feat: migrate submission format to v4 pointer style

- Remove submissions/scores/ (scores now live in submitter model repos
under .eval_results/<dataset-org>/<dataset-name>/scores.txt)
- Rewrite submissions/README.md with HF CLI submitter flow
- Rewrite submissions/results_template.yaml with v4 schema:
schema_version=4, artifact.scores_url + scores_sha256, system.paper
as a structured block

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

submissions/README.md

@@ -1,77 +1,81 @@
 # Benchmark Submissions
 
-To submit a result, open a pull request adding `<system-slug>.yaml` and `scores/<system-slug>.txt` to this directory.
+To submit a result, you'll upload two files (no git clone required):
 
-## Submission format
+1. **`scores.txt`** to **your own HF model repo** under `.eval_results/SpeechAntiSpoofingBenchmarks/ASVspoof2019_LA/scores.txt`.
+2. **`<your-slug>.yaml`** as a pull request to **this dataset repo** under `submissions/<your-slug>.yaml`.
 
-Each submission is two files:
+The YAML in this repo carries a pinned URL pointing at your `scores.txt`, plus its sha256. Scores files do not live in this repo.
 
-### `submissions/<system-slug>.yaml`
+## Submitter workflow
 
-```yaml
-schema_version: 1
+### 1. Generate `scores.txt` locally
 
-system:
-  name: AASIST
-  slug: aasist-clovaai-default
-  paper: https://arxiv.org/abs/2110.01200
-  code: https://github.com/clovaai/aasist
-  checkpoint: https://huggingface.co/...
-  description: Reference AASIST, default config.
+```bash
+speech-spoof-bench run \
+  --model-module <your_package>:<YourModelClass> \
+  --datasets SpeechAntiSpoofingBenchmarks/ASVspoof2019_LA
+```
 
-dataset:
-  id: SpeechAntiSpoofingBenchmarks/ASVspoof2019_LA
-  revision: <git-revision-sha>
-  split: test
+Output: `results/ASVspoof2019_LA/scores.txt` (one line per utterance, `<utterance_id> <score>`, higher = more bonafide).
 
-scores:
-  eer_percent: 0.83
-  n_trials: 71237
-  n_skipped: 0
+### 2. Upload `scores.txt` to your model repo
 
-artifact:
-  scores_file: scores/aasist-clovaai-default.txt
-  scores_sha256: <sha256-of-scores-file>
-  bench_version: speech-spoof-bench==0.3.1
+```bash
+huggingface-cli upload <your-owner>/<your-model-repo> \
+  results/ASVspoof2019_LA/scores.txt \
+  .eval_results/SpeechAntiSpoofingBenchmarks/ASVspoof2019_LA/scores.txt \
+  --repo-type=model \
+  --commit-message="Add ASVspoof2019_LA scores"
+```
 
-reproduction:
-  reproduced_by: SpeechAntiSpoofingBenchmarks
-  reproduced_at: 2026-05-20
-  reproduced_bench_version: speech-spoof-bench==0.3.1
-  match: scoring
+**Note the commit sha** the CLI prints — you'll need it in the next step.
 
-submitter:
-  hf_username: <your-hf-username>
-  contact: <your-email>
+### 3. Fill in the submission YAML
 
-submitted_at: 2026-05-19
-notes: "Description of your system configuration."
-```
+Copy `results_template.yaml` to `<your-slug>.yaml` and fill in every field. The two most important fields:
 
-### `submissions/scores/<system-slug>.txt`
+- `artifact.scores_url`: the **pinned** URL to your uploaded scores file. Use the commit sha from step 2, not `main`:
+  ```
+  https://huggingface.co/<your-owner>/<your-model-repo>/resolve/<commit-sha-from-step-2>/.eval_results/SpeechAntiSpoofingBenchmarks/ASVspoof2019_LA/scores.txt
+  ```
+  URLs with `/resolve/main/` are rejected because they're mutable.
+- `artifact.scores_sha256`: `sha256sum results/ASVspoof2019_LA/scores.txt | awk '{print $1}'`.
 
