Round 2: reverting power law and cycle-on-powerlaw tests; one-time holdout run.

Add TrendReversionVol: deviations from the power-law trend follow a daily
AR(1), so uncertainty levels off, optionally plus trend-parameter
uncertainty with an autocorrelation-adjusted effective sample size.

Two experiments, run under the unchanged verdict rule:
- powerlaw-ou: +21% to +45% vs powerlaw at 2-4 years, but slightly negative
  point estimates at 1 month make it inconclusive.
- cycle-on-powerlaw: inconclusive (+18% at 2 years, negative elsewhere).

The holdout (outcomes after 2024-11-26) was scored once, for the four
candidates fixed beforehand. powerlaw is the best long-horizon forecast
(+45% and +58% vs the random walk at 2 and 3 years); nothing beats the
random walk inside a year; cycle fails badly. Results are in the README.

btcmodel/experiments.py

@@ -24,7 +24,7 @@ from .models.drift import (
     TrailingMeanDrift,
 )
 from .models.shape import Empirical, StudentT
-from .models.volatility import CycleVol, EwmaVol, ReversionVol, TrailingVol
+from .models.volatility import CycleVol, EwmaVol, ReversionVol, TrailingVol, TrendReversionVol
 
 
 @dataclass(frozen=True)
@@ -64,6 +64,7 @@ def verdict(variant_summary: pd.DataFrame) -> str:
 # Idea labels (D1, V2, ...) refer to docs/2024-ideas.md.
 CYCLE = MODELS["cycle"]
 DRIFT_RW = MODELS["drift_rw"]
+POWERLAW = MODELS["powerlaw"]
 # Volatility and shape experiments use drift_rw as the control: the zero-drift
 # random walk is biased low at long horizons, so anything that merely widened
 # its intervals would look like an improvement.
@@ -138,5 +139,31 @@ EXPERIMENTS: dict[str, Experiment] = {
             CYCLE,
             (Composite("cycle_fraction", CycleDrift(phase="fraction"), TrailingVol()),),
         ),
+        # Round 2. Before running these, the holdout candidates were fixed as:
+        # the three original models, powerlaw, and any variant below that is
+        # "better" than its control.
+        Experiment(
+            "powerlaw-ou",
+            "deviations from the power-law trend fade, so long-horizon uncertainty is bounded",
+            "follow-up to diminishing-returns: powerlaw_revert beat drift_rw, but its bands"
+            " were too wide",
+            POWERLAW,
+            (
+                Composite("powerlaw_ou", PowerLawDrift(revert=True), TrendReversionVol()),
+                Composite(
+                    "powerlaw_ou_param",
+                    PowerLawDrift(revert=True),
+                    TrendReversionVol(parameter_uncertainty=True),
+                ),
+            ),
+        ),
+        Experiment(
+            "cycle-on-powerlaw",
+            "with the level set by the power law, the cycle's timing adds information",
+            "D1 on D3. This pair was already compared informally on the same data after"
+            " round 1, so only the holdout can really settle it",
+            POWERLAW,
+            (Composite("cycle_on_powerlaw", PowerLawScaledCycleDrift(), TrailingVol()),),
+        ),
     )
 }