backtest: replay the LIVE track only (v1) — shadow research tracks poisoned /backtest

replay_window reconstructed fires from ALL setup versions, so liq1/vr1/run1/pm1
(refuted shadow tracks, never traded live, 8:1 volume vs v1, tight-stop
artifacts) dominated /backtest: 17k trades / 45 sessions at -1.8R/trade —
garbage scale vs the harness-validated -0.08R/fire. reconstruct_fires_from_setups
gains an optional setup_version filter (default None: experiments unchanged);
replay_window pins it to v1.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

tests/unit/api/test_backtest_endpoint.py

@@ -225,3 +225,22 @@ def test_backtest_route_serves_html():
 
 if __name__ == "__main__":  # pragma: no cover
     pytest.main([__file__, "-q"])
+
+
+def test_replay_window_replays_live_track_only():
+    """/backtest must judge the strategy that actually trades: setup_version v1.
+    Shadow research tracks (liq1/vr1/run1/pm1 -- refuted, never traded live)
+    share setup_events and would otherwise dominate the replay 8:1 with
+    tight-stop artifacts (the 2026-07-17 -1.8R/trade garbage-scale bug)."""
+    store = _seed_store()
+    # a liq1 shadow fire on the same session/bars -- must NOT be replayed
+    store.con.execute(
+        "INSERT INTO setup_events (setup_id, symbol, session_date, setup_time, "
+        "setup_name, setup_version, entry_reference_price, invalidation_price) "
+        "VALUES (?,?,?,?,?,?,?,?)",
+        [f"liq-AAA-{SESS1}", "AAA", SESS1,
+         datetime(SESS1.year, SESS1.month, SESS1.day, 13, 35),
+         "liquid_breakout", "liq1", 10.0, 9.0])
+    from research.experiments.replay import replay_window
+    out = replay_window(store, str(SESS1), str(SESS2))
+    assert len(out["trades"]) == 4  # the 4 v1 fires only; liq1 excluded