Pokémon TCG — Documentation Corpus and Git Provenance
Complete map of the repository's 35-file docs corpus, its conceptual layers, Git introduction points and evidence status for Wikifita integration.
Pokémon TCG — Documentation Corpus and Git Provenance
Purpose
This page is the provenance index for the repository documentation, not a replacement for the documents themselves. The repository contains implementation contracts, empirical records, mathematical formalizations, future blueprints, manuscript chapters, epistemology and cross-harness handoff material. Wikifita must preserve those roles instead of flattening every document into “current architecture”.
The source hierarchy used for this integration is:
- verified live repository, Git history and local artifacts;
- inspectable project documents and their topology;
- provenance-linked Antigravity transcripts and reports;
- existing Wikifita pages as contextual and editorial records;
- user-supplied interpretations, labeled as such until independently isolated.
Corpus inventory
The docs/ tree contains 35 Markdown files at the verified develop endpoint. The inventory below is complete and keeps implementation, analysis, blueprint and philosophical material distinct. The earlier 33-file count omitted the empirical ablation monograph and the normalization heuristics document from the index prose; neither source is discarded.
Core monograph and manuscript
| File | Role | Wikifita treatment |
|---|---|---|
docs/Pokemon_TCG_AI_Monograph.md | Integrated nine-chapter technical monograph | Synthesis source; claims are reconciled against code and tournament artifacts |
docs/manuscript/01_introduction_and_teleology.md | Behavioral cloning cold start and project teleology | Historical manuscript chapter |
docs/manuscript/02_data_pipeline_and_kv_cache.md | Replay compilation, Parquet and KV cache | Data-pipeline lineage |
docs/manuscript/03_tokenizer_and_epistemology.md | Set-based tokenization and epistemic boundary | Model reference plus epistemic note |
docs/manuscript/04_engine_simulation_would_ko.md | Engine-backed would-KO oracle | Oracle/data contract |
docs/manuscript/05_action_encoder_and_pointer_heads.md | Option streams, buckets and pointer-style action scoring | Neural architecture reference |
docs/manuscript/06_scratch_registers_and_anomalies.md | Scratch registers, TBPTT and the dynamic-mean anomaly | pokemon_tcg_stage3_training_failure_postmortem |
docs/manuscript/07_empirical_results_and_elo.md | Tournament outcomes and invariant Elo | Evaluation and rating lineage |
docs/manuscript/08_curriculum_and_devops.md | Curriculum stages, optimizers and telemetry | Training operations and failure history |
docs/manuscript/09_tournament_orchestrator_and_inference.md | Tournament orchestration, SQLite and inference | Evaluation/runtime boundary |
Architecture and research blueprints
| File | Role | Status boundary |
|---|---|---|
docs/architecture/01_ropend_theory.md | N-dimensional rotary position encoding over turn, meta-epoch, time and Elo | Future architecture; not implemented |
docs/architecture/02_stochastic_elo_inference.md | Ephemeral Elo anchor and stochastic routing features | Future architecture; not implemented |
docs/architecture/moe_pipeline_blueprint.md | Magnum Opus, Pilot/Vehicle split, MoE, vehicle draft and Apex | Future blueprint; preserve in full |
docs/arena-future-architecture.md | Synchronous arena to application-boundary/worker design | Future platform architecture; explicit non-goals retained |
docs/empirical_ablation_monograph.md | 420-game matrix, deck saliency, FP16 collapse and Abelian rating interpretation | Empirical synthesis; cohorts and mechanisms remain separate |
docs/neural_engine_and_tokenization_spec.md | As-built tensor layout and target architecture comparison | Current tensor contract plus future delta |
docs/abelian_group_elo_formulation.md | Translation-invariant rating formalization | Proposed/formal model; code validation separate |
docs/pagerank_and_abelian_graph_invariance.md | PageRank and Abelian tournament-graph analogy | Research hypothesis and mathematical analogy |
Data, ETL and local platform
| File | Role | Status boundary |
|---|---|---|
docs/dataset_compilation_and_oracle_pipeline.md | Raw replay to semantic validation, oracle and Parquet | Pipeline contract and target additions |
docs/etl_architecture_and_auditing.md | Zero-trust ETL, ZIP versus SQLite, idempotency and rebuild hazards | Audit design; no rebuild authorized |
docs/database_schema.md | Current physical SQLite table and foreign-key map | Current snapshot; integrity anomalies preserved |
docs/schema-evolution.md | Relational v2 debt and target normalized schema | Design under review; not implemented |
docs/implementation-spec.md | Concrete target services, schema, dashboard and acceptance | Implementation specification; not proof of delivery |
docs/local-overhaul-design.md | Local product contract, idempotency and validation decisions | Design baseline; not implemented |
docs/normalization_heuristics.md | L1/L2 entity resolution for ghost teams | Heuristic analysis with residual uncertainty |
Platform sampling and operational context
| File | Role | Status boundary |
