Weekly-pipeline cleanup: remove deprecated workflow and follow-ups

.github/workflows/push-to-clickhouse.yaml

@@ -14,10 +14,13 @@
 # and a separate ingest one.
 #
 # At top_k=10000 per mode, a single job processing the whole list serially
-# would take far too long, so the fetch itself IS split out: generate-matrix
-# fetches the top-K list once (no Spyre hardware needed), shards it, and
-# weekly-model-scan fans out over those shards in parallel (mirrors
-# torch-spyre's generate_matrix -> matrix job pattern).
+# would take far too long, so the fetch is done once up-front and the actual
+# scan is fanned out: the generate-shards job fetches the top-K list once
+# on ubuntu-latest, splits it into per-tier shard files, uploads them as
+# the `weekly-shards` artifact, and emits matrix_x1 / matrix_x2 / matrix_x4
+# JSON pointing at those shard files. The three weekly-model-scan-x{1,2,4}
+# jobs then fan out across the Spyre runners, each downloading only its
+# own shard.
 # =============================================================================
 name: push-to-clickhouse
 
@@ -73,10 +76,9 @@ jobs:
   # job below. No Spyre hardware needed — utils/hf_model_catalog.py's import
   # chain (via hf_adapters/auto_spyre_model.py) has no module-level
   # torch_spyre dependency; it's lazily imported inside a device-guarded
-  # function body in hf_common.py. The now-deprecated test_weekly_DEPRECATED
-  # .yaml is existing proof this exact fetch path runs fine on ubuntu-latest.
+  # function body in hf_common.py.
   # ---------------------------------------------------------------------------
-  generate-matrix:
+  generate-shards:
     name: Generate weekly shards
     runs-on: ubuntu-latest
     # At top_k=10000+, fetching 2x oversampled raw candidates per mode (each
@@ -184,14 +186,14 @@ jobs:
   # 10*1 + 4*2 + 1*4 = 22 cards (~2 nodes), so PR/daily CI isn't starved.
   weekly-model-scan-x1:
     name: Weekly model scan x1 (${{ matrix.mode }} shard ${{ matrix.shard_index }})
-    needs: generate-matrix
+    needs: generate-shards
     runs-on: [x86_64, spyre_pf_x1, linux, image_torch_spyre]
     timeout-minutes: 4320
     strategy:
       fail-fast: false
       max-parallel: 20 # 20 shards x 1 card = 20 cards
       matrix:
-        include: ${{ fromJSON(needs.generate-matrix.outputs.matrix_x1) }}
+        include: ${{ fromJSON(needs.generate-shards.outputs.matrix_x1) }}
     env:
       CLICKHOUSE_HOST: ${{ secrets.CLICKHOUSE_HOST }}
       CLICKHOUSE_PORT: ${{ secrets.CLICKHOUSE_PORT }}
@@ -261,14 +263,14 @@ jobs:
 
   weekly-model-scan-x2:
     name: Weekly model scan x2 (${{ matrix.mode }} shard ${{ matrix.shard_index }})
-    needs: generate-matrix
+    needs: generate-shards
     runs-on: [x86_64, spyre_pf_x2, linux, image_torch_spyre]
     timeout-minutes: 4320
     strategy:
       fail-fast: false
       max-parallel: 4 # 4 shards x 2 cards = 8 cards
       matrix:
-        include: ${{ fromJSON(needs.generate-matrix.outputs.matrix_x2) }}
+        include: ${{ fromJSON(needs.generate-shards.outputs.matrix_x2) }}
     env:
       CLICKHOUSE_HOST: ${{ secrets.CLICKHOUSE_HOST }}
       CLICKHOUSE_PORT: ${{ secrets.CLICKHOUSE_PORT }}
@@ -338,14 +340,14 @@ jobs:
 
   weekly-model-scan-x4:
     name: Weekly model scan x4 (${{ matrix.mode }} shard ${{ matrix.shard_index }})
-    needs: generate-matrix
+    needs: generate-shards
     runs-on: [x86_64, spyre_pf_x4, linux, image_torch_spyre]
     timeout-minutes: 4320
     strategy:
       fail-fast: false
       max-parallel: 1 # 1 shard x 4 cards = 4 cards
       matrix:
-        include: ${{ fromJSON(needs.generate-matrix.outputs.matrix_x4) }}
+        include: ${{ fromJSON(needs.generate-shards.outputs.matrix_x4) }}
     env:
       CLICKHOUSE_HOST: ${{ secrets.CLICKHOUSE_HOST }}
       CLICKHOUSE_PORT: ${{ secrets.CLICKHOUSE_PORT }}

tests/spyre/weekly_generation/clickhouse_db.py

@@ -27,7 +27,7 @@
 """
 
 import os
-from datetime import date
+from datetime import date, datetime
 from pathlib import Path
 
 import clickhouse_connect
@@ -179,9 +179,38 @@ def _parse_bool(value: str) -> bool:
     return value.strip().lower() in ("1", "true", "yes")
 
 
+# Accepted date input formats. ISO first (canonical), then European day-first
+# variants we've seen in exported CSVs. We deliberately do NOT accept ambiguous
+# US month-first strings — a slash-separated date is always day-first here.
+_DATE_INPUT_FORMATS: tuple[str, ...] = (
+    "%d/%m/%Y",
+    "%d-%m-%Y",
+    "%d.%m.%Y",
+)
+
+
+def _parse_date(value: str) -> date:
+    """Parse *value* into a ``date``, tolerating DD/MM/YYYY as well as ISO.
+
+    Raises ``ValueError`` if the input doesn't match any accepted format,
+    so import_csv's existing try/except still reports the row as malformed.
+    """
+    v = value.strip()
+    try:
+        return date.fromisoformat(v)
+    except ValueError:
+        pass
+    for fmt in _DATE_INPUT_FORMATS:
+        try:
+            return datetime.strptime(v, fmt).date()
+        except ValueError:
+            continue
+    raise ValueError(f"unrecognized date format: {value!r}")
+
+
 def _parse_nullable_date(value: str | None) -> date | None:
     v = (value or "").strip()
-    return date.fromisoformat(v) if v else None
+    return _parse_date(v) if v else None
 
 
 def _parse_nullable_str(value: str | None) -> str | None:
@@ -212,7 +241,7 @@ def import_csv(sink, csv_path: str) -> tuple[int, int]:
                 )
                 continue
             try:
-                snapshot_date_val: date = date.fromisoformat(snapshot_raw)
+                snapshot_date_val: date = _parse_date(snapshot_raw)
             except ValueError:
                 malformed += 1
                 print(