H2H records collect previous meetings between two sides. Squads turn over and managers change, so a 4–0 from three seasons ago describes teams that no longer exist — as a predictor of the next result, H2H history adds almost nothing once current strength is accounted for.
The defensible uses are aggregates: pairings that persistently run high or low on goals across many meetings, or venue-specific patterns with real sample sizes. Context, not signal — our model ignores H2H entirely and works from current scoring rates.
Ten meetings averaging 3.4 goals against a league norm of 2.6 is worth a look at the totals line — as a prompt to check current attacking rates, not as a bet. Three meetings averaging anything is a coin remembered fondly.