Initial commit: KQL ↔ SDL PowerQuery proof of equivalence

harness/check_ts_collisions.py

@@ -0,0 +1,34 @@
+#!/usr/bin/env python3
+"""Count duplicate timestamps within the generated JSONL.
+
+SDL appears to dedupe addEvents by (session, ts) - events sharing a ts
+within the same session are silently dropped. If our generator emits many
+events at colliding ts_epoch_ms values, only one of each cluster survives.
+"""
+import json
+from collections import Counter, defaultdict
+from pathlib import Path
+
+JSONL = Path(__file__).resolve().parents[1] / "sample_data" / "events.jsonl"
+
+per_type_total = Counter()
+per_type_unique = defaultdict(set)
+per_type_max_collision = defaultdict(int)
+with JSONL.open() as f:
+    for line in f:
+        r = json.loads(line)
+        et = r["event_type"]
+        ts = r["ts_epoch_ms"]
+        per_type_total[et] += 1
+        per_type_unique[et].add(ts)
+
+print(f"{'event_type':30s} {'events':>8} {'uniq_ts':>8} {'collision_loss%':>16}")
+print("-" * 70)
+for et in sorted(per_type_total):
+    n = per_type_total[et]
+    u = len(per_type_unique[et])
+    loss = 100 * (n - u) / n if n else 0
+    print(f"{et:30s} {n:>8} {u:>8} {loss:>15.1f}%")
+print("-" * 70)
+print(f"{'TOTAL':30s} {sum(per_type_total.values()):>8} "
+      f"{sum(len(s) for s in per_type_unique.values()):>8}")