How Nova Thinks
Nova is a deterministic research system. It collects data, computes an independent market price, compares it to the sportsbook, and classifies the gap. No language model generates the price. No guesswork fills missing data. The process is transparent and the output is auditable.
Research
Nova collects six categories of data per game, each from a named source. Every input carries a freshness label (fresh, stale, confirmed, projected, missing, not applicable). When an input is missing, Nova says so rather than substituting a guess.
Current spreads, totals, and moneyline prices across books. Sampled continuously and stored as snapshots with timestamps.
Confirmed and projected starting lineups. Lineup confirmation is a material change trigger.
Temperature, wind, precipitation for outdoor venues. Affects scoring environment and total projection.
Player availability with severity classification. Impacts team rating adjustments.
Rest days, travel distance, back-to-back detection. Situational factors for game script.
Last-10 performance, regression indicators, and ATS context for both teams.
Price
Nova computes a fair line for each market using a deterministic engine. Team ratings, matchup context, pace, scoring environment, and situational factors feed a calculation that produces a single number. The fair line is not a prediction of the final score. It is the price at which the market would be efficient.
Spread: Nova\u2019s fair spread is computed from the team rating gap plus venue and situational adjustments. The edge is the difference between the fair spread and the book spread, in points.
Total: Nova\u2019s fair total is computed from pace, offensive and defensive efficiency, and scoring environment. The edge is the difference between the fair total and the book total, in points.
Moneyline: Nova\u2019s fair win probability is computed from the team rating gap and converted to an American odds price. The edge is the probability delta between Nova\u2019s fair probability and the book\u2019s devigged implied probability.
Classify
Nova assigns one of six states to every market it analyzes. The state is determined by the edge magnitude, data quality, and market type. The same thresholds apply across the primary and secondary pipelines for cross-market consistency.
Edge exceeds the full play threshold. Nova’s fair line differs from the book by enough to warrant action.
Edge exists but is below the full play threshold. Directional interest, not a recommendation.
Conditions are developing. A specific trigger (lineup confirmation, price movement) is needed to upgrade.
Analyzed. No qualifying discrepancy. The market is priced where Nova thinks it should be.
Critical information is incomplete, stale, or unconfirmed. Nova withholds a conclusion rather than guessing.
The original opportunity no longer qualifies. The price moved beyond Nova’s threshold.
Monitor
A read is not static. Nova continuously re-evaluates as inputs change. Material changes trigger a new read.
Lineup confirmation, significant price movement, weather shifts, or injury status changes trigger a re-analysis. The previous read is marked non-current; the new read becomes canonical.
Every read carries a writer ID, model version, and timestamp. The external Nova worker is the authoritative writer. Reads are never overwritten; previous versions are preserved with is_current set to false.
What Nova Does Not Do
- Predict game outcomes with certainty.
- Guarantee profit or winning percentages.
- Generate picks from language models without deterministic computation.
- Force a play when data is incomplete or the edge is below threshold.
- Claim market inefficiency without showing the fair line and the gap.
- Attribute league-level news to specific games without source linkage.
The fair line, the edge, and the state are all computed deterministically. The process is auditable.