Before measuring an edge, check the clock.

A comparison is only as good as its timestamps. A line collected after a game cannot establish what a forecaster knew before it.

The first import includes 2434 archived line records. Those records lack a verified bookmaker publication timestamp. At this edition, 0 observations qualify as captured before kickoff. We therefore report no prospective market benchmark.

The forecast scorecard now preserves registered predictions, source-state histories and qualifying model-versus-market comparisons for both sports.

Keep an immutable record

Each model run receives a version identifier and cutoff. Forecasts are stored with their creation time. Imported lines retain their observation time and source payload. Later evaluation must select a forecast and quote that both existed before kickoff, then join the completed result.

Closing-line comparisons need a further distinction: the most recent quote we happened to collect is not necessarily a sportsbook’s closing price. Without publisher time and bookmaker provenance, we keep the archive label.

Check the sign convention

A home spread of −3 means the home team is favored by three. A model home margin of +5 means the model favors the home team by five. The difference is +5 + (−3), or two points toward the home team. It describes disagreement; it does not by itself account for uncertainty, price, or transaction costs.

Report the misses and the denominator

A future comparison should state the number of qualifying games, excluded observations, pushes, forecast error and probability calibration. Games with no line must remain missing. Historical performance claims should always identify whether they come from a simulation or predictions recorded before the event.

Data attribution: SportsDataverse, CC BY 4.0. See the source receipts and model notebook.