How it works

cs2props projects how many kills and headshots each player will get in upcoming professional Counter-Strike 2 matches, and puts that number beside the lines PrizePicks and Underdog post for the same prop. This page explains where the number comes from, how to read the slate, and what the number has and has not been shown to do.

What gets projected

Kills and headshots for every player on both rosters, on the market the books actually price: map 1 for a best-of-one, maps 1–2 for a best-of-three, maps 1–3 for a best-of-five. Only maps 1–2 are modelled. The maps 1–3 figure is the maps 1–2 projection times a flat multiplier, which has not been checked against best-of-five results.

How a projection is built

• Results. Every finished professional match is pulled from bo3.gg into one results file, about a year deep, one row per player per map.
• A rate. A LightGBM model predicts kills per round, and separately headshots per round, from each player’s recent form. That is averaged with the player’s own trailing 90-day rate — for kills the two count about equally, for headshots the trailing rate counts for more. When the two halves land on opposite sides of the line the slate marks the row split: the published number is a compromise between two views that disagree.
• Rounds. Each fixture’s map veto is simulated and the rounds on each map are projected from both teams’ history and the match odds.
• Rate × rounds, adjusted slightly for the map, gives the raw projection.
• Calibration. A correction fitted against scored results is applied last. It pulls projections toward the average, because much of the model’s spread has turned out to be noise — it corrects a systematic error, and does not create an edge.
• Lines. PrizePicks and Underdog lines are imported from each board and matched to a projection on player, stat, market and opponent, so a line only ever lands on the identical prop.

Reading the slate

Proj ±

The projection and its typical range: how far a result usually lands from a projection of that size (about 29% of it for kills, 35% for headshots). It is the same for every player at a given projection, on purpose: tested over 1,033 players, a player’s own past spread did not predict his future spread, so the site does not rate anyone as more “consistent” or “safer”.

PP / UD

The PrizePicks and Underdog lines, when posted.

Gap

How far apart the two books’ lines are.

Decides

How often taking the better of the two lines, instead of the worse one, changes the result — the result has to land on or between the lines for it to matter. Checked against our own graded props: kills 5.6% observed against 5.3% predicted, headshots 8.0% against 8.3%. Price differences on the slate is ordered by it, which puts kills and headshots on one scale. It is about price only: where to take a side, never which side.

Better number (gold)

For the side the projection leans, the book offering the easier line: the lower line for an over, the higher for an under. It marks a price, not a side to take. Between means the projection sits between the two lines, so each book is the better price in the opposite direction.

Edge SD

Projection minus the better number, in units of the typical range. Shown for scale only; nothing is sorted by it.

few games

Marks a player without much recent play behind the projection: fewer than 600 rounds in 90 days or 100 in the last 30. The “enough recent games” filter hides them.

Markets: Priced / All

A best-of-three has both a map 1 and a maps 1–2 projection. Priced shows only the one the books actually posted; All shows both.

KPR

Kills per round — a projection or line divided by the match’s projected rounds. A player lives roughly between the bands the diagnostics show; a number far outside them is a broken input, not an opportunity.

What gets hidden

A player who has not played for more than 21 days, and whose projection sits more than 30% away from the posted line, is left off the slate. That combination is a stale trailing rate that cannot see a change in the player’s level of competition, not a real disagreement.

What it cannot tell you

There is no demonstrated edge. Over 2,240 scored kills props, bigger gaps between projection and line hit no more often than small ones (50.0%, 48.5%, 52.0% and 49.9% by size of gap, against a break-even near 53%), which is why nothing on the site is ranked by edge. The one thing that held up in testing is price: where the two books disagreed, the better number hit 48.3% against 44.0% for the worse one — a real difference, still short of break-even. The live record is on the track record page. Nothing here is betting advice.

How often it updates

The slate is rebuilt every hour while matches are upcoming, so lines and projections move as the boards do. Finished slates are scored against the actual results, usually the next day, and that score is what the track record shows.

Last reviewed 2026-10-15.