A transparent arithmetic example
Assume, only for illustration, a two-point conversion succeeds 52.5% of the time and a one-point attempt succeeds 94% of the time. Ignoring other outcomes, the expected points are 2 × 0.525 = 1.05 versus 1 × 0.94 = 0.94. These are hypothetical probabilities, not league averages, team-specific estimates or live Pick6Labs outputs.
This calculation answers which option yields more expected immediate points under those assumptions. It does not yet answer which option most improves the chance of winning a particular game. Treating the two questions as interchangeable discards the game state that makes the decision interesting.
The state after the attempt matters
The score, time remaining, expected possessions and opponent behavior affect the value of different outcomes. A successful two-point attempt, a failed attempt and a made one-point kick can leave strategically different paths. Overtime assumptions can matter as well and must match the relevant rules and season.
A decision model would assign a win probability to each resulting state and combine them with the probabilities of reaching those states. That requires more inputs than the short expected-points calculation above. Without them, an exact universal coaching rule would be overconfident.
Keep outcome evaluation separate from decision evaluation
If a choice was reasonable using the information available at the time, a failed attempt does not automatically make it a bad decision. Conversely, a successful attempt does not validate every probability estimate used to justify it. Evaluating decisions from results alone rewards luck and hides poor assumptions.
For an audit, record the state and probabilities before the decision, explain the source or assumptions, then preserve the later result separately. Do not replace the earlier estimate with one informed by the observed outcome. This is the same reason pregame and in-play prediction records need distinct timestamps and purposes.
How to read this on an analytics site
A simulation may explore a coaching scenario without having a verified, team-specific decision model. Labeling an explanation as illustrative is important. A qualitative AI assessment should not be mistaken for a measured win-probability-added value unless the relevant calculation and inputs actually exist.
This article establishes no measured conversion advantage for any coach and no claim that Pick6Labs optimizes every fourth-down or conversion decision. It gives a reproducible way to distinguish an expected-value example from a game-specific forecast, and a checklist for the additional evidence the latter would require.