zeroluck

NFL bets that survived the luck test.

BETAzeroluck is in beta and we are still updating it. Data may change while we finish building it.

Most winning trends are luck. We bet the ones that aren't.

Anyone can find a trend that went 12-3. Dig through enough past games and thousands of them turn up by pure chance. zeroluck bets only when dozens of rules, built different ways, land on the same side, and every one of them has already survived a luck test. Our deep-search rules were pitted against scrambled results, round after round; only the searches that beat the noise made the cut.

Every rule is sealed with a date before the games it bets. Every bet is recorded, wins and losses.

36.2 million rules tested in one search alone, and a fair coin would have found more winners. A winning record proves nothing here. Surviving the luck tests does.

Rules tested, in one search36.2 millionjust our August search, one of several; every rule kept on file, bet or not

Looked like winners7.2 millionin that one search, but a fair coin would give 7.75 million: on its own, a winning record proves nothing

Rules that made the cut2,161Deep rules come only from searches that beat scrambled results, strong winning margins, a luck number on every crew rule, and every rule dated before the games it bets

2026 game lines—weeks 1-6 as today's rules pick them; every bet is listed below

2026 game lines, weeks 1-6

Every pick, win or lose.

Week	Game	Bet	Strength	Rules for / against	Result (final score)

Four ways we kill luck

Locked before kickoff

A rule is written down and dated before the games it bets, and is never changed after; a change makes a new, dated rule.

A number on luck

We shuffle the results hundreds of times and count how often the shuffled games do as well as the rule. Lower is better.

Noise vs. noise

Our deep search was also run on scrambled results. Of 90 rounds, 34 found more than the noise did and were kept; the 1,002 rules from the other 56 were thrown out.

Consensus or nothing

A bet needs 20 or more of our rules on its side and almost nothing against it; with fewer, it is not a bet. Every bet is STRONG or STANDARD.

This week

See this week's card How we beat the noise

Nothing here is betting advice.

This week

Every bet this week, strongest first. Every bet is STRONG or STANDARD; some also carry the PUBLIC PRIMETIME tag.

When a bet goes up: 24 hours before its kickoff.
When it locks: at the last check before kickoff, an hour or two before the game. Its line is fixed then and never changed.
Pulled: a bet already up that stops qualifying before it locks is marked PULLED, with the reason, and is never placed.

Check back on game days. New bets go up 24 hours before each kickoff, and the card is rechecked several times a week, the last time an hour or two before each game.

Kickoff (ET)	Game	Pick	Strength	Best angle (luck)	Angle	Best subangle	Result

Track record

How each strength has done, season by season:

Season

How to read the card

Strength

Every bet has 20+ of our rules behind it, at least one a crew rule, and almost nothing against it.

STRONG: 50+ rules agree. Not one disagrees.
STANDARD: 20+ rules agree. No crew rule disagrees, and at most 2 others do.

PUBLIC PRIMETIME: a tag, not a bet of its own. A prime-time pattern that fades the public's side, shown only when several of our other rules already make that bet.

Luck number

The odds that pure chance made the record. We shuffle the results hundreds of times and count how often the fakes do as well. Lower is better.

Track record

Wins-losses-pushes, hit rate and ROI at -110. Angle: the angle's record on past games. Best subangle: its best single rule's. Both show the average points past the line in wins.

Angles

An angle is a family of rules. Every rule with 15+ past games gets its own number, like #5.13. Best subangle is the rule on the bet with the best past hit rate (8+ games).

Not advice

A research project. Nothing here is betting advice.

The story

Most betting records are noise. Here is how we know, what we do about it, and everything we threw away.

The problem: winning records are cheap

Give 1,000 people a coin and have each one call 20 flips. About 410 of them will call 11 or more right, and about 20 will go 15-5 or better. None of them can predict a coin. Betting trends work the same way. Search enough past games, by team, weather, rest, line, public money, and thousands of rules will look brilliant by chance alone. Most sites show you those rules and stop there.

Rules tested, in one search36.2 millionjust our August search, one of several; every rule kept on file, bet or not

Winning record on past games7.2 millionin that one search: rules with 20+ games, beating the 52.4% break-even

Coin flips would produce7.75 millionwinning records from a fair coin over the same games: more than we found

Hit 60%+ on past games2.0 milliona fair coin gives 2.27 million

So in our own research, a winning record, even a 60% one, says almost nothing by itself. We checked that directly: we re-tested about 115,000 cells on games nobody used to pick them. One cell from each of 5,427 families scattered exactly the way pure chance does (1.006, against 1.000 for chance), and cells that hit 65% or better on their past games hit about 50% afterwards. A real 55% rule needs about 3,500 bets before its own record can prove it. Counts as of 2026-10-14.

Our answer

Seal the rule first. Every rule is written down and dated before the games it bets, and is not changed after that; a change is a new, dated rule.

Measure the luck. Every crew rule carries its own luck number (deep-search rules are judged as whole searches, below): the chance that a record this good came from random results. We shuffle the results hundreds of times and count how often the shuffled version does as well.

Test the test. A luck test is only as good as its proof. So we plant a fake edge in the data and make sure the test catches it, then run the entire search on scrambled results. If scrambled games produce as many "great" rules as real ones, the great rules mean nothing. Our deep search ran 90 rounds this way: 34 beat the noise and their 1,311 rules were kept; 56 did not, and their 1,002 rules were thrown out, good-looking records and all. The crew rules then faced the same check and, as a group, found about six times what noise did.

Bet only when many rules agree. We bet with 2,161 rules built from different angles, so each one carries a different piece of the edge. One rule can get lucky; dozens pointing the same way is a different story. A game-side is a bet only when at least 20 rules pick it, at least one of them a crew rule, and almost nothing picks the other side (no crew rule and at most 2 others). Its strength, STRONG or STANDARD, says how strong the agreement is.

The graveyard

Plenty. Besides the 1,002 deep-search rules thrown out with the 56 rounds that did not beat the noise, we tested whether a list of 210 group picks could make bets on its own. Every version failed our bar: the agreement versions lost at about the bookmaker's cut or worse, and the rest did not beat scrambled results or had no recent record. So the list does not make bets. It can only stop one: a side the 210 list picks against, or splits on, is not a bet. We keep every test we run, pass or fail, so a failure cannot quietly disappear.

Most sites	zeroluck
Show a record	Show the record and the chance luck alone made it
Count the games used to find the trend	Keep those games apart. The real test is games the rule never saw: for us, the live record from 2026
Never test the method itself	Run the same search on scrambled results first (null calibration). If shuffled games produce as many "great" rules, the great rules mean nothing
Pick the trend after the game	Seal every rule with a date before the games it bets
Quietly drop losing picks	Record every bet placed under the bet rule from 2026, wins and losses, and never remove one

This is the record of a research project. Nothing here is betting advice.