Update spec pin, CI validation, and sample naming

.github/workflows/ci.yml

@@ -27,6 +27,11 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
+      - name: Check out the spec for record validation
+        uses: actions/checkout@v4
+        with:
+          repository: apache/sourcelume-spec
+          path: sourcelume-spec
       - uses: actions/setup-python@v5
         with:
           python-version: '3.12'
@@ -38,12 +43,7 @@ jobs:
         run: uv run python src/train.py --epochs 1 --chars 20000
       - name: Validate records against the spec
         env:
-          # TODO: pin to the spec release once PR #10 (TrainingProvenanceRecord)
-          # is merged and published; until then CI validates offline only.
-          SOURCELUME_SPEC: ''
-        run: |
-          if [ -n "$SOURCELUME_SPEC" ]; then
-            uv run python scripts/validate-records.py
-          else
-            echo "record validation skipped: no spec checkout available in CI yet"
-          fi
+          # Unpinned main for now: the spec has no published release yet, and
+          # validating against the current spec is the point of this check.
+          SOURCELUME_SPEC: ${{ github.workspace }}/sourcelume-spec
+        run: uv run python scripts/validate-records.py

README.md

@@ -64,9 +64,9 @@ make demo
 |---|---|
 | Dataset record: spec + registry validation | ✅ works today |
 | Dataset record: publish to Atlas | ✅ works with the Atlas stack up |
-| Training record: spec validation | ✅ works — **requires a spec checkout at the `model-usage-record` (PR #10) state or later**; set `SOURCELUME_SPEC` accordingly |
+| Training record: spec validation | ✅ works — validate against a `sourcelume-spec` checkout at `main` (the `TrainingProvenanceRecord` type was merged with PR #10); set `SOURCELUME_SPEC` accordingly |
 | Training record: plugin validate/publish | ❌ **expected to fail** — the Python plugin itself only models `ProvenanceRecord` (its Pydantic model rejects the record client-side). The plugin needs the same type-dispatch the spec validator has (tooling follow-up; see `make registry-validate`). |
-| Training record: registry validate/publish | ❌ **expected to fail** until the registry learns the `TrainingProvenanceRecord` type (registry follow-up to PR #10). Direct `curl -X POST .../records/validate` shows the registry's own rejection (HTTP 422, `must be the constant value 'ProvenanceRecord'`) — the demo's exhibit for that follow-up. |
+| Training record: registry validate/publish | ❌ **expected to fail** until the registry learns the `TrainingProvenanceRecord` type (registry follow-up to PR #10, currently in review as sourcelume-registry#15). Direct `curl -X POST .../records/validate` shows the registry's own rejection (HTTP 422, `must be the constant value 'ProvenanceRecord'`) — the demo's exhibit for that follow-up. |
 
 This demo **builds on PR #10 as given**: the training record type, the
 type-dispatching validator and the record examples come from that branch.

data/provenance-record.json

@@ -28,7 +28,7 @@
   "created": "2026-10-15T00:00:00Z",
   "added": "2026-10-15T00:00:00Z",
   "contentCreated": "1865-11-26T00:00:00Z",
-  "origin": "Plain text of Lewis Carroll's 'Alice's Adventures in Wonderland' (first published 26 November 1865), fetched verbatim from Project Gutenberg (ebook #11, https://www.gutenberg.org/ebooks/11) in October 2026 and bundled unchanged at data/corpus/alice-in-wonderland-1865.txt as demo training data. The work is in the public domain worldwide (published 1865, author died 1898); the Project Gutenberg framing text included in the file is distributed under the Project Gutenberg License (www.gutenberg.org). This record is authored and published by the Sourcelume training demo itself: the demo bootstraps its own registry preconditions instead of assuming records it did not create. Form: verbatim.",
+  "origin": "Plain text of Lewis Carroll's 'Alice's Adventures in Wonderland' (first published 26 November 1865), fetched verbatim from Project Gutenberg (ebook #11, https://www.gutenberg.org/ebooks/11) in October 2026 and bundled unchanged at data/corpus/alice-in-wonderland-1865.txt as demo training data. The work is in the public domain worldwide (published 1865, author died 1898); the Project Gutenberg framing text included in the file is distributed under the Project Gutenberg License (www.gutenberg.org). This record is authored and published by the Sourcelume sample itself: the sample bootstraps its own registry preconditions instead of assuming records it did not create. Form: verbatim.",
   "custodyChain": [
     {
       "agent": "https://www.gutenberg.org/",

scripts/validate-records.py

@@ -18,8 +18,8 @@
 SHACL), reusing the spec's own validator.
 
 SOURCELUME_SPEC must point at a sourcelume-spec checkout that includes the
-TrainingProvenanceRecord type — i.e. the model-usage-record branch (PR #10)
-until it is merged into main. Defaults to the local worktree of that branch.
+TrainingProvenanceRecord type (spec main since PR #10). There is no default:
+point it at your local checkout, e.g. SOURCELUME_SPEC=../sourcelume-spec.
 """
 
 import json
@@ -28,7 +28,14 @@
 from pathlib import Path
 
 BASE = Path(__file__).resolve().parent.parent
-SPEC = Path(os.environ.get("SOURCELUME_SPEC", "/private/tmp/wt-model")).resolve()
+spec_env = os.environ.get("SOURCELUME_SPEC", "")
+if not spec_env:
+    print(
+        "SOURCELUME_SPEC must point at a sourcelume-spec checkout (e.g. ../sourcelume-spec)",
+        file=sys.stderr,
+    )
+    sys.exit(2)
+SPEC = Path(spec_env).resolve()
 if not (SPEC / "tools" / "validate.py").is_file():
     print(f"SOURCELUME_SPEC is not a spec checkout: {SPEC}", file=sys.stderr)
     sys.exit(2)

src/train.py

@@ -140,12 +140,12 @@ def main() -> int:
         "id": MODEL_ID,
         "type": "TrainingProvenanceRecord",
         "identifier": "tiny-lm",
-        "name": "Tiny char-LM (Apache Sourcelume training demo)",
+        "name": "Tiny char-LM (Apache Sourcelume sample)",
         "version": "v1",
         "creator": [
             {
                 "type": "schema:Organization",
-                "name": "Apache Sourcelume training demo",
+                "name": "Apache Sourcelume sample",
                 "role": "originator"
             }
         ],
@@ -167,7 +167,7 @@ def main() -> int:
             f"{tokens_per_epoch}-character slice of the referenced corpus "
             f"({len(vocab)}-character vocabulary, {params} parameters, "
             f"final loss {history[-1]:.4f}, see model/training-run.json). "
-            f"The referenced dataset record is published by this demo itself "
+            f"The referenced dataset record is published by this sample itself "
             f"(data/provenance-record.json). Unsigned producer claim — see README."
         ),
     }