How this works

An honest summary of the methodology — enough to trust the numbers, not enough to rebuild the pipeline. Where a sport's data isn't fully ready yet, that's stated plainly rather than smoothed over.

Two systems, two different jobs

Jackdaw is an in-house projection, computed independently from official league statistics. It starts from a player's recent performance, weights recent seasons more heavily than older ones, and regresses each player's rate toward a positional baseline in proportion to how much evidence exists for them — a rookie's few games move the projection less than a veteran's several seasons. On top of that sit age curves (fitted from how players' own numbers actually change year to year, not assumed) and a playing-time model, since a great rate on the bench is worth nothing. Every one of these pieces is fit and checked out-of-sample, on seasons the fitting process never saw, before it's trusted.

Magpie is not "one hand-picked best projection system." It's an accuracy-weighted blend across every independent public projection system available for that sport, plus Jackdaw — weighted by how each one has actually performed historically, re-validated every season rather than assumed to hold. The weighting is granular (per stat category in some sports, per position in others) because a system can be excellent at one stat and mediocre at another. In practice this often concentrates almost entirely on whichever source has been most accurate for that particular stat — but it's arrived at by measurement, not by a fixed preference, and a source with no track record contributes nothing rather than an assumed-average weight.

Backtesting: how "accurate" gets measured

Every system on this site — public or in-house — is scored the same way: its historical projections are compared against what actually happened, over multiple prior seasons, using error metrics (how far off a projection was, on average) and rank correlation (whether it got the order of players right, which matters more for draft/roster decisions than exact totals). Fitting always happens on data prior to the season being scored — a system is never allowed to "predict" a season using information from that same season. Shortened or interrupted seasons are handled carefully so a system isn't credited or penalized for something no projection could have known in advance.

The accuracy comparisons shown on each sport page are anonymized by design — public system names aren't attached to their numbers here, out of respect for the licensing terms some of them publish under. You'll see the shape of the comparison (how many systems, how they rank against Jackdaw and Magpie) without a named leaderboard of third-party products.

Where the data comes from

Jackdaw is built from official league statistics. Magpie's inputs are a range of publicly available projection systems, each reviewed individually for what its terms permit before being used at all — some are fetched automatically, some are entered by hand after manual review, and a few are excluded outright where their terms don't allow this kind of use. What's stored is always the raw projected components (yards, makes and attempts, and so on) — never a single derived number — specifically so that your own league's scoring rules can be applied after the fact, in your browser, without asking any upstream system to recompute anything.

Current status, honestly, by sport

• NFL — both systems are fully built: Jackdaw, an accuracy-weighted Magpie blend per position, and a full position breakdown with value-over-replacement rankings.
• MLB — both systems are fully built. Magpie is a real accuracy-weighted blend of Jackdaw and public projection systems, full coverage across batters and pitchers.
• NBA — both systems are fully built, full position coverage. A real third-party accuracy comparison exists here too now — a couple of independent, anonymized public systems currently score better than Jackdaw on error, an honest result shown as-is rather than smoothed over.
• NHL — both systems are fully built, with several prior seasons of backtested history. Magpie is published for every current-season skater and goalie now (a licensing question that used to leave some players on Jackdaw with a marker has since been resolved).

Your league's scoring, computed in your browser

Nothing about turning raw projections into your league's rankings happens on a server — there isn't one. The site fetches one JSON file of projections and one of accuracy data, and every scoring calculation, ranking, and CSV export happens with JavaScript already running on your device. Your scoring settings never leave your browser.

What this page won't tell you

Deliberately: the exact fitted parameters, the precise weighting formulas, or the full list of every data source and how each is combined. That level of detail lives with the people building each sport's engine, not on a public page — this is an explanation of the approach, not a specification you could rebuild the system from.