---
type: reference
title: "Pokémon TCG - Harness Session Provenance"
description: "Portable provenance ledger for the August 12-14 Antigravity sessions and the 19 Claude project histories, with evidence boundaries and unresolved locators."
tags: [pokemon-tcg, provenance, transcripts, handoff, orchestration, harness]
timestamp: "2026-08-15T18:13:21-03:00"
---

# Pokémon TCG - Harness Session Provenance

## Purpose

This page records the session-level provenance behind the August 12-14
Pokémon TCG handoff. It is a locator and classification ledger, not a
transcript archive. It does not reproduce private conversation content,
credentials, machine-specific absolute paths or unverified implementation
claims.

The evidence progression is:

```text
transcript -> artifact or report -> code, database and Git
```

The arrow describes claim progression from origin, through a reviewable
derivative, to operational verification. A transcript establishes chronology,
dispatch and stated decisions. An artifact or report makes a claim easier to
inspect and cross-check. The live source tree, database and Git history decide
whether a module, result or acceptance condition exists in the project.

| Evidence layer | Records | Supports | Does not establish by itself |
|---|---|---|---|
| Transcript | User inputs, tool events, session UUIDs and dispatch history | Origin, chronology, scope and stated decisions | Implementation, passing tests or numerical reproducibility |
| Artifact or report | Handoff notes, matrices, audits, blueprints and summaries | Reviewable synthesis and targets for verification | Current runtime state when detached from the source tree |
| Code, database and Git | Source modules, persisted rows, run outputs and commits | Implementation, integrity, reproducible results and delivered changes | The full rationale that led to a design choice |

## Portable transcript locators

The locators below are intentionally portable. Resolve `<id>` only on a
machine that owns the corresponding local session store.

Antigravity full transcripts:

```text
~/.gemini/antigravity-cli/brain/<id>/.system_generated/logs/transcript_full.jsonl
```

The shorter sibling transcript, when present, is:

```text
~/.gemini/antigravity-cli/brain/<id>/.system_generated/logs/transcript.jsonl
```

Claude project histories:

```text
~/.claude/projects/-Users-alefita-workdir-pokemon-tcg/<id>.jsonl
```

The repository-side durable handoffs are project-relative, primarily under
`.agents/` and `docs/`. They are reviewable derivatives of the local session
stores and must not be treated as replacements for the source transcripts.

## Primary Antigravity session

The principal Antigravity session is:

```text
9189fa2e-93c2-4a04-9bf0-6d090880de27
```

The observed full transcript contains 4,907 JSONL records, including 438
`USER_INPUT` records, 327 `CODE_ACTION` records and 336 completed
`RUN_COMMAND` records. The same file also contains 127 `RUN_COMMAND` records
whose status was still `RUNNING`; therefore 336 is the completed-command
count, not the total number of command records.

The following ten artifacts were present in the main brain artifact root:

| Artifact | Role in the handoff | Evidence class |
|---|---|---|
| `ablation_matrix_results.md` | FP32 Stage 1-4 matrix and deck-conditioned results | Empirical artifact; cross-check against the tournament ledger |
| `fp16_precision_crisis.md` | MLX versus PyTorch precision-collapse narrative | Historical empirical artifact; runtime contract requires independent checking |
| `project_architecture_and_state.md` | Stage chronology, scratch-register interpretation and later transition | Source-reported chronology and interpretation |
| `tournament_analysis_2274.md` | Separate cross-agent tournament analysis | Distinct cohort; do not merge with the 420-game matrix |
| `kaggle_timezone_derivation.md` | Episode-ID boundary and UTC cutoff inference | Quantitative inference, not an official platform contract |
| `schema_mermaid.md` | Legacy/current SQLite table-family and foreign-key topology | Structural artifact; live counts come from the database audit |
| `moe_pipeline_blueprint.md` | RoPEND, MoE, vehicle and Apex design direction | Blueprint; not an implemented runtime |
| `master_project_state_and_handoff_plan.md` | Locked Meta roadmap, component inventory and acceptance checklist | Frozen handoff blueprint; unchecked claims remain open |
| `prompt_draft.md` | R1-R4 teamwork requirements and acceptance contract | Normative request; not evidence that gates passed |
| `deck_analysis_fp32.md` | FP32 deck analysis and checkpoint comparison | Analytical artifact; verify against stored run records |

The main handoff also reports a stop at the available compute or quota
boundary. The resulting continuation rule was documentation-only: preserve
the project tree, database and future architecture, and do not silently
repair or implement the planned work.

