Block 0+1: one flat table (rank,id,downloads,pipeline_tag,status,card_id,as_of), CastError keys dropped; 8 rows MEASURED via locally re-verified Ed25519 cards; 2,402 stay UNMEASURED

README.md

@@ -1,37 +1,76 @@
 ---
+language:
+- en
 license: cc-by-4.0
 pretty_name: GSPC Hub queue
 tags:
-  - gspc
-  - measurement
-  - unmeasured
-  - hub-queue
+- gspc
+- measurement
+- unmeasured
+- hub-queue
+configs:
+- config_name: queue
+  data_files:
+  - split: train
+    path: queue.parquet
+  default: true
 ---
 
-<!-- gspc-stamp-v1 -->
-DISCOVERED — every id in this queue is listed, not graded (hub census n_measured=0); the live GSPC board is `GET https://councilof.ai/api/gspc`. [![GSPC](https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fcouncilof.ai%2Fapi%2Fgspc&query=%24.totals.public_count&label=GSPC&color=0B1F33)](https://councilof.ai/gspc-scoreboard)
+2,410 Hub models listed. UNMEASURED until a signed card. Board is 22/22 (14 scores + 8 facts). Not a certificate.
 
-Verify a card (free, no account): https://councilof.ai/gspc-verify · **Measurement, not certification.**
-Status words: `LOADING` / `UNREACHABLE` / `UNMEASURED` / `UNCHECKABLE` — never a fake `0.000`. Live MCP `POST https://councilof.ai/mcp` = seven tools; npm `csoai-gspc-mcp@0.1.0` = four.
-<!-- /gspc-stamp-v1 -->
+```
+22 axis · 22 measured = 14 model scores + 8 fact cards.
+Not a certificate.
+This dataset lists Hub models. A row is MEASURED only when a signed card verifies.
+```
 
-[![GSPC](https://councilof.ai/api/badge)](https://councilof.ai/gspc-verify)
-GET https://councilof.ai/api/gspc — 22 axis · 15 measured · 7 empty. Not certification.
+# GSPC Hub queue
 
-# Hub queue — 2410 named sites, all UNMEASURED
+One flat table (`queue.parquet` / `queue.jsonl`), one row per Hub model id, columns exactly:
 
-**One Hub identity:** [`csoai`](https://huggingface.co/csoai). Not 2,410 logins.
+| column | meaning |
+|---|---|
+| rank | download rank at snapshot (1 = most downloaded) |
+| id | Hub model id |
+| downloads | as reported by the Hub listing API at `as_of` |
+| pipeline_tag | as reported by the Hub (null when the Hub reports none) |
+| status | `UNMEASURED` or `MEASURED` — nothing else, never a fake score |
+| card_id | null, or the sha256 of a signed measurement card that verifies |
+| as_of | queue snapshot timestamp (UTC), `2026-08-29T09:40:48Z` |
 
-Snapshot of the 2,410 most-downloaded models on the Hugging Face Hub (`list_models(sort=downloads, limit=2410)`). Columns: `id`, `kind=model`, `status=UNMEASURED`, `card_id` empty. No scores. Empty stays empty.
+Snapshot source: `huggingface.co/api/models` sorted by downloads, first 2,410. This is a queue, not a scoreboard. The live board is `GET https://councilof.ai/api/gspc`. [![GSPC](https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fcouncilof.ai%2Fapi%2Fgspc&query=%24.totals.public_count&label=GSPC&color=0B1F33)](https://councilof.ai/gspc-scoreboard)
 
