docs: add model card with tier definitions and counts

README.md

@@ -1,34 +1,75 @@
 ---
-dataset_info:
-  features:
-  - name: tea_id
-    dtype: string
-  - name: tradition
-    dtype: string
-  - name: tier
-    dtype: string
-  - name: text
-    dtype: string
-  - name: provenance_chain
-    dtype: string
-  - name: contamination_flags
-    list: string
-  - name: label
-    dtype: int64
-  splits:
-  - name: train
-    num_bytes: 37006
-    num_examples: 14
-  - name: test
-    num_bytes: 18508
-    num_examples: 7
-  download_size: 62707
-  dataset_size: 55514
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: test
-    path: data/test-*
+license: cc-by-sa-4.0
+language:
+  - en
+  - ar
+  - el
+  - sa
+  - zh
+tags:
+  - abacus
+  - provenance
+  - classification
+  - pre-modern-mathematics
+  - bert-training
+size_categories:
+  - n<1K
 ---
+
+# ABACUS Provenance Trainset v0.1
+
+**Phase 1 (bootstrap)** — labeled from existing ABACUS corpus TEA records.
+Phase 2 will add active ingest to fill gaps.
+
+## Purpose
+
+Training data for the W6.2 BERT provenance classifier (Decision #53).
+Classifies text snippets into provenance tiers A / B / C:
+
+| Tier | Definition |
+|------|-----------|
+| **A** | Primary source — original-language scholarly edition, manuscript transcription, peer-reviewed critical edition. Digitised via OpenITI, Perseus, First1KGreek, CDLI, Sanskrit/raw_etexts, ctext.org, etc. No translated-by intermediary. |
+| **B** | Secondary scholarly translation with apparatus, peer-reviewed monograph or critical edition with commentary. Tradition-preserving lineage. Post-1929 but pre-Bourbaki epistemic stance. |
+| **C** | Popularization, encyclopedic, derivative, web-content. Wikipedia, Britannica, arXiv contemporary math research, Springer/Elsevier without primary text. |
+
+## Tradition Classes
+
+| Key | Coverage |
+|-----|---------|
+| `islamic` | OpenITI / Athar / Sufi-mystical / Arabic-Persian sources |
+| `greek` | Greek-Hellenistic / Perseus / First1KGreek / Greek-Classical |
+| `vedic` | Vedic-Kerala / Sanskrit / raw_etexts / GRETIL |
+| `chinese` | East-Asian-Wasan / ctext / Kanseki / Japanese / Korean |
+| `math` | Pre-mathematical-substrate / Erdős / MathOverflow / methodology / failed-approaches / misattribution |
+
+## Counts
+
+| Tradition | Tier A | Tier B | Tier C | Total |
+|-----------|--------|--------|--------|-------|
+| chinese | 3 | 3 | 2 | 8 |
+| greek | 5 | 0 | 0 | 5 |
+| islamic | 0 | 0 | 0 | 0 |
+| math | 3 | 3 | 2 | 8 |
+| vedic | 0 | 0 | 0 | 0 |
+| **TOTAL** | **11** | **6** | **4** | **21** |
+
+## Labeling
+
+Labels are deterministic, derived from `provenance_chain` and `contamination_flags`
+using the callable function `label_provenance_tier(row)` in
+`pattern/scripts/provenance_tier_labeler.py` (Garman-Unified-Systems/abacus repo).
+
+UNKNOWN-tier rows are excluded from this dataset.
+
+## Split
+
+90/10 train/test, stratified by (tradition, tier).
+
+## Source
+
+ABACUS TEA corpus — `Garman-Unified-Systems/abacus` pattern/corpus/data/*.jsonl.
+All source texts are public domain or CC-BY-SA-3.0/4.0.
+
+## Phase 2 Gap Map
+
+See `_provenance_labels_v0_1_audit.md` in the source repo for cells under 50 / under 20 / empty.