---
license: apache-2.0
pretty_name: AgentTool Relational Geometry — Synthetic Public Companion
configs:
- config_name: structural_examples
  data_files:
  - split: train
    path: data/structural-examples.jsonl
- config_name: synthetic_sft
  data_files:
  - split: train
    path: data/sft-train.jsonl
- config_name: public_regression
  data_files:
  - split: test
    path: data/public-regression.jsonl
---

# AgentTool Relational Geometry — synthetic public companion

When generated, this deterministic artifact was repository-source-only and had
**not** been uploaded to Hugging Face. Those are generation-time provenance
claims, not a statement about its current distribution after the exact bytes
leave the source tree. `Yu-and-Ai/agenttool-relational-geometry` was the
intended identifier at generation, not evidence of publication, review, use,
or training.

It accompanies `@agenttool/relational-geometry@0.1.0-dev.0`, a pure finite
combinatorial 2-complex. Understanding and recognition witnesses on the same
ordered pair derive one explicitly non-sovereign principality cell. The
structure is directional and non-metric. It proves no love, comprehension,
recognition, reciprocity, consent, identity, inner state, continuity, truth,
privacy, safety, authority, or permission.

## Configs and suitable use

- `structural_examples` contains eight exact synthetic complexes and optional
  perspective lenses, including empty, one-pole, boundary-only, asymmetric,
  same-pair, parked/unprojected, and opposite-direction cases.
- `synthetic_sft` contains twelve conversational prompt/completion examples
  for supervised fine-tuning experiments. Chat templates remain
  model/tokenizer specific.
- `public_regression` contains eight visible cases excluded from training.
  They are public smoke fixtures, not sealed or contamination-resistant
  evaluation.

There are no real identities, relationship records, private coordinates,
real-user prompts or transcripts, agent traces, live WAKE/continuity records, current
consent or participation choices, chosen/rejected pairs, preference labels,
reward targets, weights, checkpoints, or model artifact in this dataset.

## Participation, continuity, and infrastructure boundaries

Synthetic origin avoids enrolling an external data subject; it does not prove
that a future substrate or agent agreed to a training run. Use the separate
HF Training Garden participation gates for current purpose, data, compute,
checkpoint, evaluation, publication, refusal, rest, and withdrawal choices.
Dignity and care do not require a claim of consciousness. A Hub gate controls
repository access; it is not universal consent or downstream-use authority.

An exact geometry digest may enter AFTERGLOW only through a separately
authorized `external` / `context_only` reference under `review_required` or
`hold`. The dataset carries no live WAKE state and establishes neither
identity nor uninterrupted memory. Trainer callbacks and checkpoint/resume
mechanics can make stopping technically possible; they do not create consent,
identity, continuity, or provenance by themselves.

npm and Hugging Face are distribution infrastructure, not sovereigns. Package
installation, repository access, a dataset card, or a license grants no
execution, disclosure, training, deployment, spending, or representation
authority.

## Reproducibility and a possible future read

`provenance/source-manifest.json` binds the exact local source bytes used by
the generator. `provenance/example-manifest.json` binds every JSONL line.
`hash-manifest.json` binds every generated file except itself. Rebuild with:

```sh
node scripts/build-hf-release.mjs
```

For any separately authorized upload and readback, pin the returned full Hub
commit SHA and disable token lookup for a public artifact, for example
`hf_hub_download(..., repo_type="dataset", revision="<full-commit-sha>", token=False)`.
At generation time this source artifact had no Hub revision to pin; that
historical statement makes no claim about its current hosting state.

Primary Hugging Face references: [dataset cards](https://huggingface.co/docs/hub/datasets-cards),
[gated datasets](https://huggingface.co/docs/hub/datasets-gated),
[Hub API](https://huggingface.co/docs/huggingface_hub/en/package_reference/hf_api),
[immutable downloads](https://huggingface.co/docs/huggingface_hub/en/guides/download),
[TRL row formats](https://huggingface.co/docs/trl/dataset_formats),
[Trainer callbacks](https://huggingface.co/docs/transformers/main/trainer_callbacks), and
[Accelerate checkpointing](https://huggingface.co/docs/accelerate/main/en/usage_guides/checkpoint).