docs: point v2 configs at Parquet; record the single-builder rule

README configs/dataset_info regenerated from the actual Parquet (all
nine configs loaded and split-counts cross-checked, 9/9 OK; features
round-tripped through Features._from_yaml_list). Adds the 2026-08-16
changelog entry, the serving-format note in the v2 section, and two new
hard rules in CLAUDE.md: one data format across ALL configs (the hub
infers a single builder from the first config's files), and dataset_info
dtypes must be YAML-format strings. dataset_card.json updated in sync;
.agents/.claude tooling caches ignored.

Claude-Session: https://claude.ai/code/session_0188gRcGZdDN828oyp4TGkt7
Co-Authored-By: Claude Fable 5 (max) <noreply@anthropic.com>

.gitignore

@@ -1 +1,5 @@
 .mimocode/
+
+# Local agent-tooling caches (hf CLI skill files) -- never dataset content.
+.agents/
+.claude/

CLAUDE.md

@@ -100,7 +100,7 @@ Standing implications for any work here:
 
 ## Editing the dataset card
 
-`README.md` frontmatter drives the Hugging Face Dataset Viewer. Two rules, both
+`README.md` frontmatter drives the Hugging Face Dataset Viewer. Four rules, all
 learned from breakage:
 
 1. **One config per file.** Never list several files under one `config_name` unless
@@ -110,6 +110,21 @@ learned from breakage:
    old `samples/*_SAMPLE_*.jsonl` were byte-exact prefixes of their parents; listing
    both counted those rows twice. Samples are now random draws and stay out of the
    configs entirely.
+3. **One data format across ALL configs.** A script-less hub dataset resolves a
+   single packaged builder from the *first* config's data files and applies it to
+   every config (`datasets/load.py`, `HubDatasetModuleFactoryWithoutScript`).
+   Listing JSONL and Parquet configs together made the viewer parse Parquet bytes
+   with the JSON loader (2026-08-16). Everything listed is Parquet now; the v2
+   JSONL in `full_data/` stays canonical but unlisted (served via `v2_parquet/`).
+   A 0-row Parquet file also cannot be listed — `datasets` ≤ 5.0.x fails with
+   `ArrowInvalid: BatchSize must be greater than 0` — so empty schema artifacts
+   (`v3/action_proposals`, `v3/safety_filter_log`) stay out of `configs:` until
+   populated.
+4. **`dataset_info` feature dtypes must be YAML-format strings** (`float64`,
+   `timestamp[ms]`), generated with `Features._to_yaml_list()` — repr-style
+   strings (`Value('float64')`) break the viewer's config-names step for the
+   whole dataset. Round-trip any generated block through
+   `Features._from_yaml_list()` before committing.
 
 After editing, validate before pushing — a malformed block breaks the viewer silently:
 

README.md

@@ -14,22 +14,22 @@ configs:
   data_files:
   - split: train
     path:
-    - full_data/neuromorphic_data.jsonl
+    - v2_parquet/gpu_telemetry/train-*.parquet
 - config_name: mining
   data_files:
   - split: train
     path:
-    - full_data/node_sync_harvest.jsonl
+    - v2_parquet/mining/train-*.parquet
 - config_name: hft
   data_files:
   - split: train
     path:
-    - full_data/ghost_market_log.jsonl
+    - v2_parquet/hft/train-*.parquet
 - config_name: qubic_ticks
   data_files:
   - split: train
     path:
-    - full_data/qubic_ticks_snn.jsonl
+    - v2_parquet/qubic_ticks/train-*.parquet
 - config_name: gpu_telemetry_v3
   data_files:
   - split: full
@@ -100,7 +100,7 @@ dataset_info:
   - name: train
     num_bytes: 84653192
     num_examples: 813973
-  download_size: 221699172
+  download_size: 6864325
   dataset_size: 84653192
 - config_name: mining
   features:
@@ -128,10 +128,10 @@ dataset_info:
     dtype: int64
   splits:
   - name: train
-    num_bytes: 12670092
+    num_bytes: 12670089
     num_examples: 120322
-  download_size: 31783185
-  dataset_size: 12670092
+  download_size: 5763683
+  dataset_size: 12670089
 - config_name: hft
   features:
   - name: timestamp
@@ -188,12 +188,12 @@ dataset_info:
   - name: train
     num_bytes: 7591295
     num_examples: 31573
-  download_size: 19143651
+  download_size: 2604075
   dataset_size: 7591295
 - config_name: qubic_ticks
   features:
   - name: timestamp
-    dtype: timestamp[s]
+    dtype: timestamp[ms]
   - name: tick
     dtype: int64
   - name: tick_rate
@@ -212,7 +212,7 @@ dataset_info:
   - name: train
     num_bytes: 1755520
     num_examples: 27430
-  download_size: 6029655
+  download_size: 637843
   dataset_size: 1755520
 - config_name: gpu_telemetry_v3
   features:
@@ -564,7 +564,7 @@ Each gains a config entry with its first populated release; the schemas are
 already frozen in the Parquet files and in
 [`v3_build.py`](https://github.com/rmems/spikenaut-telemetry-etl/blob/main/src/spikenaut_etl/v3_build.py).
 
