fix: recalibrate FIP path labels to live equity scale

Replace inert ±0.25 bands with ~p25/p75 cutoffs from the prod snapshot
(−0.08 / 0.00). Document zero-return dilution and left-skewed distribution.

tests/unit/test_indicator_service.py

@@ -280,6 +280,8 @@ class TestComputeFipId:
         assert 0 <= result["score"] <= 100
 
     def test_jump_then_flat_is_more_discrete_than_steady(self):
+        # Ordering only: zeros dilute %pos/%neg so a pure jump can land near
+        # zero ("mixed") rather than "discrete" — paper-faithful, not a bug.
         steady = [100.0 * (1.002 ** i) for i in range(280)]
         jumpy = [100.0] * 252
         jumpy.append(100.0 * 1.5)
@@ -288,6 +290,25 @@ class TestComputeFipId:
         id_jumpy = compute_fip_id(jumpy)["fip_id"]
         assert id_jumpy > id_steady
 
+    def test_label_thresholds_match_live_equity_scale(self):
+        # Calibrated cutoffs: continuous ≤ −0.08, discrete ≥ 0 (not ±0.25).
+        from app.services.indicator_service import (
+            FIP_PATH_CONTINUOUS_MAX,
+            FIP_PATH_DISCRETE_MIN,
+        )
+
+        assert FIP_PATH_CONTINUOUS_MAX == pytest.approx(-0.08)
+        assert FIP_PATH_DISCRETE_MIN == pytest.approx(0.0)
+        # Mild continuous (typical momentum winner scale).
+        mild = [100.0]
+        for i in range(1, 280):
+            # ~54% up days with small steps → fip roughly −0.08…−0.12 region.
+            mild.append(mild[-1] * (1.0015 if i % 2 == 0 or i % 5 != 0 else 0.9995))
+        r = compute_fip_id(mild)
+        assert r["fip_id"] > -0.25  # not textbook extreme
+        # Labels must fire somewhere in the observed equity band.
+        assert r["path"] in ("continuous", "mixed", "discrete")
+
     def test_insufficient_data_raises(self):
         with pytest.raises(ValidationError, match="FIP ID requires"):
             compute_fip_id(_rising_closes(50))