pa_pvp mini is designed to find structural problems, not to guarantee that an artifact is correct. Its output is only as good as the model interpreting it and the structure of the artifact submitted. Understanding what the protocol cannot do is as important as understanding what it can do — particularly before acting on its findings or treating a clean output as a certificate of quality.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.
What pa_pvp mini does not do
- Does not prove an artifact is correct or complete. A round with zero findings means no structural findings met the bar — not that the artifact is free of problems. The reviewer may miss things.
- Does not find every problem. The protocol caps output at five structural findings per round by design. Problems that fall below the strongest five are intentionally excluded. Additional rounds can surface them.
- Does not replace domain expertise or human review. The reviewer operates on the artifact’s internal structure. It does not know whether the artifact’s goals are the right goals, whether the domain assumptions are valid, or whether the artifact is solving the right problem.
- Does not enforce its own output. Findings are candidates, not mandates. Whether a finding is acted on, deferred, or rejected is a human decision. The protocol produces review items that can be checked, challenged, merged, or applied — it does not apply them.
- Does not have memory between sessions. Each new chat session starts with no knowledge of prior runs unless that knowledge is explicitly provided through
MEMORY_NOTEorPREVIOUS_ROUND_OUTPUT. Without those blocks, each round is independent.
Model dependence
The quality of the review depends directly on the model running it. pa_pvp mini is a behavioral contract — a set of rules the model accepts. It is not a mechanism that enforces compliance. More capable models produce more precise findings with better evidence anchoring, tighter severity judgments, and more useful fix language. Less capable models may technically follow the output format while producing findings that are shallow, evidence lines that are vague, or fixes that do not change anything structural. This means that a pa_pvp mini output from a weaker model is not equivalent to one from a stronger model, even if the format is identical. If you are using the protocol for high-stakes review, run it on the most capable model available and consider running multiple rounds across different models for additional coverage.Artifact quality effects
A poorly structured artifact produces poor findings. If the artifact you submit has no internal logic, steps, constraints, or rules, the reviewer has nothing to anchor findings to. The output may be technically valid — zero findings, or low-severity findings about missing structure — but it will not be useful. Before submitting to pa_pvp mini, consider the following:- Break your artifact into explicit sections or steps. A wall of prose gives the reviewer no structure to evaluate. Numbered steps, labeled sections, and named constraints are all reviewable. Undifferentiated paragraphs are not.
- Add labels to constraints and decision rules. If the artifact has a rule, name it. If it has a constraint, state it explicitly. Named elements give the reviewer something to target.
- Make expected behavior explicit. If the artifact is supposed to do something under specific conditions, say so. Implicit expectations cannot be found as gaps — the reviewer can only find gaps in what is stated.
The 5-finding cap
pa_pvp mini caps findings at five per round. This is by design, not a limitation of the protocol. The goal of each round is the strongest findings, not the most. A round with three high-quality HIGH findings is more useful than a round with five findings diluted by marginal LOW ones. The cap forces the reviewer to prioritize. If you want to surface lower-priority issues, run additional rounds. Each round can build on the previous one usingPREVIOUS_ROUND_OUTPUT, and issues that were cut in earlier rounds can be picked up when stronger issues have been addressed.
License
pa_pvp mini is licensed under the Creative Commons Attribution-ShareAlike 4.0 International License. SPDX identifier:CC-BY-SA-4.0
You are free to use, share, and adapt the protocol. Attribution is required. Derivatives must be shared under the same license.
For a conceptual overview of how pa_pvp mini works and when to use it, see the Introduction.