March 2026 was a more difficult month on both accuracy and return. We placed 43 bets across 17 leagues and 75 teams, finishing down $240.46 with a 20.9% hit rate. Compared with February (68 bets, -$95.34), the volume fell again while league coverage widened, and the outcomes never really caught a stable rhythm. With a bankroll-linked flat stake, the unit size adjusted as intended, but it does not change the simple fact that a month with fewer wins will leave a clear mark.
The most usable signal, again, came from BTTS_YES. We hit 4 of 10 (40%), which is not a standout number in isolation, but it continues to look like the market where our modelling assumptions translate most cleanly into actionable edges. At league level, the Europa League went 4 from 5 (80%). It is the kind of spike that can be either insight or variance, and the sensible view is to log it as a watchlist item rather than a conclusion, particularly given the small sample and the different incentive structures that come with knockout football.
The main lesson from March is about where we ask the model to do its hardest work. Spreading 43 bets across 17 leagues is broad for the month’s volume, and it increases the chance that we are taking thinner, less repeatable edges to keep coverage high. Going into April, the process priority is to tighten the filter: lean more into markets where we repeatedly see model-to-market gaps (BTTS_YES remains the clearest candidate), and be more cautious about expanding league count unless we can show that the edge appears frequently and survives normal variance.