four of the following criteria. Failing any one disqualifies the finding. Observable — The problem must be directly visible inORIGINAL_ARTIFACT. If it requires inference, assumption, or reading between the lines, it does not qualify. The reviewer must be able to point to the specific content that exhibits the problem.
Specific — The finding must identify a particular element, rule, section, field, or behavior. Vague statements about general quality, completeness, or tone are not findings.
Actionable — There must be something concrete that can be changed. A finding that describes a problem with no possible structural response is not a finding.
Structurally relevant — The problem must affect structure, behavior, constraints, decision rules, flow, validation, or operational clarity. Surface issues that do not touch any of these dimensions are excluded.
Evidence Anchoring
Every finding and every evidence line must cite content fromORIGINAL_ARTIFACT specifically. Evidence drawn from PREVIOUS_ROUND_OUTPUT, MEMORY_NOTE, chat context, or reviewer assumptions is invalid. This requirement applies independently of the four validity criteria above — a finding that meets all four criteria but cites evidence from outside ORIGINAL_ARTIFACT is still dropped.
Fix Validity
A fix is valid only if it changes at least one of the following: structure, behavior, constraints, decision rules, flow, validation, or operational clarity. Pure wording fixes are invalid unless the rewording removes an ambiguity that has a structural or behavioral consequence. If two readers could interpret a phrase differently and each interpretation would produce different execution behavior, fixing that phrase is a valid structural fix. If the two readings produce the same behavior, the fix is cosmetic and is excluded. A finding with no valid fix is not a finding. The protocol does not include it.Finding Count Rules
- The reviewer identifies up to 5 structural findings per round.
- There is no floor — if fewer than 5 genuine findings exist, fewer are reported. The output is not padded.
- If more than 5 findings exist, the reviewer keeps the strongest and cuts the weakest. The cut is not disclosed.
- Do not interpret a short finding list as incomplete. Fewer findings is a valid outcome.
Section Concentration Rule
By default, no more than 2 findings may reference the same section ofORIGINAL_ARTIFACT. This prevents the reviewer from exhausting the finding budget on a single area while ignoring the rest of the artifact.
Exception: if one section contains concentrated structural failure — meaning multiple independent structural problems that each meet all validity criteria — the reviewer may report up to 4 findings for that section. When this exception is invoked, the output must include the marker SECTION OVERLOAD JUSTIFIED on that section’s findings.
Mechanics vs. Heuristics
When a rule inORIGINAL_ARTIFACT requires human judgment to apply — rather than a deterministic check — the reviewer must name that dependence as a structural weakness. Heuristic rules are weaker than mechanical rules because their results vary by interpreter. If the artifact relies on judgment calls without defining the criteria for those calls, that gap is a reportable structural finding.
No Praise Rule
The reviewer produces no praise, encouragement, narrative commentary, or positive framing. Output contains findings and fixes only. A statement that something works well is not included — absence of a finding for a section is sufficient signal that no structural problem was identified there.FAQs
What if I disagree with a finding?
What if I disagree with a finding?
Supply the updated artifact as a new
ORIGINAL_ARTIFACT in the next round alongside the prior output in PREVIOUS_ROUND_OUTPUT. The reviewer will re-examine the artifact independently, then check whether the prior finding still holds. If the problem was addressed, the finding will simply not reappear — or the prior fix will appear in confirmed_fixes_ready in the MERGE STATUS block as resolved. If the reviewer still observes the problem, it will reappear with a CONFIRMED label and updated evidence, meaning the independent review reached the same conclusion. You can also use CURRENT_ROUND to force a fresh Round 1 on the revised content.Can the reviewer produce zero findings?
Can the reviewer produce zero findings?
Yes. There is no minimum finding count. If
ORIGINAL_ARTIFACT contains no observable, specific, actionable, structurally relevant problems that meet all validity criteria, the output contains zero findings. An empty findings list is a valid result, not a failure of the review.What does 'anchored to ORIGINAL_ARTIFACT' mean?
What does 'anchored to ORIGINAL_ARTIFACT' mean?
It means every claim in a finding — both the problem statement and the evidence — must be traceable to specific content inside the
ORIGINAL_ARTIFACT block. The reviewer cannot cite chat history, general knowledge about the topic, assumptions about intent, or content from PREVIOUS_ROUND_OUTPUT or MEMORY_NOTE as evidence for a finding. If the evidence for a finding does not exist in ORIGINAL_ARTIFACT, the finding is dropped.