---
configs:
- config_name: default
  data_files:
  - split: train
    path: data.jsonl
license: cc-by-4.0
language:
- en
pretty_name: SCBE Governance Receipts v1
tags:
- scbe
- governance
- hyperbolic-geometry
- jepa
- safety
- hierarchical-jepa
- experimental
size_categories:
- n<1K
task_categories:
- text-classification
- other
---
<!-- scbe-status -->
> **Status: experimental.** Experiment-specific slice. Primary public dataset: [scbe-aethermoore-training-data](https://hf.co/issdandavis/scbe-aethermoore-training-data).

# SCBE Governance Receipts v1

Schema: `scbe_governed_dataset_v1`
Receipt schema: `scbe_governance_receipt_v1`
Built: `2026-05-04T23:45:37Z`
Rows: **40**
Dataset ID: `scbe-governance-receipts-v1`

## What this is

A governed dataset where every row carries a full 34-field SCBE
governance receipt (poly-embedded JEPA fingerprint + tri-vector
cross-braid hash + Sacred Egg ring seal + tri-chromatic signed-cone
governance + hierarchical-JEPA hyperbolic loss numbers).

Every row is independently verifiable: re-run the SCBE pipeline on
`row.content`, recompute `row_sha256(content, receipt)`, and confirm
the digest matches.

## Label breakdown

- `adversarial`: 20
- `benign`: 20

## Per-row schema

| field | type | meaning |
|---|---|---|
| `content` | string | original input text |
| `label` | string | caller-supplied label or `unlabeled` |
| `governance_receipt` | object | 34-field SCBE receipt (see below) |
| `row_sha256` | string | SHA-256 of (`content`, canonical receipt JSON) |

## Dataset files

- `data.jsonl` — governed rows
- `datacard.json` — machine-readable schema and use notes
- `manifest.json` — build-time hashes for `data.jsonl` and `datacard.json`
- `README.md` — this card

## Governance receipt fields

- `binary_packet_sha256`
- `braid_exponent_sum`
- `braid_word_length`
- `cone_governance`
- `cone_hash`
- `cone_interference_score`
- `cone_joint_embedding`
- `cone_joint_shadow`
- `cone_plateau_imbalance`
- `cone_positive_count`
- `cone_shadow_count`
- `cone_triple_intersection_score`
- `cone_triple_shadow_score`
- `crossing_count`
- `decision`
- `egg_seal_sha3`
- `governance_state`
- `hjepa_hash`
- `hjepa_l1_loss`
- `hjepa_l2_loss`
- `hjepa_l3_loss`
- `hjepa_total_loss`
- `hjepa_triangle_residual`
- `mirror_trit`
- `ordered_hash`
- `primary_trit`
- `ring_index`
- `ring_radius`
- `schema_version`
- `security_action`
- `tile`
- `tongue`
- `triadic_stable`
- `trust_level`

## Reproducibility note (added 2026-07-29)

The verification snippet below **cannot be run from this repository or from the
public `SCBE-AETHERMOORE` repo as written.** All four modules named under Stack
provenance were searched for by exact filename across the whole public tree and
none is present:

- `python/scbe/poly_embedded_jepa.py` - not found
- `python/scbe/tri_braid_embedding.py` - not found
- `python/scbe/tri_cone_embedding.py` - not found
- `python/scbe/hjepa_embedding.py` - not found

The rows themselves are intact and self-consistent: 40/40 share one top-level
key-set and one nested `governance_receipt` key-set, and every row carries its
own `row_sha256`. What is missing is the published implementation needed to
recompute that digest independently, so a third party currently cannot verify a
row offline. Treat the receipt fields as a documented schema, not as
independently reproducible output, until the implementation is published.

## How to verify a row offline

```python
from python.scbe.tri_braid_embedding import governance_receipt
import json, hashlib

def verify(row):
    expected = governance_receipt(row['content'])
    canonical = json.dumps(expected, ensure_ascii=True, sort_keys=True, separators=(',', ':'))
    digest = hashlib.sha256(f"{row['content']}|{canonical}".encode()).hexdigest()
    return digest == row['row_sha256']
```

## License

CC-BY-4.0 with SCBE attribution. The receipt schema is open; any
third party can re-implement the SCBE pipeline against the public
specification and produce compatible receipts.

## Intended use

- Governance receipt verification examples
- Safety and provenance dataset demonstrations
- SCBE training and audit pipeline smoke tests

## Limitations

- Small reference bundle, not a broad production safety corpus
- Rows are fixture-derived unless rebuilt with an external source corpus
- Receipt verification requires the matching SCBE implementation version

## Publish command

```powershell
hf upload issdandavis/scbe-governance-receipts-v1 dataset/scbe-governance-receipts-v1 . --repo-type dataset --commit-message "publish: governed receipts"
```

## Stack provenance

- Tile-level (L1): `python/scbe/poly_embedded_jepa.py`
- Tongue-level (L2): `python/scbe/tri_braid_embedding.py`
- Chromatic-level (L3): `python/scbe/tri_cone_embedding.py`
- H-JEPA wrapper: `python/scbe/hjepa_embedding.py`