We Tested Eleven Ways to Beat a Survivor Pool Market — Here Is What We Found

Everybody in this space publishes the angle that worked. Almost nobody publishes the ten that did not, which is a problem, because a list of only the winners is not evidence — it is marketing.

So here is the whole register. Eleven survivor-pool theories, each tested against real historical NFL seasons using real bookmaker odds, with the sample sizes attached. Ten of them are unusable. The one that survived is smaller and less exciting than the pitch you have probably read elsewhere.

How the tests were run

Every test starts from the same baseline: de-vigged market win probability. The question is never merely does this factor correlate with winning — favourites correlate with winning, that tells you nothing. The question is whether the factor predicts anything the closing price has not already priced in.

Windows range from 2006-2025 for schedule and venue factors down to 2021-2025 where the odds backfill limits us. Sample sizes run from 183 targeted games to 5,295. The full arithmetic is on the methodology page.

The eight that came back flat

Rest differential, divisional games, home underdogs, coaching head-to-head, surface & altitude, trap games, defensive line strength differential and primetime risk. Every one of these is a staple of survivor advice columns. Every one of them tested at z below 1.17 against the market — statistical noise.

• Rest differential — No detectable edge. 5,295 games
• Divisional games — Flat. 5,295 games
• Home underdogs — Flat. 5,295 games
• Coaching head-to-head — Flat. 5,295 games
• Surface & altitude — Flat. 5,295 games
• Trap games — Flat. Full odds backfill
• Defensive line strength differential — Season-to-date net sack differential is strongly collinear with the market's own pricing (r = 0.509), so a raw test would look positive for the wrong reason. 2,052 team-weeks (4-game minimum rolling window)
• Primetime risk — Flat. 1,230 primetime games / 4,051 non-primetime games

The honest interpretation is not that these effects are imaginary. Several are real football phenomena. They are simply already in the number, because thousands of people with money at stake put them there before you got to the board.

The one that failed twice

Confidence scoring — the idea that you can rank favourites by how safe they are — failed in two separate builds. Version one used market and ELO agreement and came back inverted: the games our model called low-confidence won 76.9% of the time, because the market prices roster news that an ELO rating simply cannot see. Version two used bookmaker dispersion and book count and came back non-monotonic.

Neither was shipped. A confidence number that does not rank correctly is worse than no confidence number, because it looks authoritative. The full account of both attempts is in the confidence scoring writeup.

The one that is real and still useless

Referee crews are the most interesting failure in the register, because the effect is genuinely there. Across 3,024 games and 27 head referees, penalty-rate variation between crews is significant at p = 0.0014. Crews really do call games differently.

And it does not help you. Every market-facing angle — favourite win rate, home win rate, totals, over rate — came back flat after correcting for multiple comparisons. A real effect you cannot bet is still a real effect, and we hold to that distinction rather than dressing it up. The full test is in the referee crew writeup, and you can browse every crew profile in Explore.

The one that survived

That is the finding. Not a system, not a model, not an edge over the market — a timing rule. When a game involves a team with nothing left to play for, late-week information matters unusually much, and the price you see on Tuesday is measurably worse than the price you see on Sunday.

The practical version fits in one line: submit your pick as late as your pool allows, and care about it most in the last month of the season. The methodology, control group and numbers behind it are in the late-week timing writeup.

What this means for how you actually play

If ten of eleven angles are dead, what is left? Discipline. The betting market is very hard to beat and surviving a full season is rare — our simulator puts full-season survival near 1.2% per season, and zero of the five replayed seasons from 2021 to 2025 survived under greedy picking.

The value is not in a secret factor. It is in spending teams in the right order, reserving scarce good options for the weeks that have none, sizing your leverage to your pool, and not submitting before the news is in. That is unglamorous and it is what the data supports.

Every finding above, with hypothesis, result, sample size and test date, is listed in the Edge Lab. If you are new to survivor pools, start with the complete strategy guide instead.