Raw 12-factor signal points one way; regime-adapted composite points the other. A single regime classification is doing the steering — caution warranted.
When the equal-weighted (or prior-weighted) factor composite would suggest BUY but the regime-adapted composite — using the weight vector calibrated for the current HMM regime — drops by 20+ points to MIXED or SELL, the engine is essentially saying "company fundamentals look fine but the regime makes me distrust the bullish factors right now". The reverse case (raw says SELL, adapted says BUY) is rarer but symmetric: company looks weak, regime tailwind compensates.
Hamilton 1989 showed that financial-market regimes persist but classifications are noisy at boundaries. Asness-Moskowitz-Pedersen 2013 documented that factor weights should rotate with regime — momentum thrives in risk-on, quality/low-vol in risk-off, both struggle in transition. When the rotation drives a multi-tier verdict swing, the bullish thesis is structurally exposed to a single classification. Da-Engelberg-Gao 2011 found that "narrative vs technicals" divergences predict near-term volatility — the regime divergence here is a quantitative variant of the same phenomenon.
Position-size at HALF the size you would on an aligned-tier ticker. Watch for regime confirmation (VIX trend + market breadth + yield-curve shape) before scaling up. If the regime classification has been near a boundary (BOCPD posterior entropy high) the divergence may resolve in either direction within a week — wait it out rather than acting on the spread.