Preliminary findings propagation @57dca65

tools/hf_space/runner.py

@@ -456,6 +456,7 @@ def _validate_zip_streaming(*, api: HfApi, dataset: str, token: str | None,
 
     merged_results: list[dict] = []
     merged_layout: list[dict] = []
+    merged_preliminary: list[dict] = []
     # Set when ANY processed unit's results.json carries
     # preliminary_check_failed=true (the validator's strict pre-check fired).
     # Propagated into the final dict so the dashboard sees the flag
@@ -786,8 +787,10 @@ def _validate_zip_streaming(*, api: HfApi, dataset: str, token: str | None,
                         asset["rel_path"] = asset_rel
                 zip_results = rj.get("results", [])
                 zip_layout = rj.get("layout_findings") or []
+                zip_preliminary = rj.get("preliminary_findings") or []
                 merged_results.extend(zip_results)
                 merged_layout.extend(zip_layout)
+                merged_preliminary.extend(zip_preliminary)
                 if rj.get("preliminary_check_failed"):
                     any_preliminary_check_failed = True
                 out(f"    {len(zip_results)} asset(s); rc={rc}")
@@ -900,6 +903,7 @@ def _validate_zip_streaming(*, api: HfApi, dataset: str, token: str | None,
         "schema_version": 1,
         "results": merged_results,
         "layout_findings": merged_layout,
+        "preliminary_findings": merged_preliminary,
         "preliminary_check_failed": any_preliminary_check_failed,
         "profile_coverage": {},
         "streaming_zips": len(zip_entries),
@@ -918,12 +922,28 @@ def _summarize(results_json: dict) -> tuple[str, str]:
     # knows what to do (forward the report to the partner; address
     # these before re-validating to surface deeper USD findings).
     if results_json.get("preliminary_check_failed"):
-        violations = len(results_json.get("preliminary_findings")
-                         or results_json.get("layout_findings") or [])
+        # Count actual issues by summing across results — robust to
+        # whichever sidecar field the validator populated.
+        violations = sum(len(r.get("issues") or [])
+                          for r in (results_json.get("results") or []))
+        if violations == 0:
+            # Fall back to the sidecar list when results is empty
+            # (shouldn't happen, defensive).
+            violations = len(results_json.get("preliminary_findings")
+                             or results_json.get("layout_findings") or [])
         files_affected = len(results_json.get("results") or [])
-        return "fail", (f"PRELIMINARY CHECK FAILED — {violations} foundation-spec "
-                        f"issue(s) across {files_affected} file(s); address these "
-                        f"before re-validating")
+        # Per-code breakdown for the chip text — the partner-facing
+        # summary is more useful when it names the failing rules.
+        code_counts: dict[str, int] = {}
+        for r in (results_json.get("results") or []):
+            for iss in (r.get("issues") or []):
+                c = iss.get("code") or "UNKNOWN"
+                code_counts[c] = code_counts.get(c, 0) + 1
+        top_codes = sorted(code_counts.items(), key=lambda kv: -kv[1])[:3]
+        codes_text = ", ".join(f"{c} ×{n}" for c, n in top_codes) if top_codes else "0 issues"
+        return "fail", (f"PRELIMINARY CHECK FAILED — {codes_text} "
+                        f"({files_affected} file(s) affected). Address these "
+                        f"before deeper validation runs.")
     counts = {"error": 0, "failure": 0, "warning": 0}
     total = len(results_json.get("results", []))
     failed = 0