Datasets:
Add DATA_STANDARD.md for unified raw/gold policy
Browse files- DATA_STANDARD.md +64 -0
DATA_STANDARD.md
ADDED
|
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# SemantaAI Dataset Standard
|
| 2 |
+
|
| 3 |
+
This document defines the mandatory dataset standard for SemantaAI datasets published under `Grencape` on Hugging Face.
|
| 4 |
+
|
| 5 |
+
## 1) Repository Naming
|
| 6 |
+
- Namespace: `Grencape`
|
| 7 |
+
- Repo format: `semantaai-<domain>-<scope>`
|
| 8 |
+
- Examples:
|
| 9 |
+
- `Grencape/semantaai-main-assets`
|
| 10 |
+
- `Grencape/semantaai-synthetic-ru`
|
| 11 |
+
- `Grencape/semantaai-crypto-gold`
|
| 12 |
+
|
| 13 |
+
## 2) Canonical Layout (Required)
|
| 14 |
+
- `raw/`
|
| 15 |
+
- Immutable source history (OHLCV or source-native records)
|
| 16 |
+
- Per-file metadata (`*.meta.json`)
|
| 17 |
+
- Required indexes:
|
| 18 |
+
- `raw/DATASET_INDEX.json`
|
| 19 |
+
- `raw/DATASET_INDEX.csv`
|
| 20 |
+
- Required docs:
|
| 21 |
+
- `raw/README.md`
|
| 22 |
+
|
| 23 |
+
- `gold/`
|
| 24 |
+
- Training-ready enriched layer (`raw` + engineered features + targets + labels)
|
| 25 |
+
- Required indexes:
|
| 26 |
+
- `gold/DATASET_INDEX.json`
|
| 27 |
+
- `gold/DATASET_INDEX.csv`
|
| 28 |
+
- `gold/DATASET_INDEX_FILES.json`
|
| 29 |
+
- `gold/DATASET_INDEX_FILES.csv`
|
| 30 |
+
- Required docs:
|
| 31 |
+
- `gold/README.md`
|
| 32 |
+
- `gold/METRICS.md`
|
| 33 |
+
- `gold/BUILD_REPORT.json`
|
| 34 |
+
|
| 35 |
+
## 3) Naming Rules
|
| 36 |
+
- Do not use version folders in public paths (`v1`, `v2`, `labeled_v*`).
|
| 37 |
+
- Do not use ambiguous folder names like `main`, `final`, `new`.
|
| 38 |
+
- Use semantic layer names only: `raw`, `gold`.
|
| 39 |
+
- Partitioned gold files must follow:
|
| 40 |
+
- `gold/files/<SYMBOL>__<EXCHANGE>__<TF>__gold.parquet`
|
| 41 |
+
|
| 42 |
+
## 4) Versioning Policy
|
| 43 |
+
- Versioning is metadata-based, not folder-name-based.
|
| 44 |
+
- Track iterations in:
|
| 45 |
+
- `gold/BUILD_REPORT.json` (`built_at_utc`, source snapshot, feature set)
|
| 46 |
+
- git/HF commit history
|
| 47 |
+
- optional tags/releases
|
| 48 |
+
|
| 49 |
+
## 5) Compatibility Policy
|
| 50 |
+
- `raw/` should remain append-only where possible.
|
| 51 |
+
- `gold/` may evolve, but schema changes must be reflected in:
|
| 52 |
+
- `gold/METRICS.md`
|
| 53 |
+
- `gold/README.md`
|
| 54 |
+
- index files
|
| 55 |
+
|
| 56 |
+
## 6) Minimum Quality Gates Before Publish
|
| 57 |
+
- Indexes present and readable (`json` + `csv`)
|
| 58 |
+
- Metric glossary updated (`gold/METRICS.md`)
|
| 59 |
+
- Build provenance written (`gold/BUILD_REPORT.json`)
|
| 60 |
+
- Root `README.md` links to this standard
|
| 61 |
+
|
| 62 |
+
## 7) Mandatory Rule For Future Datasets
|
| 63 |
+
Every new SemantaAI dataset in `Grencape` must be published in this standard by default.
|
| 64 |
+
Any exception must be temporary and resolved to `raw/gold` structure before production use.
|