Skip to main content

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.

PA-PVP Mini produces a list of findings with fixes — but not every finding is equally ready to act on. Applying fixes automatically without judgment defeats the purpose of the review. The right approach is to use the protocol’s own output signals — labels, severity, and MERGE STATUS — as a triage guide before touching anything.

​
The core principle

Do not apply every finding automatically. A finding is worth acting on when it has clear evidence anchored to the artifact and points to a structural change — a change in flow, constraints, decision rules, validation, behavior, or operational clarity. Findings that address only wording are only worth acting on if the wording creates real ambiguity in how the artifact works.

​
What makes a finding worth acting on

A finding has earned action when it meets most of the following criteria:

Clear evidence

The finding’s evidence line is anchored directly to the ORIGINAL_ARTIFACT — not to the prior review, not to a general pattern, and not to something outside the artifact.

Confirmed across rounds

The finding carries a CONFIRMED label, meaning a second independent AI reviewed the artifact and reached the same conclusion without being influenced by the prior finding.

Structural relevance

The fix changes at least one of: structure, behavior, constraints, decision rules, flow, validation, or operational clarity. A fix that changes only surface wording without affecting meaning is not structural.

Not challenged

The finding has not been labeled CHALLENGED by a subsequent round. A challenge means a second AI found the finding invalid, overstated, duplicated, or not anchored to the artifact.

​
What to skip or defer

A finding that appears in Round 1 and has not been reviewed by a second pass is an unconfirmed signal. It may be valid — but it has not been pressure-tested. Defer HIGH-severity single-round findings to investigation; deprioritize MED and LOW until a second round confirms them.
If a subsequent round labels a prior finding as CHALLENGED, a second AI examined the artifact independently and found the finding invalid, overstated, duplicated, or weak. Do not apply the fix until you have investigated the disagreement yourself. The challenged_items field in MERGE STATUS will explain what the challenge is and whether a replacement fix was offered.
PA-PVP Mini excludes pure wording fixes from its findings unless the wording change removes ambiguity that affects how the artifact actually works. If a finding slips through that only proposes a rephrasing without any structural consequence, skip it.

​
Using MERGE STATUS as a triage guide

Round 2+ outputs include a MERGE STATUS block. Use it as a prioritized action list before reading individual findings:

confirmed_fixes_ready → High priority

These fixes survived an independent re-review of the artifact. They have been validated by two separate passes and are the strongest candidates for immediate action.

challenged_items → Investigate before acting

A prior finding or fix was disputed. Read both positions, check the artifact yourself, and decide whether the original finding, the challenge, or neither is correct.

new_fixes → One more round of validation

Fresh findings from the current round that were not in any prior round. They are unconfirmed. Run another round or review them manually before applying.

conflicts → Human judgment required

Direct contradictions between the current independent review and the prior audit position. These are declared explicitly and cannot be resolved by the protocol — they require you to read both positions and make a call.

​
Severity and urgency

Severity levels give you a second axis for prioritization independent of confirmation status:
SeverityMeaningAction
HIGHBreaks core function, correctness, execution viability, or downstream trustReview immediately, even if unconfirmed. A HIGH finding that breaks core function is worth investigating before the next round.
MEDWeakens reliability, robustness, auditability, or prioritization without core collapsePrioritize confirmed MED findings; defer unconfirmed ones to the next round.
LOWReal structural weakness with limited blast radiusAct on confirmed LOW findings when you have bandwidth; skip unconfirmed LOW findings until further rounds.
A finding that appears only once in a single round is a weaker signal than one that has survived a cross-AI challenge. Treat single-round findings — especially at MED and LOW severity — as hypotheses to be tested in a subsequent round, not conclusions to act on immediately.
Use round progression as a natural triage mechanism. A fix that was INDEPENDENT in Round 1 and CONFIRMED in Round 2 by a different AI has passed two separate reviews with no shared memory. That convergence is the strongest confidence signal the protocol produces — stronger than any single finding regardless of its severity label.

Build docs developers (and LLMs) love

Get started for freeTalk to us