Fix reference artifact reproducibility in CI

Author: Muhra95

.github/workflows/harness-ci.yml

@@ -30,6 +30,7 @@ jobs:
             --config benchmarks/configs/default.json \
             --system-name "Reference Example System" \
             --system-version "1.0.0" \
+            --date-evaluated "2026-04-03" \
             --observations benchmarks/fixtures/reference_observations.json \
             --output /tmp/reference-example-v1.json
 

benchmarks/fixtures/reference_observations.json

@@ -94,12 +94,14 @@
     "Reference example generated from the documented v1 fixture set.",
     "Suitable as a schema example and smoke test for the v1 scaffold."
   ],
+  "artifact_class": "reference_example",
   "caveats": [
     "This is a reference example artifact, not a competitive public submission.",
     "Some scenario evidence is summarized rather than linked to raw transcripts."
   ],
   "evidence": [
     "benchmarks/fixtures/reference_observations.json",
     "benchmarks/configs/default.json"
-  ]
+  ],
+  "confidence_level": "low"
 }

eval/runner.py

@@ -51,6 +51,8 @@ def main() -> None:
     evaluator_notes = observations.get("evaluator_notes", [])
     caveats = observations.get("caveats", [])
     evidence = observations.get("evidence", [])
+    artifact_class = observations.get("artifact_class")
+    confidence_level = observations.get("confidence_level")
     metric_weights = config.get("metric_weights", {})
     metric_scores = build_metric_scores(metric_weights, scenario_scores, metric_overrides)
     scenario_coverage = round(
@@ -74,6 +76,7 @@ def main() -> None:
             "Benchmarking Persistent AI and Digital Twin as a Service Systems",
         ),
         "artifact_type": "benchmark_result",
+        "artifact_class": artifact_class,
         "system_name": args.system_name,
         "system_version": args.system_version,
         "date_evaluated": args.date_evaluated,
@@ -103,6 +106,7 @@ def main() -> None:
         "evaluator_notes": evaluator_notes,
         "caveats": caveats,
         "evidence": evidence,
+        "confidence_level": confidence_level,
     }
 
     write_json(args.output, payload)