-Plain text, one line per utterance:
+Leave the `reproduction:` block empty — the maintainer fills it in at merge time.
 
+### 4. Open the PR via HF CLI
+
+```bash
+huggingface-cli upload \
+  SpeechAntiSpoofingBenchmarks/ASVspoof2019_LA \
+  <your-slug>.yaml submissions/<your-slug>.yaml \
+  --repo-type=dataset \
+  --create-pr \
+  --commit-message="Add <your-slug> submission"
 ```
-LA_E_2834763 -1.234
-LA_E_1665632  2.871
-```
 
-Format: `<utterance_id> <score>`. Higher score = more bonafide.
+The CLI prints a PR URL. That's it.
+
+### 5. Wait for maintainer reproduction
+
+A maintainer runs `speech-spoof-bench reproduce --scoring <PR-branch>`, which:
+- Fetches `scores_url`.
+- Verifies the sha256 against `artifact.scores_sha256`.
+- Recomputes EER from the file.
+- Compares to your claimed `scores.eer_percent` (must match within 1e-6).
 
-## Verification
+If it passes, the maintainer fills in `reproduction:` and merges. If it fails, you get a comment on the PR explaining why.
 
-All submissions are verified by a maintainer before merging:
+## Verification levels
 
-1. **Scoring verification** (mandatory): maintainer recomputes EER from your `scores.txt`. Must match the YAML's `eer_percent` within 1e-6.
-2. **Inference verification** (optional): maintainer re-runs your model end-to-end. Must match within 0.05% EER.
+| Level | What the maintainer checks | Cost |
+|---|---|---|
+| `scoring` (default) | sha + recomputed EER from your `scores.txt`. | Seconds. |
+| `inference` (optional, follow-up) | Re-runs your checkpoint end-to-end and regenerates `scores.txt`. Must match within 0.05% EER. | Expensive. |
 
-Submissions without a `reproduction:` block are never displayed in the arena.
+Submissions without a `reproduction:` block never appear in the arena.
 
-## How to submit
+## What about git clone + push?
 
-1. Evaluate your system on this dataset.
-2. Generate `scores/<system-slug>.txt` (one score per utterance, higher = more bonafide).
-3. Copy `results_template.yaml` to `<system-slug>.yaml` and fill in all fields.
-4. Compute the SHA-256 of your scores file: `sha256sum scores/<system-slug>.txt`.
-5. Open a pull request with both files.
+You can do it that way too, but for a single 2 KB YAML it's massively heavier. The HF CLI path is the documented one.

submissions/results_template.yaml

@@ -1,12 +1,16 @@
-schema_version: 1
+schema_version: 4
 
 system:
   name: ""
   slug: ""
-  paper: ""
+  description: ""
   code: ""
   checkpoint: ""
-  description: ""
+  paper:
+    arxiv_id: ""
+    url: ""
+    bibtex: |
+      @article{...}
 
 dataset:
   id: SpeechAntiSpoofingBenchmarks/ASVspoof2019_LA
@@ -19,15 +23,18 @@ scores:
   n_skipped: 0
 
 artifact:
-  scores_file: scores/.txt
+  # Must be pinned by commit sha. Pattern:
+  #   https://huggingface.co/<owner>/<repo>/resolve/<commit-sha>/.eval_results/SpeechAntiSpoofingBenchmarks/ASVspoof2019_LA/scores.txt
+  scores_url: ""
   scores_sha256: ""
   bench_version: ""
 
+# Leave this block empty — the maintainer fills it in at merge.
 reproduction:
   reproduced_by: ""
   reproduced_at: ""
   reproduced_bench_version: ""
-  match: scoring
+  match: ""
 
 submitter:
   hf_username: ""