## Relevant Antigravity satellites

The handoff names 27 relevant satellites. All 27 have a matching brain
directory and full transcript locator in the current local store.

| Session ID | Scope | Classification | Current locator status |
|---|---|---|---|
| `245904ec-c5c7-4dca-a796-e04db0afb510` | FP32, checkpoints, Stage 1-3 | Empirical explorer | Brain root present |
| `687d75b1-19ff-4999-a56e-c5d40d869f0f` | Decks 633 and 21 | Deck and tournament explorer | Brain root present |
| `7d940327-ddbf-4352-85d8-eda03d63fc39` | FP16 legacy path | Historical runtime explorer | Brain root present |
| `be9ea07b-da36-4ff3-9b8d-93a69d094cfb` | `CLAUDE.md` audit | Repository-instruction audit | Brain root present |
| `da74ca6c-c155-4353-82d1-5c5695a60da1` | Sentinel and stop-state control | Coordination sentinel | Brain root present |
| `cd851a4f-6875-4819-9f25-1b23dd14cc1b` | Root orchestration | Root orchestrator | Brain root present |
| `4be1b993-121e-48d4-80bc-30b0269d271d` | Survey R1 | Repository survey explorer | Brain root present |
| `92697d23-559c-40ba-8d87-3e16a52bc678` | Survey R2 | Repository/data survey explorer | Brain root present |
| `f81eba1c-fa62-47bd-9d4a-bf47e90c85c5` | Survey R3 and R4 | Extended survey explorer | Brain root present |
| `9a189410-43b1-4cdc-bc2a-7a942180e59c` | M1 orchestration | Milestone sub-orchestrator | Brain root present |
| `607511a6-cff5-4500-bc83-612199d23b85` | RoPEND | M1 architecture explorer and blueprint | Brain root present |
| `33480337-1f72-44b7-b470-a9ff3955c574` | MoE, vehicle and Apex | M1 architecture explorer and blueprint | Brain root present |
| `2306715e-9fc9-4434-9175-5065874716a0` | Strict FP32 contract | M1 contract explorer | Brain root present |
| `eb2a3aa0-0b9f-4e52-9290-54dcb7e3df1d` | M1 worker | M1 implementation worker, interrupted | Brain root present |
| `f5143692-4dba-4e8a-aa34-f7465d296f9b` | M2 orchestration | Milestone sub-orchestrator | Brain root present |
| `4f2002e7-7972-4adb-a21f-95806599d6c4` | Oracle and auxiliary targets | M2 oracle explorer and handoff | Brain root present |
| `85613988-6d11-4bc9-9fac-20d973ad499b` | Database and foreign keys | M2 database explorer | Brain root present |
| `d2204035-8f25-4c8e-8e7b-dde82ad0dccb` | Elite dataset | M2 dataset explorer | Brain root present |
| `4877bc7d-bfc2-44d3-bc55-1a9dd628ba39` | M3 orchestration | Milestone sub-orchestrator | Brain root present |
| `9c810271-60ff-4b94-aae9-8348d0e534fb` | PageRank | M3 mathematical explorer | Brain root present |
| `7eb49539-1e33-49ad-8219-b274dca5eb2c` | Abelian and PageRank | M3 mathematical explorer | Brain root present |
| `b97c6afd-7027-4283-afed-98ff8590b4f1` | Metanoia and RFC | M3 research/documentation explorer | Brain root present |
| `f386a1cd-3536-45f1-855c-3e7003e85d98` | E2E orchestration | End-to-end test orchestrator | Brain root present |
| `df46139c-195d-4b55-b5e8-26abe6699282` | Tier 1; test file reached Git | E2E test writer; partial delivered artifact | Brain root present |
| `c8f415bf-2604-4699-8793-49bd9c6c2946` | Tier 2 attempt not in `HEAD` | E2E test writer; not in current snapshot | Brain root present |
| `24cdbd60-d9ce-4b9b-ab18-5c4f1e25982b` | Tier 3 briefing only | E2E briefing; no test delivery established | Brain root present |
| `e2cb3cca-541d-4f19-a464-c9c95ddf07d6` | Tier 4 attempt not in `HEAD` | E2E test writer; not in current snapshot | Brain root present |

The satellite identifiers are provenance labels. They do not establish that
the requested worker, reviewer, test writer or acceptance gate completed.
The repository `.agents` artifact ledger remains the source for orchestration
topology, while code, database and Git remain the source for delivery status.

