Round Win Probability Added
RWPACS2v0.1.0 · baseline · validated 2026-07-16What it measures
How much each player's actions moved their team's chance of winning rounds. A 1v3 clutch kill swings far more probability than a cleanup frag in a 5v2.
Formal definition
Round win probability (interpretable state model over alive counts, equipment, bomb state, and clock) is evaluated before and after each attributable event (kills, plants, defuses). ΔWP is credited to the acting player; per-round average over rounds played.
Inputs
- kill/plant/defuse events
- round state timeline
- round_wp_cs2 model
Exclusions
- rounds with missing state (confidence < 0.5)
Sample & population
- Minimum sample
- 100 rounds
- Baseline population
- Players in the same game/window with ≥ minimum rounds
- Direction
- higher better
- Unit
- win prob/round
Percentile interpretation
Captures round-impact density rather than volume.
Confidence = source confidence × model confidence. Values from baseline models are capped at 0.7. Below-minimum samples display the value grayed out with no percentile.
Example
Opening kill in a gun round: WP moves 0.50 → 0.63; the killer is credited +0.13.
Known limitations
- Assists and non-event contributions (space, info) are not credited in v0
- WP model is a baseline prior, not fitted on pro data
Current distribution
career window · n=35 eligible playersModel card
round_wp_cs2 v0.1.0Hand-specified priors, sanity-calibrated on synthetic fixture corpus v1. NOT fitted on production data.
Terminal states pinned to truth; mid-round states are directional priors pending real-data fit.
- alive_diff(CT alive − T alive) / 5
- alive_ratio_edgeNear-wipe indicator (±0.5 when a side is down to 1)
- equip_diff(CT equipment − T equipment) / 5000, clamped
- bomb_plantedBomb planted indicator
- bomb_time_pressurePlanted × (1 − bomb-time fraction remaining)
- late_no_plantNo plant × (1 − round-time fraction remaining)
Live betting · Cross-game comparisons
Change log
- v0.1.0 · 2026-07-16Initial baseline release