How It Works

The CFL King is a research tool for CFL player props. For every player and market β€” passing, rushing and receiving β€” we project a realistic number and turn it into a probability at the posted line. We publish the projection and that probability; we do not publish an edge, a confidence tier or a stake. The one place we name a side is the free gate page, which surfaces a single projection with the side it favours as a sample of the board β€” with no edge, no tier and no recommended stake attached. Here's the whole approach, in plain English.

1Real game-by-game production

We start from official CFL.ca game logs β€” every player's passing, rushing and receiving numbers, game by game, with history back to 2022. We work in per-game terms so short seasons and missed games don't distort the picture.

2Recent games count more

What a player has done lately tells you more about his current role than a game from a year ago. We lean on the last few weeks most heavily, while still letting older form count.

3Steady, not streak-chasing

A three-game hot streak isn't a new normal. We temper small hot or cold stretches against a player's fuller body of work, so the projection stays stable instead of overreacting to noise.

4A projected number, and a chance to clear it

That gives us a projected number for each stat β€” passing, rushing and receiving. When a line is posted, we also show the chance the player goes over it, accounting for how much his results normally bounce around game to game.

5The line, and our number next to it

When a line is posted we put our projection and the model's win probability beside it and leave the comparison to you. We used to publish the gap as an edge percentage and sort every pick into Elite / Strong / Lean / Pass tiers. We no longer do either.

6Why the edge and the tiers are gone

We graded them, and the grading does not support them. Our own track record is 327-245 = 57.17% over 572 decided picks across 24 slates. A cluster bootstrap over those slates β€” resampling whole slates, because picks on the same slate rise and fall together β€” puts the 95% interval at [51.5%, 62.3%]. That lower bound sits below the 54.34% a βˆ’119 single leg has to clear and below the 57.74% a two-pick PrizePicks power play has to clear, so the record cannot tell this model apart from no edge at all.

The tier ladder is worse off than that. Our own model_perf.json holds 614 posted rows with the grade field null on every single one β€” nothing was ever actually tagged β€” and 407 of those 614 (66%) settled as void, so the 205-pick "record" above is a minority survivor subset of what the board really posted. A conviction we never assigned, over a sample we mostly never settled, is not a conviction. It is gone.

What did hold up is calibration, and only calibration: on accuracy our Brier score is 0.2367 against a 0.2500 coin-flip baseline. That says our percentages are honest percentages. It says nothing about beating a price, and that file has always carried not_a_betting_record.

7Nothing hidden after the fact

Once games finish, every posted pick is graded β€” win, loss, or push β€” and kept in the track record. Nothing gets quietly deleted, including the numbers above.

See how accurate our numbers are β†’ β€” a plain check of whether our percentages hold up against what actually happens. It's an accuracy check, not a betting record.

What the numbers mean. A projection is the number we expect a player to post. When a line is posted, the win % is how often we'd expect him to clear it. There is no edge column, no confidence tier and no play tag on the board or anywhere in the members’ area; the free gate page’s single sample projection is the only place a side is named. Rows marked projection-only have no posted line yet, so we show the number on its own.

Common questions

Are these guaranteed winners?

No. Player props are inherently uncertain and even our best reads lose regularly. Lines are sharp, so we make no claim to beat the market β€” our aim is an honest, useful number, which is exactly why we publish a transparent track record.

How far back does the data go?

Player game logs run back to 2022, sourced from official CFL.ca box scores. More history means each projection is grounded in real production, not a tiny sample.

Where did the edge % and the confidence tiers go?

We retired them on 2026-07-27. Sharp lines rarely leave a real gap on a single prop, and when we finally measured ours the 95% interval on 205 graded picks ran from 49.2% to 64.2% β€” straddling break-even. The tier labels were worse: 614 posted rows carried a null grade and 66% of them voided. Publishing a number we cannot defend is worse than publishing none, so the projection and the win % are all that is left.

How should I use this?

As an information tool, not financial advice. Only ever risk money you can afford to lose, and treat every projection as an estimate carrying real uncertainty.

The CFL King is for entertainment and informational purposes only. 21+. If gambling is no longer fun, it's time to stop β€” please play responsibly.

For entertainment & information only. 21+. Please play responsibly.