## Claude project histories

The 19 Claude histories were classified from the `entrypoint` field in each
JSONL file, not from filename style or application-support directory names.
The observed split is four 1P histories with `entrypoint: claude-desktop` and
fifteen 3P histories with `entrypoint: claude-desktop-3p`.

| History ID | Entrypoint | Classification |
|---|---|---|
| `b99ca436-d9c3-4f23-ad6e-9ce870a00023` | `claude-desktop` | 1P substantive |
| `7bdae6cf-c7a4-4faf-8a06-3b2d8580d44a` | `claude-desktop` | 1P substantive |
| `8621fb6c-e691-4f13-a75e-64a33845bfba` | `claude-desktop` | 1P substantive |
| `3a872e38-b2b2-41d6-9def-0f98828b5b0a` | `claude-desktop` | 1P short |
| `8c046344-a11e-44d0-bb73-951e7cf972e2` | `claude-desktop-3p` | 3P substantive |
| `22aea1b3-75c0-4fed-90e3-270af2d0b7e5` | `claude-desktop-3p` | 3P substantive |
| `affa6e4d-db3d-4b62-81e1-08d34dd15a08` | `claude-desktop-3p` | 3P substantive |
| `6fb56962-f570-47a4-8ecc-b52f765aa37c` | `claude-desktop-3p` | 3P substantive |
| `43be699c-2bc5-4ea0-8391-243cf5578bf1` | `claude-desktop-3p` | 3P substantive |
| `a066e4b6-7edc-44a6-9a40-b63513753f0a` | `claude-desktop-3p` | 3P substantive |
| `086d2d94-c845-4906-b3ab-3306694fb37b` | `claude-desktop-3p` | 3P orientation fork |
| `207406e5-e580-4fef-aa85-305f0e110a53` | `claude-desktop-3p` | 3P orientation fork |
| `26b88d3d-a2dd-4a86-a150-888c224f95fb` | `claude-desktop-3p` | 3P orientation fork |
| `45a44c12-60fe-44ea-bc41-7fbb159ead14` | `claude-desktop-3p` | 3P orientation fork |
| `08545a97-f9f2-4ab6-9d86-b0d7cbb2eba3` | `claude-desktop-3p` | 3P short or interrupted |
| `765b2efe-d5b0-47f1-9e22-046579de5503` | `claude-desktop-3p` | 3P short or interrupted |
| `815449d4-b24a-4d6d-aaa8-aeb396f45ac5` | `claude-desktop-3p` | 3P short or interrupted |
| `50b0ee9f-9922-4b39-b4fe-e69ec4d6e78d` | `claude-desktop-3p` | 3P short or interrupted |
| `dacd7d1b-b433-4360-9d35-7e377726fb96` | `claude-desktop-3p` | 3P short or interrupted |

### Non-authoritative application metadata

The local path `~/Library/Application Support/Claude-3p/claude_desktop_config.json`
suggests a 3P application-support tree, while its neighboring
`sentry/scope_v3.json` identifies the telemetry application as `Claude`.
That mismatch is non-authoritative for project-history classification. The
Sentry scope is telemetry metadata and the directory name is a storage
convention; neither replaces the per-history `entrypoint` field.

## Substantive decisions and evidence gaps

The session family supports the following decisions and boundaries:

- Keep transcript, report, source, database and Git roles separate. A
  handoff sentence must not be promoted to a current implementation fact
  without a matching code, database or commit observation.
- Preserve strict FP32 as the current runtime contract while retaining FP16
  material as historical evidence of the precision incident. Legacy filenames
  and report language do not override the live source boundary.
- Keep the 420-game Stage 1-4 matrix, the separate 3,600-game cross-agent
  analysis and other curriculum cohorts separate. Their rates and stage labels
  must not be merged into a synthetic benchmark.
- Preserve RoPEND, MoE, vehicle-draft, Apex, PageRank-Abelian and later
  Metanoia work as planned, experimental or research material unless direct
  implementation evidence promotes a specific component.
- Treat the M1, M2, M3 and E2E tracks as handoff topology with open gates. The
  500-match acceptance run, database-parity gate and full E2E validation remain
  pending in this provenance snapshot.
- Continue the immediate work as documentation-only after the quota boundary.
  No project code, model, database or test suite was changed by this Wikifita
  operation.

The principal gaps are the two missing current Antigravity brain roots noted
above, the distinction between completed and still-running command records,
and the fact that session UUIDs alone cannot prove implementation or test
completion. The repository `.agents` tree also contains attempts and briefings
whose files are absent from `HEAD`; those remain negative-space evidence, not
material to reconstruct.

## Cross-references

- [[pokemon_tcg_cross_harness_provenance]]
- [[pokemon_tcg_orchestration_artifact_ledger]]
- [[pokemon_tcg_aug14_architecture_and_handoff_audit]]
- [[pokemon_tcg_aug14_data_etl_database_audit]]
- [[pokemon_tcg_aug14_ablations_tournaments]]
- [[pokemon_tcg_docs_corpus_provenance]]
- [[pokemon_tcg_decision_trail]]