Results

Four different things, kept apart: how the projections tracked the scores, how the ownership projection tracked the field, how the pages' plays graded, and where our own entries finished. None of them is profit.

Projection accuracy

Correlation between projected and actual DraftKings points across every projected player, by week, against ranking by salary alone. 1.00 would be perfect.

Loading…

Ownership accuracy

Two datasets, kept apart. The first is the validation log: the slate's ownership as logged before lock against the standings of contests Clarity scored, the same figures the home page shows. The second is the public recaps, which exist only for contests with a published breakdown.

No logged contests yet.
No recaps with an ownership comparison yet.

The pages, graded

A play hit when he scored at least his projection and paid when he reached 85% of his ceiling. Every contest, including the bad ones. All recaps

No contest graded yet this season. A page is graded when the standings file of a contest it was written for is uploaded after the games, usually on the Tuesday; until then the projection record above is the only grade on the board.

Our entries

Where the owner's lineups finished, in the contests he chose to show. The grades above are the record and appear for every scored contest; these are personal finishes, not returns: entry fees and payouts are not recorded, so no ROI is shown.

No finishes shown yet. They appear for the contests Clarity chooses to show after they are scored.

Method and limits

Projection accuracy is scored from the slate as it was logged before lock, against the official DraftKings points after the games, for every player the slate projected. It is a ranking measure, not a profit measure.
Ownership accuracy compares the ownership the slate projected at its contest size with the share of entries that drafted each player in a specific contest's standings file. Different contests of the same slate differ.
Page grades use the page as it stood when the results were uploaded; while a slate is open the page reads live, so results are uploaded after the games. Historical pages and recaps are kept as files and are not rewritten after the fact.
Simulated metrics on a build (top-1% rate, cash rate) describe 10,000 simulated slates, not any real contest, and are never shown here as results.