docs(data): correct the path patterns in data/README.md

brand/ has a country folder, and soc/ and smartphone/ have a year
folder; the file showed none of them and left out eight categories. It
now lists every category, explains _verify/, and points at the table in
the root README. The note about curation no longer says every record is
hand-verified: `verified` on the record says that.

Author: Seungpyo1007

data/README.md

@@ -1,24 +1,39 @@
 # data/ — curated dataset
 
-These JSON files are the **curated, validated** records the API serves. Layout is
-singular folder names, organised by brand:
+These JSON files are the **curated, validated** records the API serves: one
+file per record, named `<slug>.json`, in singular category folders.
 
 ```
-brand/<slug>.json
-soc/<manufacturer>/<slug>.json
-smartphone/<brand>/<slug>.json
-gpu/<manufacturer>/<year>/<segment>/<slug>.json   # GPUs are split by release year then segment (consumer | enterprise)
-cpu/<manufacturer>/<year>/<segment>/<slug>.json   # CPUs split by release year then segment (consumer | enterprise)
+brand/<country>/<slug>.json                                 # country code in lower case, or unknown/
+soc/<manufacturer>/<year>/<slug>.json
+cpu/<manufacturer>/<year>/<consumer|enterprise>/<slug>.json
+gpu/<manufacturer>/<year>/<consumer|enterprise>/<slug>.json
+smartphone/<brand>/<year>/<slug>.json                       # or <year>/<base_model>/<slug>.json
+tablet/<brand>/<year>/<base_model>/<slug>.json              # watch/, pda/, laptop/ and monitor/ are the same
+device_catalog/<brand>/<slug>.json                          # device identities with no known release date
+software/<prefix>/<slug>.json                               # prefix = first two characters of the slug
+website/<prefix>/<slug>.json
+_verify/                                                    # not records: see below
 ```
 
-GPU `<segment>` is:
+The [root README](../README.md#layout) has the same table with file counts and
+an example for each category.
+
+The `consumer` / `enterprise` folder is a coarse split. For GPUs:
 * `consumer` — gaming/desktop & laptop dGPUs (GeForce GTX/RTX, Radeon RX/R9/HD, Arc, Voodoo, MTT S60/S70/S80, …)
 * `enterprise` — datacenter compute, AI accelerators, workstation pro cards (Tesla, A100/H100/B200, Quadro/RTX A/RTX PRO, Radeon Pro W, Instinct MI, MTT S3000/S4000, Vega Frontier, AI-focused TITAN V/RTX, …)
 
+For CPUs, `enterprise` holds server and most HEDT parts. The record's own
+`segment` field is finer: `desktop`, `laptop`, `hedt` or `server`.
+
+`_verify/` holds the verification ledger (`ledger.jsonl`), the status aggregate
+(`status.json`) and the caches under `state/`. TechEngine's verify tools write
+them; don't edit them by hand.
+
 > ⚠️ **This is a curated subset — NOT an exhaustive list of every device/chip.**
-> It is hand-verified and intentionally partial. Breadth is expanded out-of-band
-> through an internal pipeline that publishes curated records here after review;
-> each record carries `source_urls`. Don't assume a device is missing-by-error —
-> it may simply not be curated yet.
+> It is intentionally partial. New records come from TechEngine's collectors and
+> land here through a reviewed PR; each record carries `source_urls`, and
+> `verified` says whether it has passed verification. Don't assume a device is
+> missing-by-error — it may simply not be curated yet.
 
 Validate after edits: `python -m app.validate`. Add only real, sourced models.