Datasets:

GSMA
/

oran

@@ -59,76 +59,16 @@ configs:
     path: data/WG7/train.parquet
 ---
-# O-RAN Specifications Corpus
-Technical Specifications (TS) and Technical Reports (TR) from the O-RAN Alliance working groups (Release R005).
-Source documents (DOCX, PDF) were converted to GitHub-flavoured Markdown via the
-[Datalab Marker](https://www.datalab.to) API (`mode=accurate`) on 2026-05-14.
-Only the parsed `.md` text is published here — original binaries and extracted images are not redistributed.
-## Stats
-- **Documents**: 66
-- **Working groups**: 11
-- **Total characters**: 25,581,453
-- **Total words**: 2,681,736
-## Working groups
-Each row below corresponds to a `config_name` in the dataset viewer — pick one to browse only that
-working group, or pick `default` to see every document.
-| WG | Docs | Description |
-|---|---:|---|
-| `WG11` | 12 | Security — threat modeling, zero-trust, post-quantum, certificate management |
-| `WG2` | 12 | Non-RT RIC & A1 / R1 Interfaces — non-real-time intelligent controller |
-| `WG6` | 8 | Cloudification & Orchestration — O-Cloud, OAM, deployment |
-| `WG10` | 7 | OAM and SMO — Service Management & Orchestration, O1 interface, information models |
-| `WG1` | 6 | Use Cases & Overall Architecture — RAN architecture, NTN, CCIN use cases |
-| `WG3` | 6 | Near-RT RIC & E2 / Y1 Interfaces — near-real-time intelligent controller, xApps |
-| `WG5` | 5 | Open F1 / W1 / E1 / X2 / Xn Interfaces — midhaul and CU-DU split |
-| `TIFG` | 4 | Test & Integration Focus Group — certification badge, E2E test specifications |
-| `WG9` | 3 | Open X-Haul Transport — fronthaul/midhaul/backhaul transport, XPSAAS |
-| `WG4` | 2 | Open Fronthaul Interfaces — CUS-plane, M-plane, IOT, MP-YANG |
-| `WG7` | 1 | White-Box Hardware — open hardware reference designs |
-## Schema
-| Column | Type | Description |
-|---|---|---|
-| `wg` | string | Working group code (matches the config name) |
-| `doc_id` | string | Document filename stem |
-| `title` | string | First H1 in the markdown, or `doc_id` when absent |
-| `rel_path` | string | Path relative to the corpus root |
-| `text` | string | Parsed markdown body |
-| `n_chars` | int64 | `len(text)` |
-| `n_words` | int64 | Whitespace word count of `text` |
-| `n_lines` | int64 | Line count of `text` |
-## Loading
-```python
-from datasets import load_dataset
-# Every document across every working group
-ds = load_dataset("GSMA/oran", "default", split="train")
-# Just one working group (e.g. WG11)
-ds_wg = load_dataset("GSMA/oran", "WG11", split="train")
-print(ds_wg[0]["title"])
-print(ds_wg[0]["text"][:500])
-```
-## Layout
-```
-data/<wg>/train.parquet     one parquet shard per working group
-README.md                   this dataset card
 ```
-## License and use
-Private dataset; access restricted to GSMA organisation members on HuggingFace. The underlying
-documents are subject to GSMA / O-RAN Alliance redistribution terms — internal research use only.

     path: data/WG7/train.parquet
 ---
+# O-RAN source mirror
+This dataset mirrors the O-RAN ALLIANCE's published specifications and technical reports. Every source is kept twice: `original/` is the document as O-RAN released it (PDF and Word), and `marked/` is that same document converted to Markdown for search and retrieval, one `raw.md` per document with any figures extracted beside it. The two trees share identical paths, so a file in `original/` has a parsed twin at the matching path under `marked/`; both are grouped by O-RAN ALLIANCE working group (WG1 through WG11, plus the focus groups SFG, SuFG, and TIFG). `manifest.jsonl` carries one row per file (path, SHA-256, parse status), and `STATUS.md` reports how much of the corpus has been parsed.
+```text
+oran/
+  original/         # specs as published, by working group
+    WG1/  WG2/  WG4/  TIFG/  ...
+  marked/           # Markdown mirror of original/, same paths
+    WG4/.../raw.md  #   one per document, images beside it
+  manifest.jsonl    # one row per file: path, sha256, parse status
+  STATUS.md         # parse-coverage summary
 ```