Scores, schedules and box scores
ESPN NBA and ESPN NHL public feeds supply fixtures, team-season schedules, standings, box scores and play-by-play. The NHL publisher scoreboard is used to cross-check and update current NHL scores when available.
These are public publisher feeds, not contracted APIs with guaranteed availability. The NBA live CDN was tested but rejected requests from this deployment environment. Source failures retain the last successful snapshot and its original timestamp.
Live updates and source delay
The collector targets a 15-second cycle during live games, and a 60-second cycle otherwise. Game details refresh during live cycles; injuries and standings refresh approximately every five minutes. Schedule and model refreshes run every six hours and after newly final games.
The browser polls snapshots every ten seconds during live games and pauses polling in hidden tabs. Publisher delay and request duration add latency. The displayed game clock is the last reported clock; it is not an invented countdown.
Injuries, lineups and goalies
ESPN injury feeds provide player status, notes, report time and estimated return dates. A report’s age is visible; an empty list does not prove a full-strength roster. NHL probable starters are labelled projected until confirmed by a reliable source.
Current injuries do not enter model v1. Using today’s report to backtest an older game would leak hindsight. Pregame injury snapshots are now saved with forecasts so future player-impact models can be evaluated prospectively. Player-level impact estimates and enough archived games are still needed. NBA’s official reporting site and licensed injury/lineup feeds are options for stronger coverage.
Rest, travel and “jet lag”
We measure calendar days off, hours between scheduled starts, back-to-backs, games in seven days, approximate direct venue-city distance, and local UTC-offset changes including daylight saving time. Future scheduled games can inform future travel context without assuming their results.
Actual flights, sleep, chronotype and acclimation are not observed. Travel is a proxy: research reports associations in NBA performance and NHL performance, not a universal causal penalty to subtract from every team.
Advanced stats worth adding
NBA: pace-adjusted offensive/defensive rating, effective field-goal percentage, turnover rate, offensive rebounding, free-throw rate, lineup continuity and player minutes. Current box scores already expose shooting, rebounding, turnovers and player performance where supplied.
NHL: five-on-five expected goals, shot share, special teams, goalie workload and goals saved above expected. MoneyPuck’s published downloads are a promising next source with required attribution and non-commercial-use terms. Its xG and GSAx are not yet inputs here; ordinary scoring margin is not relabelled as expected goals.
Synpath prediction markets
Synpath 0.5.1 now supplies public Kalshi market quotes and five-level order books through an isolated read-only collector. NBA, NHL, NFL, MLB and UFC contracts enter game details only after unique date, matchup and outcome checks. F1 contracts appear in the shared market explorer; F1’s existing Kalshi model benchmark remains separate.
Prices, resting contract sizes, trading volume, quote gaps, settlement rules and collection timestamps remain distinct. Local price history starts with this integration. Hosted tick history and cross-venue matching are not connected; those require a Synpath account/key. No trading credentials or order-entry service are configured. Explore prediction markets.
PREDICTION INTEGRITY
What a probability means here
A pregame probability describes the winner, including overtime and NHL shootouts. It is frozen after the scheduled start. If no forecast was saved beforehand, the page says so. We do not regenerate an apparently “pregame” forecast using the final score.
Any ESPN live win estimate is explicitly labelled as the publisher’s estimate, separate from Sports Desk’s frozen pregame model. Historical model evaluations use time-ordered features and held-out 2025–26 games; NBA record and NHL record show calibration and baselines.
Preseason and postseason games are tracked but forecasts are withheld. Early-season form may span previous seasons and rosters can change. The model is an initial analytical baseline, not a proven betting edge.