CSC FantasyRules
How drafting, scoring, and standings work in CSC Fantasy
Drafting
Each week you draft a lineup of 5 players from your tier. Your roster slots are:
AWPer
Support
Rifler
Flex
Flex
- Lineups lock before the first regularly scheduled match day of each week in that tier.
- Each player can only fill one slot in your lineup.
- You can pick yourself.
- You cannot pick your teammates or players on teams you are scheduled to play against that week.
- Points are scored per match — if there are two match days in a week, each player can earn points from both games.
Auto-Substitution
If one of your drafted players did not play in a match but their team did, the system looks for a temporary-contract sub who played in that match. If multiple temp subs played, the one closest in MMR to your absent player is chosen.
The substitute's stats are used for that match only. Your original player still earns points for any other matches they played that week.
Fantasy Scoring
Points are calculated per match for each player in your lineup. Your weekly total is the sum of all player points across all matches.
AI Predictions
Work in Progress — These predictions are experimental and actively being improved. They're meant to be a helpful starting point, not gospel. If you notice something that seems off, that's expected — we're still tuning the models and welcome feedback.
The "Projected" fantasy points shown on the picks page are generated by a machine learning system trained entirely on CSC match data. The goal is to give you a data-driven starting point when building your lineup.
Step 1: Match Win Probability
For each upcoming match, a logistic regression model estimates each team's chance of winning. It compares 14 features between the two teams, including:
- Average rating, ADR, and KAST differences
- Recent form (win/loss record over the last 5 matches)
- Opening duel rates (first kills vs first deaths)
- Multi-kill and clutch round rates
- Average MMR difference between rosters
- Rating consistency (how volatile each team's players are)
All features are normalized and the model is trained using gradient descent with regularization to prevent overfitting. The resulting probability is also dampened based on how much training data is available — with only a few matches played, predictions stay closer to 50/50.
Step 2: Individual Player Stat Prediction
Each player's stats are predicted using an exponentially weighted moving average (EWMA) of their recent match history. This means recent performances matter more than older ones — a player who's been on a hot streak will have a higher projection than their season average might suggest.
These base predictions are then adjusted for context:
- Opponent strength — Playing against a team with a higher average rating and ADR lowers the projection by up to 30%. Weaker opponents raise it.
- Win probability — Players on the favored team get a slight boost (up to 20%), since winners tend to have better stat lines.
- Win/loss splits — If a player has enough history, the model separately tracks their performance in wins vs losses and blends based on expected match outcome.
- New players — Players with fewer than 10 games have their predictions regressed toward the tier and role average to avoid wild swings from small sample sizes.
Step 3: Fantasy Point Calculation
The predicted stat line (kills, deaths, assists, ADR, rating, multi-kills, clutches, etc.) is run through the exact same scoring formula used for real games. The final projected FPts you see is actually a blend: 60% from the EWMA of past fantasy point totals, and 40% from the recomputed stat breakdown. This blend helps capture bonus points (clutches, aces, etc.) that a pure stat average might miss.
If a player has multiple matches in a week, the projections for each match are summed.
Model Caching
Training a model from scratch on every page load would be slow, so each tier's model (weights, feature stats, and tier baselines) is cached and periodically retrained as new matches are completed.
Known Limitations
- Early in the season, predictions lean heavily on tier averages since there isn't enough match data yet.
- The model doesn't account for map picks, roster chemistry changes, or players having an off night — it's purely statistical.
- Bonus-heavy stats (aces, clutches) are inherently hard to predict and may be over- or under-estimated.
- Projections will improve as the season progresses and more data becomes available.
These projections are one tool in your drafting toolkit. Your own knowledge of the league, player tendencies, and matchup dynamics will always be valuable — use the projections as a starting point, not a final answer.
Standing Points
Each week, players are ranked by their fantasy point total within their tier. The top 7 earn standing points. Your overall season ranking is determined by your cumulative standing points.
Players ranked 8th or lower receive 0 standing points for that week.