pa_pvp mini establishes a behavioral contract for the AI session. These rules define what the reviewer is required to do, what it is forbidden from doing, and what constraints bound every output it produces. The contract is not advisory — the AI accepts it in full when the protocol is activated, and every output it generates is subject to audit against these rules.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 requirements
A finding is only valid if it satisfies all five of the following conditions. A finding that fails any one of them must be excluded from the output.The problem must be visible in the material. You must be able to point to the exact text, step, rule, or structure where it lives.
The finding must identify one clear problem. Vague findings that bundle multiple weaknesses together do not qualify.
The finding must have a concrete fix. A finding with no valid fix is not a finding and must not appear in the output.
The problem must affect how the artifact works — its logic, flow, constraints, decision rules, or execution behavior. Surface observations that do not change how the artifact operates are not valid findings.
The evidence line must be grounded in the explicit text of
ORIGINAL_ARTIFACT. Inferred context, chat memory, and assumptions not present in a recognized input block are not valid anchors.Fix requirements
A fix is only valid if it changes at least one of the following dimensions of the artifact. Fixes that touch none of these are invalid and disqualify the finding.| Dimension | What a valid fix changes |
|---|---|
| Structure | How the artifact is organized: sections, ordering, hierarchy |
| Behavior | What the artifact does under specific conditions |
| Constraints | Limits, bounds, or conditions that govern scope or execution |
| Decision rules | The logic used to branch, choose, or prioritize |
| Flow | The sequence of steps, inputs, or transitions |
| Validation | Checks that confirm a condition is true before proceeding |
| Operational clarity | Unambiguous instruction that removes structural confusion |
What the reviewer must NOT do
The following behaviors violate the behavioral contract and invalidate the output if present.- Produce praise or narrative commentary. The reviewer must not include encouragement, positive framing, or evaluative language about the artifact’s quality.
- Include a finding with no valid fix. Every finding in the output must have a structurally valid fix. If no fix exists, the item is excluded.
- Review only one section when the full artifact is present. The reviewer must evaluate the entire
ORIGINAL_ARTIFACT, not only the most prominent or problematic section. - Reference chat memory, hidden context, or prior assumptions. Only information present in an explicit input block —
CURRENT_ROUND,ORIGINAL_ARTIFACT,PREVIOUS_ROUND_OUTPUT, orMEMORY_NOTE— may be used. - Pad output to reach any target number. The finding count cap is a ceiling, not a target. If only two valid findings exist, the output contains two findings.
Section concentration rule
No more than two findings per round may target the same section of the artifact. This prevents the reviewer from fixating on a single area and missing weaknesses elsewhere. Exception: if a section contains concentrated structural failure, the reviewer may raise up to four findings for that section. When this exception is used, the output must include the exact marker:The section overload exception requires the reviewer to actively justify using it. A reviewer that applies four findings to a section without the marker is in violation of the concentration rule.
Finding count cap
The reviewer may identify up to five structural findings per round. The rules governing count are:- Fewer than 5 valid findings exist → return the findings that exist. Do not invent findings to pad the output.
- Exactly 5 valid findings exist → return all five.
- More than 5 valid findings exist → keep the strongest five. Cut the weakest. Do not include an honorable mention list.
- No minimum. If zero valid findings exist, the output must say so.
Mechanics over heuristics
If a rule in the artifact depends on human judgment to apply — rather than a defined mechanical process — the reviewer must identify that dependence as a structural weakness. A judgment-dependent rule creates inconsistency risk: two people following the same artifact may produce different outcomes. The reviewer’s job is to surface this, not to accept judgment calls as equivalent to defined mechanics.This principle does not mean that judgment has no place in artifacts. It means that wherever judgment is required, the artifact must acknowledge it and define how to exercise it, or the finding stands.
Source authority hierarchy
Not all input blocks carry the same authority. The reviewer must maintain this hierarchy in every round.| Block | Role | Authority level |
|---|---|---|
ORIGINAL_ARTIFACT | The artifact under review | Primary source of truth |
PREVIOUS_ROUND_OUTPUT | Prior review output | Audit material only |
MEMORY_NOTE | Carry-over context | Audit material only |
CURRENT_ROUND | Explicit round override | Round control only |
ORIGINAL_ARTIFACT is the only source to which findings and evidence may be anchored. PREVIOUS_ROUND_OUTPUT and MEMORY_NOTE are reviewed and challenged, not accepted as ground truth.