Forecasting, explanation and outcome tracking

Cyrus Insight Engine

Turn historical measurements into an estimate of what may happen next, explain the factors behind that estimate, and compare it with the eventual result.

The engine is a repeatable statistical workflow: data preparation, a defined model, confidence estimates, publication and evaluation. Its current implementation is specialized; applying the workflow to a business problem requires a suitable dataset, model and validation.

Applications Across Industries

The same workflow could support demand forecasting, staffing and workload planning, inventory estimates or service-volume projections. For example, a contact center could estimate next week's hourly arrivals from its historical traffic and known events, log the forecast, and compare it with actual arrivals.

Those business integrations are potential adaptations, not existing features of this demonstration. Each needs relevant data, domain-specific assumptions and evaluation before operational use.

Read the implementation and methodology on GitHub →

Worked example · NFL player statistics

Applying the Insight Engine to player projections

This implementation uses NFL player statistics to demonstrate two measurable outcomes: quarterback passing yards and the probability that a running back scores at least one rushing or receiving touchdown. These are the example's measures, not the engine's product name.

Inputs and calculation

Passing yards: The model uses an exponentially weighted average of prior starts, with a half-life of eight games, shrunk toward the league average. It applies an opponent pass-defense factor and adjustments fitted on 2021–2025 starts for home field, division games, a dome, a missing top target, the spread and the game total.

Its estimate of the chance of finishing above or below the reference line uses a 50/50 blend of the projection and that line, with an assumed spread of 30% of the mean. This makes the estimate market-informed; it is not independent of that reference.

Touchdown probability: A shrunk per-game scoring rate is scaled by the team's implied points. The chance of at least one touchdown is calculated as 1 − exp(−λ), where λ is the estimated touchdown rate. The documented categories are Strong at 70% or higher, Moderate from 60% to below 70%, and Slight below 60%.

Publication and evaluation

The documented schedule runs Wednesday and Friday at 6:45 PM Central. The last projection published before kickoff counts for grading. Grading starts no earlier than 12 hours after kickoff using weekly player statistics; a missing player statistics row when the team played is treated as void under the current rules. The public record begins with 2026 Week 8.

Use the example

Search for a player or select a tab below. Compare the estimate, benchmark, confidence and explanatory notes. The track record shows recorded outcomes; it does not establish that future predictions will be accurate.