fix(research): honor configured fib scale_mode in level-event detection (PR #33 P2)

Repo configs default fib.scale_mode to "log", but run_level_events,
_aggregate_leg_events and collect_candidates called detect_level_events (and the
row-price helper) without scale_mode, so they scanned/reported LINEAR levels —
different prices than the log-scale protocol and charts.

Thread settings.fib.scale_mode through all three call sites and
_level_rows_from_swing. Adds a regression test asserting the level ladder is log.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: JohnCCarter

src/fibengine/research/human_review_rows.py

@@ -75,8 +75,10 @@ def _relation_for_bar(df: pd.DataFrame, bar_idx: int, price: float) -> str:
     )
 
 
-def _level_rows_from_swing(swing: Swing, ratios: list[float]) -> list[dict]:
-    prices = fib_levels(swing, ratios)
+def _level_rows_from_swing(
+    swing: Swing, ratios: list[float], scale_mode: str = "linear"
+) -> list[dict]:
+    prices = fib_levels(swing, ratios, scale_mode=scale_mode)
     return [
         {"ratio": f"{ratio:g}", "price": round(float(prices[ratio]), 6)} for ratio in sorted(prices)
     ]
@@ -222,8 +224,15 @@ def collect_candidates(
         else:
             lo = swing.end.index
             hi = n
-        streams = detect_level_events(df, swing, level_cfg, ratios, settings.pivots.atr_period)
-        level_rows = _level_rows_from_swing(swing, ratios)
+        streams = detect_level_events(
+            df,
+            swing,
+            level_cfg,
+            ratios,
+            settings.pivots.atr_period,
+            scale_mode=settings.fib.scale_mode,
+        )
+        level_rows = _level_rows_from_swing(swing, ratios, settings.fib.scale_mode)
         for stream in streams:
             for ev in stream.events:
                 if lo <= ev.bar_index < hi:

src/fibengine/research/level_events.py

@@ -233,7 +233,12 @@ def run_level_events(
     streams: list[LevelInteractionStream] = []
     if swing is not None:
         streams = detect_level_events(
-            df, swing, level_cfg, settings.fib.levels, settings.pivots.atr_period
+            df,
+            swing,
+            level_cfg,
+            settings.fib.levels,
+            settings.pivots.atr_period,
+            scale_mode=settings.fib.scale_mode,
         )
 
     record = {
@@ -320,7 +325,14 @@ def _aggregate_leg_events(
         # Attributionsfönster [lo, hi) för denna leg.
         lo = t if non_overlapping else swing.end.index
         hi = legs[i + 1][0] if (non_overlapping and i + 1 < len(legs)) else n
-        streams = detect_level_events(df, swing, level_cfg, ratios, settings.pivots.atr_period)
+        streams = detect_level_events(
+            df,
+            swing,
+            level_cfg,
+            ratios,
+            settings.pivots.atr_period,
+            scale_mode=settings.fib.scale_mode,
+        )
         leg_events = 0
         for stream in streams:
             kept = [e for e in stream.events if lo <= e.bar_index < hi]

tests/research/test_human_review_level_events.py

@@ -186,6 +186,36 @@ def test_collect_candidates_single_mode_uses_selected_swing(monkeypatch):
     assert all(r["event_bar"] > 40 for r in rows)
 
 
+def test_collect_candidates_honors_configured_log_scale(monkeypatch):
+    """Regression (PR #33 P2): collect_candidates must thread settings.fib.scale_mode
+    into the detector and the level ladder, not silently fall back to linear."""
+    from fibengine.core.fib import fib_levels
+    from fibengine.research.human_review_rows import decode_levels
+
+    df = _trend_df()
+    settings = _settings()
+    assert settings.fib.scale_mode == "log"  # repo default
+    swing = _up_swing(df)
+    monkeypatch.setattr(hr, "select_swing", lambda _df, _p, _s: swing)
+    rows = collect_candidates(df, settings, mode="single")
+    assert rows
+
+    ratios = settings.fib.levels
+    log_ladder = {
+        f"{r:g}": round(float(p), 6) for r, p in fib_levels(swing, ratios, scale_mode="log").items()
+    }
+    lin_ladder = {
+        f"{r:g}": round(float(p), 6)
+        for r, p in fib_levels(swing, ratios, scale_mode="linear").items()
+    }
+    assert log_ladder != lin_ladder  # scales must differ for this swing, else the test is moot
+
+    for r in rows:
+        decoded = {d["ratio"]: d["price"] for d in decode_levels(r)}
+        assert decoded == log_ladder  # ladder honors log scale, not linear
+        assert r["fib_price"] == log_ladder[r["fib_level"]]
+
+
 def test_end_to_end_package_files_created(monkeypatch, tmp_path):
     df = _trend_df()
     settings = _settings()