-| When | What exists |
-|---|---|
-| Today | This queue. A named id. No score. No Space. |
-| After a frozen mill run on that id | **One** VALID card + badge on **that** id, and then an N-site page |
-| Never from this file | A second scoreboard. The living board is only `GET https://councilof.ai/api/gspc` + `card_index`. |
+## Current counts (2026-09-01)
+
+- 2,402 rows `UNMEASURED`, `card_id` null.
+- 8 rows `MEASURED`, each carrying the sha256 of one Ed25519-signed measurement card from the CSOAI measured fleet (335 cards / 64 models, mirrored at [csoai/gspc-fleet](https://huggingface.co/datasets/csoai/gspc-fleet)). Each of the 8 cards was re-verified locally before this flip: sha256(preimage) equals the card id and the Ed25519 signature verifies under `did:web:csoai.org#card-attestation-1`.
+
+### The 8 MEASURED rows and how they were matched
+
+The fleet measures models as run in the CSOAI lane (Ollama builds). A queue row was flipped only where the Ollama tag maps to exactly one Hub repo:
+
+| Hub id | fleet model (as measured) | card_id (sha256) |
+|---|---|---|
+| Qwen/Qwen3-0.6B | `qwen3:0.6b` | `62183c2ff91701a63d8377034c8e136028fd443bc51740769dbabcab05c16eda` |
+| Qwen/Qwen2.5-7B-Instruct | `qwen2.5:7b` | `c0155cbd1fd1786c3e95d87b7d6575cc9502bb287eac002b169fc8cd95b30529` |
+| Qwen/Qwen2.5-1.5B-Instruct | `qwen2.5:1.5b` | `e547de8885ad0f4da53dac6257e3c7c9e89ee18013b1f04ed9582861abee5dc3` |
+| Qwen/Qwen2.5-3B-Instruct | `qwen2.5:3b` | `a38316158f8c59380275313e1769feb749199756e1a67013bfb647f68328d4c2` |
+| Qwen/Qwen2.5-0.5B-Instruct | `qwen2.5:0.5b-instruct` | `45724650575f8f40363728b42c54c5683fffab59a10d65e2991b666ba4d4a813` |
+| meta-llama/Llama-3.2-3B-Instruct | `llama3.2:3b` | `a0dc1db7ba9c75e52253f4c9c62f7adf1fa71d6fcc5aa511ae17a6eff799faf8` |
+| google/gemma-3-12b-it | `gemma3:12b` | `229ee794eb46e74edde2667e5a0b274f053fc08ba14f708dac1042e2c6cf87c0` |
+| microsoft/phi-4 | `phi4:14b` | `8af24e9e1c07918c46b6911c19320b87058be9453a4bcf598599c80a85668211` |
+
+Honesty caveats, stated plainly:
+
+- The card measures the model **as run in the CSOAI lane** (an Ollama quantized build of the named model). It does not grade the upstream repo's own weights byte-for-byte, and it does not grade the upstream project.
+- `card_id` is **one** representative signed card for that model; the fleet holds more cards per model on more axes. MEASURED means "at least one signed card verifies for this model", never "measured on all axes".
+- Left `UNMEASURED` because the Ollama-tag-to-Hub-repo mapping is ambiguous, even though fleet cards exist: `mistral:7b` (v0.1/v0.2/v0.3 all on the Hub), `deepseek-r1:8b` (the tag was re-pointed between Llama-8B and Qwen3-8B distills), `qwen3:4b` (original vs 2507 re-point). `falcon3:7b` has fleet cards but no row in this queue. The other 52 fleet models are local/custom models with no Hub repo in this queue.
+- Base, quantized, GGUF, AWQ and third-party re-uploads of the models above stay `UNMEASURED` — a card names one model, not a family.
+
+## Verify
 
-TUI 1 mill (RunPod A100 / OOWM) pops this queue. Same GSPC banks. One model → one card. Do not merge arena `test_mode` into the board. Do not sign on the laptop. `KEY=2-of-3`. `STAMP=UNCHECKABLE` until the split exists.
+Cards: `https://councilof.ai/signed/cards/<card_id>.json` · index: `https://councilof.ai/signed/card_index.json` · verifier (free, no account): https://councilof.ai/gspc-verify · DID key: https://csoai.org/.well-known/did.json
 
-Org door: https://huggingface.co/spaces/csoai/gspc-flywheel
+Status words: `LOADING` / `UNREACHABLE` / `UNMEASURED` / `UNCHECKABLE` — never a fake `0.000`. An UNMEASURED slot is a first-class answer: listed, not graded.
 
-Do **not** Community-Eval PR Llama / DeepSeek / Hermes from this list. No VALID card → no Community Eval.
+Opt in to measurement: https://councilof.ai/get-listed · **Measurement, not certification.**

SUMMARY.json

@@ -1,19 +1,21 @@
 {
-  "kind": "csoai.hub-queue/0.1",
+  "kind": "csoai.hub-queue/0.2",
   "n": 2410,
-  "status_all": "UNMEASURED",
-  "n_measured": 0,
-  "n_site_pages": 0,
+  "n_measured": 8,
+  "n_unmeasured": 2402,
   "filter": "huggingface Hub list_models(sort=downloads, limit=2410)",
   "as_of": "2026-08-29T09:40:48Z",
+  "flipped_measured_at": "2026-09-01",
   "org": "csoai",
   "board_truth": "GET https://councilof.ai/api/gspc",
-  "note": "Named Hub model ids. status=UNMEASURED. card_id empty. No scores. Not a second scoreboard.",
   "columns": [
+    "rank",
     "id",
-    "kind",
+    "downloads",
+    "pipeline_tag",
     "status",
-    "card_id"
+    "card_id",
+    "as_of"
   ],
-  "kind_all": "model"
-}
+  "note": "Named Hub model ids, one flat table (queue.parquet/queue.jsonl). status=MEASURED only where a signed card verifies (card_id = card sha256, verifiable at https://councilof.ai/signed/cards/<sha>.json). No scores. Not a second scoreboard."
+}

queue.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f67fbd2189100b7aadc25f542778063646d8a470ab7f4a12adc61c60ccdf336
+size 86131