|---|---|---|
docs/kaggle_platform_dynamics.md | Official-team coverage, sampling bias and leaderboard blind spots | Empirical platform analysis |
docs/kaggle_timezone.md | EpisodeId interpolation and archive-boundary timezone inference | Inference from monotonic IDs; not a hard platform fact |
docs/cross_harness_and_tokenizer_adaptation.md | Model/channel mappings, Antigravity ledger and portability | Provenance and adaptation guide |
docs/technical_handoff_rfc.md | Master index, chronology, transcripts, runs, schema and handoff state | Handoff assertions; direct claims require source reconciliation |
Metanoia suite
| File | Role | Status boundary |
|---|---|---|
docs/metanoia/01_channel_protocol_and_cognitive_swarm.md | Generate/debate/rank/evolve/meta-review state machine | Research protocol |
docs/metanoia/02_rule_provenance_and_epistemic_evolution.md | CLAUDE → GEMINI/.agents rule evolution and epistemic laws | Research history and governance |
docs/metanoia/03_model_adherence_and_failure_mode_analysis.md | Model-family adherence, leakage and failure categories | Source-reported telemetry; not independently benchmarked |
docs/metanoia/04_tensorized_scaling_and_subagent_orchestration.md | Cognitive tensor, provenance ledger and subagent topology | Conceptual architecture |
docs/metanoia/05_the_halt_protocol_and_hypersigil_epistemology.md | HALT boundary, hypersigils and tokenization interpretation | Philosophical/speculative research |
docs/metanoia/06_holographic_tokenization_and_liberatory_pedagogy.md | Holographic tokenization, pedagogy and research philosophy | Philosophical/speculative research |
Git provenance timeline
The dates below describe when the files entered or were last materially edited on the verified develop history. They are not claims about when the ideas were first imagined in conversation.
| Date | Commit | Documentation event |
|---|---|---|
| 2026-07-27 | 05c4435, babcded | Local arena, implementation specification, schema evolution and overhaul design were introduced/replaced as detailed architecture documents. |
| 2026-07-29 | 98b906f | Operational contract was refreshed after the earlier migration and training work. |
| 2026-08-11 | f90ac1e, 6dc5a0c, 4844ef9, 1545f48, c330547, fccf2dd | Curriculum sweep orchestration and successive GEMINI.md rule, memory and communication-governance revisions were added. |
| 2026-08-12 | d9c67a7 | Monograph and nine manuscript chapters were added or refreshed alongside training fixes. |
| 2026-08-12 | 86aa528 | Abelian-group Elo formulation was added during the Stage 3 tournament phase. |
| 2026-08-12 | 0dec40f, 26b1e87, 89deb7d, 9bc5021 | Trainer reduction changes, selected deck artifact, local-Elo rule changes and Stage 1–3 FP32 checkpoint archives were recorded. |
| 2026-08-13 | 459f9d6 | RoPEND/MoE blueprint entered the architecture lineage. |
| 2026-08-13 | f697ce9 | RoPEND theory, stochastic Elo inference and the blueprint were consolidated in the pre-dawn handoff. |
| 2026-08-13 | 328d4f9 | Current physical database schema was documented. |
| 2026-08-13 | eb83bef | ETL auditing, platform dynamics, timezone inference and normalization heuristics were documented. |
| 2026-08-13 | dbf298a | ETL idempotency, bulk Elo decoupling and KaTeX directive changes were recorded in source and project memory. |
| 2026-08-13 | 14ec607 | ResultsDB, card-stat build behavior and agent-rule files were updated; the commit also added the follow-me operational rule. |
| 2026-08-14 | cec2eab | Neural engine/tokenization, dataset/oracle pipeline and empirical ablation specification were added. |
| 2026-08-14 | 955149b | The master technical handoff RFC was created. |
| 2026-08-14 | c104f5b | Metanoia Specs 01–04 and the provenance-ledger material were added. |
| 2026-08-14 | 41438ec | Metanoia Spec 05 documented HALT, hypersigils and tokenization mechanics. |
| 2026-08-14 | 4321ad2 | Metanoia Spec 06 and the Wikifita integration context were added. |
| 2026-08-14 | 936d301 | PageRank and Abelian Graph Invariance was authored and linked from the RFC. |
| 2026-08-14 | f988eb0 | Blueprint, milestone, survey, scratch-probe and E2E-scaffolding artifacts were added; their gates remained pending. |
| 2026-08-14 | ea9317a | SQLite binaries and sidecars were ignored and GEMINI.md was synchronized; this changed repository hygiene, not database contents. |
| 2026-08-14 | a71318e | Compute-quota context and the sovereign handoff horizon were registered. |
| 2026-08-14 | 12595d4 | Cross-harness adaptation and the complete transcript filesystem ledger were added. |
| 2026-08-14 | 20d7d0d | AGENTS.md added the Odyssey Diary and multi-agent convergence contract; it is a handoff/instruction layer, not proof that every blueprint module shipped. |
The dense August 12–14 sequence is itself part of the project's history: the repository moved from implementation and local-platform design, through empirical training and rating analysis, into a formal handoff containing both executable state and later architectural proposals. The current verified endpoint is 20d7d0d on develop; the reconciliation anchor records 144 reachable commits and resolves the older 147-commit snapshot without erasing it.