-### v2 configs (unchanged, still authoritative for the raw captures)
+### v2 configs (same records, still authoritative for the raw captures)
 
 | Config | Records | Window | Description |
 |--------|--------:|--------|-------------|
@@ -575,6 +575,15 @@ already frozen in the Parquet files and in
 
 These are **disjoint capture windows**, not one continuous run.
 
+**Serving format:** the v2 configs load from verified Parquet conversions
+under `v2_parquet/` — a script-less Hub dataset gets exactly one builder,
+inferred from the first config's files, so JSONL and Parquet configs cannot
+share a card (see the 2026-08-16 changelog). The original JSONL files remain
+in `full_data/`, byte-untouched and canonical; the conversions are produced
+and fidelity-checked by the ETL (identical rows, columns, and nulls; the one
+representation change is `qubic_ticks.timestamp`: `timestamp[s]` →
+`timestamp[ms]`, same values at finer unit).
+
 ```python
 from datasets import load_dataset
 
@@ -818,8 +827,8 @@ no reproducible basis.
 ```
 rmems/Theseus-Quarry           Rust collectors → raw JSONL
         ↓
-rmems/spikenaut-telemetry-etl  ingest → validate → clean → publish   (v2 files)
-        ↓                       └─ spikenaut-etl build-v3 ──────────→ (v3/ Parquet)
+rmems/spikenaut-telemetry-etl  ingest → validate → clean → publish       (full_data/ JSONL)
+        ↓                       └─ spikenaut-etl build-v3 ─────────────→ (v3/ + v2_parquet/)
 rmems/Spikenaut-SNN-Telemetry  this dataset
         ↓
 rmems/Spikenaut-SNN            model training
@@ -847,6 +856,26 @@ or `qubic_ticks` today, nothing changes for you.
 
 ## 📝 Changelog
 
+### 2026-08-16 — all configs Parquet-served; viewer restored
+
+The 2026-08-15 release broke the Dataset Viewer for every v3 config
+(`JSON parse error: Invalid value. in row 0`): a script-less Hub dataset
+resolves **one** packaged builder from the **first config's** data files
+(`datasets/load.py`, `HubDatasetModuleFactoryWithoutScript`) and applies it to
+every config — so the JSON builder was fed Parquet bytes. Mixed JSONL/Parquet
+configs cannot work on the Hub.
+
+Since the v3 contract mandates Parquet, the four v2 configs now load from
+Parquet conversions under `v2_parquet/`, generated and fidelity-verified by
+`spikenaut-etl build-v3` (ETL PR #9): identical row counts, rows, columns,
+and nulls, with one representation change — `qubic_ticks.timestamp` is
+`timestamp[ms]` instead of `timestamp[s]` (Parquet has no seconds resolution;
+values unchanged). `full_data/*.jsonl` are byte-untouched and remain the
+canonical cleaned exports. Same-day earlier fix: `dataset_info` dtypes were
+serialized as repr strings (`Value('float64')`), which the viewer's
+config-names step cannot parse; now generated with `Features._to_yaml_list()`
+and round-trip-validated before every card commit.
+
 ### 2026-08-15 — v3.0.0: action-proposal trajectory restructure
 
 Built by [`spikenaut-etl build-v3`](https://github.com/rmems/spikenaut-telemetry-etl)

dataset_card.json

@@ -25,6 +25,7 @@
     "full_data/ghost_market_log.jsonl": 31573,
     "full_data/qubic_ticks_snn.jsonl": 27430
   },
+  "serving_note": "All nine configs load from Parquet. The four v2 configs are served from fidelity-verified conversions under v2_parquet/ because a script-less Hub dataset applies ONE builder (inferred from the first config's files) to every config, so JSONL and Parquet configs cannot share a card. full_data/*.jsonl remain the canonical cleaned exports, byte-untouched. Only representation change: qubic_ticks.timestamp is timestamp[ms] (was timestamp[s]); values identical.",
   "v3": {
     "schema_version": "3.0.0",
     "teacher_policy_version": "1.0.0",