ci: add training-data/ path trigger + scripts/validate-scene-data.py (#647)

.gitea/workflows/validate-training-data.yml

@@ -5,7 +5,9 @@ on:
     paths:
       - 'training/data/**/*.jsonl'
       - 'training/data/**/schema.json'
-      - 'training/data/**/validate.py'
+      - 'training-data/*.jsonl'
+      - 'training-data/schema.json'
+      - 'scripts/validate-scene-data.py'
 
 jobs:
   validate:
@@ -18,21 +20,11 @@ jobs:
         with:
           python-version: '3.11'
 
-      - name: Validate scene description JSONL
+      - name: Validate scene descriptions (training-data/)
+        run: |
+          python3 scripts/validate-scene-data.py training-data/scene-descriptions-*.jsonl
+
+      - name: Validate scene descriptions (training/data/)
         run: |
           cd training/data/scene-descriptions
           python3 validate.py *.jsonl
-      - name: Validate training data provenance
-        run: |
-          cd training
-          python3 -c "
-          from training_pair_provenance import validate_provenance
-          import json, sys, glob
-          issues = 0
-          for f in glob.glob('data/*.jsonl'):
-              report = validate_provenance(f)
-              print(f'{f}: {report["coverage"]:.0f}% coverage ({report["with_provenance"]}/{report["total"]})')
-              if report['missing_provenance'] > 0:
-                  print(f'  WARNING: {report["missing_provenance"]} pairs missing provenance')
-          sys.exit(0)
-          "