Credence Sports
Better sports decisions start with knowing what you don't know.
Credence Sports builds fantasy decision tools and betting intelligence around a simple idea: a projection is more useful when you can see the uncertainty around it, understand what is driving it, and judge it by what actually happened.
Why Credence exists
Sports analytics is full of precise-looking numbers. A player is projected for 17.4 fantasy points. A team has a 61% chance to win. A market supposedly has a 4% gap to close.
The number is rarely the whole answer. How uncertain is it? What assumptions produced it? How often has the model been right before? Does the apparent difference survive proper testing? What should you actually do differently because of it?
Credence Sports exists to make those questions part of the product rather than something hidden behind the model.
We are not trying to eliminate uncertainty from sports. We are trying to measure it well enough to make better decisions inside it.
Two products, one standard
Fantasy decision tools
Credence turns player projections into rankings and decisions that reflect the league you actually play in rather than a generic scoring format.
Available now
- Draft rankings
- Player projections
- Projection ranges
Free to read without an account: every player we project, our rank for him, the 80% range on his season, and the market's ADP beside it. A membership re-scores that board for your own league's scoring and roster.
Coming this season
- Weekly projections
- Start/sit decision support
- Waiver tools
- Matchup-aware recommendations
These are not built yet. Nothing on the site treats them as available, and this list is where they are tracked until they are.
Betting intelligence
Credence compares model-implied probabilities with the betting market, shows where they disagree, and keeps the uncertainty around the estimate attached to it. A model-market gap is information. It is not automatically a bet.
Available now
- MLB betting intelligence
Model probabilities, the de-vigged market consensus beside them, the distance between the two, and the range around our estimate — graded against the market afterwards, losses included.
Coming this season
- NFL betting intelligence
- NCAAF betting intelligence
Both are the same product as the MLB one — model projections, the market's number beside them, and the record kept in the open. Neither has shipped.
What we believe
Show the uncertainty.
Every estimate carries a range, a distribution, or another measure of how sure we are, wherever the model supports one. A wide interval is information, not a flaw — it tells you how much confidence the evidence actually warrants.
Grade the whole record.
We publish wins and losses rather than the observations that happened to go well. Our MLB picks are scored against the market after the game on the members' scorecard, and every past fantasy season is scored against what happened, one row per player, in a track record open to anyone.
Show the work.
A projection is more useful when you can see what moved it. Where the model supports it we expose the inputs, the drivers, the market comparison, the uncertainty and the evaluation history — so you can disagree with us and see exactly where.
Retire what fails.
Models and hypotheses have to clear evaluation gates set before the answer is known. When an idea does not survive proper testing we close it out, in the changelog, with the reason — rather than changing the test until it passes. Several have been.
Separate disagreement from edge.
A model disagreeing with a market or a consensus does not establish that the model is right. Model-market disagreement, evidence of signal, and something you can act on are three different things, and we do not treat them as interchangeable.
An honest maybe beats a confident guess.
Especially when the maybe arrives with a probability, a range, and a record you can inspect.
We model ranges, not just answers.
Sports outcomes are noisy, so a projection should carry more than a single point estimate. Credence uses predictive distributions, probability estimates and calibrated ranges wherever the model supports them, so you can see not only what the model expects but how much confidence the evidence warrants.
Today that shows up in
- —Fantasy player projection ranges
- —MLB model probabilities and their uncertainty
- —Model-versus-market comparisons
The model does not get to grade its own homework.
Every serious model change is tested against explicit benchmarks and held-out data. New features have to beat simpler alternatives. More complex models have to justify their complexity. Calibration matters, and so does whether a result survives reasonable validation.
When a result is underpowered, we call it underpowered. When a signal disappears under proper testing, we say so. When the market remains the better estimate, we say that too.
Our goal is not to make every research question produce a winning model. It is to make the answer trustworthy.
Wins and losses both belong on the page.
Credence keeps a record of the forecasts and model-market comparisons it publishes. Every MLB pick is scored against the market consensus after the game and lands on the members' scorecard; every past fantasy season is scored against what actually happened, in a track record open to anyone with no account required.
The point is not to turn every correct call into a victory lap. It is to keep the whole record inspectable, including the forecasts that were wrong and the stretches where the market was the better estimate.
What that record has not shown is a durable advantage over the closing market, and we would rather say so here than let the page imply otherwise. Transparency only means anything if the losing observations stay on it too.
See the fantasy track record →Credence is for the person who wants to know why.
The fantasy player who wants to know why a player is ranked differently in their league. The bettor who wants to see the model probability sitting beside the market consensus. The fan who already knows that a projection without uncertainty is incomplete.
And the person who would rather be told the evidence is weak than be sold confidence the model has not earned.
If you have ever opened a player page because a ranking looked wrong, and wanted to understand the mechanism rather than simply accept the number, this was built for you.
What Credence is not
- —Not a tout account, and not a service that sells you winners.
- —We do not place, size or submit bets on your behalf. Every wager is your own decision and your own action.
- —We do not treat every model-market difference as something you can act on.
- —We do not label the featured MLB read as the day's best bet or its biggest opportunity.
- —And we do not hide uncertainty because a cleaner number is easier to market.
Credence provides decision intelligence. The decision remains yours.
Questions about how Credence works?
The FAQ covers how the market consensus is calculated, what the ranges mean, how MLB projections update through the day, how fantasy rankings are personalized, and how a model earns its way into the product.
A product of Penumbra Partners.