Standard 12-1 cross-sectional momentum has decayed in DM large caps but remains economically and statistically significant in EM small caps after costs, with a roughly 4–6% annualized spread. The decay does not look like a Bayesian update toward zero — it looks like a liquidity-tier story.

Reading notes on the 2024 update to 'The Devil in HML's Details'. The headline result — that book/price using current-month price recovers most of the value premium — survives, but the construction choices that get you there are not innocent.

When backtests improve from vol-targeting, the gain is almost entirely from changing the effective leverage profile across regimes, not from any predictive content in realized vol. This matters because the cost model people use for the unlevered strategy is wrong for the levered one.

Post-earnings announcement drift weakened sharply in 2023–24 but the cross-sectional dispersion of the surprise→drift slope widened. A pooled coefficient hides a regime-conditional one. Decomposition by analyst-coverage tercile recovers most of the lost spread.

Most published microcap anomalies are real; almost none survive a realistic implementation simulation that respects participation caps and price impact. We build a tier-aware backtest harness and show that the implementable frontier is roughly 30–40% the size of the paper one.