diff --git a/README.md b/README.md
index 252551e..1219d8c 100644
--- a/README.md
+++ b/README.md
@@ -70,7 +70,7 @@ The same day, a real NVDA earnings guidance 8-K produced conviction=0.38 with a
 
 ---
 
-## Research methodology
+## Research methodology 
 
 Eleven factors were tested using HLZ multiple-testing correction at M=400 (Bonferroni threshold |t| >= 3.78). Every null result is documented in `docs/progress/` with full IC tables, regime splits, and bootstrap confidence intervals. An earlier backtest produced CAGR +14.3% and Sharpe 0.554, which was explicitly rejected after discovering that full-sample factor weights were applied at every rebalance -- approximately five years of look-ahead at the start of the window. The corrected result is CAGR +8.72%, Sharpe 0.488. The Phase 6 numbers do not appear anywhere in the current codebase as valid results. Three factors (momentum_12_1, size_proxy, gnn_drift) were flagged as declining by the trailing IC decay detector and removed from the aggregator before the corrected backtest was run.