---
type: reference
title: "Pokémon TCG Metanoia Spec 01 — Channel Protocol and Cognitive Swarm"
description: "Finite-state research protocol for generation, debate, ranking, evolution, meta-review, delegation and HALT."
tags: [pokemon-tcg, metanoia, channel-protocol, orchestration, provenance, halt]
timestamp: "2026-08-15T03:05:00-03:00"
---

# Pokémon TCG Metanoia Spec 01 — Channel Protocol and Cognitive Swarm

## State machine

Spec 01 defines a staged research loop:

```mermaid
stateDiagram-v2
    [*] --> INIT
    INIT --> GENERATE
    GENERATE --> DEBATE
    DEBATE --> RANK
    RANK --> EVOLVE
    EVOLVE --> META_REVIEW
    META_REVIEW --> RESPOND
    META_REVIEW --> DELEGATE
    META_REVIEW --> HALT
    RESPOND --> [*]
    DELEGATE --> [*]
    HALT --> [*]
```

The states are a coordination protocol, not a claim that one model internally executes a human-like deliberation. Each transition should have an artifact, a decision rule and a provenance pointer.

## Anti-pollution boundary

The protocol's anti-pollution idea is directly relevant to this project: hypotheses, evidence, criticism and final synthesis should not be allowed to silently collapse into one undifferentiated context. For the Pokémon pipeline, the same principle maps to:

- replay-derived observations;
- engine-generated oracle labels;
- model-generated synthetic vehicles;
- validation metrics;
- tournament outcomes;
- user interpretations and future blueprints.

The Stage 3 incident demonstrates why this matters technically. If training-only signals, sparse auxiliary targets and validation summaries share an unclear contract, an apparently coherent run can conceal objective interference.

## Ranking and evolution

The protocol proposes ranking candidate explanations or hypotheses before evolution. In this project, that should mean comparing evidence quality and reproducibility, not using an Elo label as a universal truth score. A candidate architecture should carry its source commit, test result, data cohort and unresolved assumptions.

## HALT and delegation

HALT is a valid result when evidence is insufficient, a destructive action would be required, or the next phase is outside the authorized scope. Delegation is valid when another bounded source or human decision is required. Neither state is failure; both preserve epistemic control.

## Primary source

`docs/metanoia/01_channel_protocol_and_cognitive_swarm.md`

## Cross-references

- [[pokemon_tcg_metanoia_suite]] — suite index.
- [[pokemon_tcg_stage3_training_failure_postmortem]] — anti-pollution relevance to training/validation.
- [[pokemon_tcg_docs_corpus_provenance]] — source-role separation.