Document-to-Wikifita integration map
The source files are represented through a graph of focused pages rather than one imported monolith. The mapping below records where each conceptual atom is now maintained:
This map is intentionally many-to-many. A source document can introduce a concept that is validated by a different artifact, and a Wikifita page can join code, Git history, database evidence and transcript provenance without replacing any of them.
Provenance roles inside the corpus
The documents have four non-interchangeable roles:
- Current implementation records: neural tensor shapes, database snapshot, ETL observations and the trainer contract. These must be checked against live code/artifacts before making “implemented” claims.
- Empirical analyses: ablations, sampling coverage, timezone interpolation, normalization heuristics and Elo/PageRank work. These preserve denominators and uncertainty rather than becoming production guarantees.
- Blueprints and specifications: RoPEND, MoE, vehicle draft, Apex, normalized schema v2 and future arena workers. These are valuable design lineage and must remain visible even when unimplemented.
- Research philosophy and governance: the Metanoia suite and manuscript teleology. These explain how the project framed evidence, orchestration and epistemic boundaries; they are not mechanistic proof about model cognition.
What was incorporated in this Wikifita pass
The incorporation layer now adds dedicated pages for the Stage 3 failure, derived-target versus vehicle-draft augmentation, the implementation contract, schema evolution, future arena boundary, entity normalization and manuscript chapter lineage, while preserving the full corpus here as a navigable map. Follow-on pages should deepen, rather than replace, the existing August 14 audit pages:
- neural engine and tokenization, including as-built versus target tensor flows;
- RoPEND, stochastic Elo, MoE, data augmentation and vehicle-draft lineage;
- dataset, oracle, ETL and audit contracts;
- manuscript synthesis and empirical result lineage;
- all six Metanoia specifications as a linked suite;
- cross-harness and Antigravity transcript provenance;
- local platform design, implementation gates, schema evolution and arena future architecture;
- entity normalization, ghost-team resolution and residual uncertainty;
- chapter-level manuscript lineage and narrative-to-evidence routing;
- sampling, timezone, Abelian Elo and PageRank analysis.
Editorial boundary
No future-stage material was deleted, shortened into a current-state claim or relabeled as implemented. The absence of a live rl/ropend, rl/moe, vehicle-draft or Apex runtime path remains an observed boundary, while the corresponding documents remain part of the project's forward lineage.
Cross-references
- pokemon_tcg_ai_battle — project hub and evidence boundary.
- pokemon_tcg_repository_timeline — earlier semantic timeline of repository phases.
- pokemon_tcg_cross_harness_provenance — transcript and Antigravity artifact ledger.
- pokemon_tcg_stage3_training_failure_postmortem — Stage 3 incident reconstruction.
- pokemon_tcg_ropend_moe_blueprint — future architecture and data-augmentation lineage.
- pokemon_tcg_metanoia_suite — linked Metanoia index.
- pokemon_tcg_current_state_reconciliation — current code, database, milestone and contradiction ledger.
- pokemon_tcg_handoff_acceptance_reconciliation — PROJECT, TEST_INFRA, AGENTS, RFC and acceptance claims reconciled against live evidence.
- pokemon_tcg_implementation_specification — target services, workflows and acceptance gates.
- pokemon_tcg_schema_evolution — current physical debt and relational v2 proposal.
- pokemon_tcg_arena_future_architecture — future application/worker seam.
- pokemon_tcg_entity_normalization — heuristic identity-resolution analysis.
- pokemon_tcg_manuscript_chapter_lineage — chapter-level manuscript provenance.