Results Docs
What the Results page is showing, what is counted in tracked performance, and how to cross-check it against backtesting.
Methodology Reference
Current product links stay exactly the same. Once you are in methodology, use this reference strip to move between related sections without going back through the app UI.
FR means FiteQuant Rating: the internal fighter rating framework used to organize model inputs across the platform.
On This Page
Use this page to check what Results is counting, where the time-safety boundary sits, and how to verify the tracked numbers elsewhere in FiteQuant.
The Results page is a record of settled bout outcomes against predictions already stored by the system. It is not a retroactive simulation page.
Recent Results shows bout-level records in settlement order.
Model Prediction Data summarizes the stored prediction set for the selected model using the same underlying settled rows.
This page is intended to reflect tracked predictions made before settlement. For users concerned about leakage, the main cross-check is the Backtest page in strict time-safe mode.
Results and strict backtests are designed to be comparable at the policy level. If a model has changed since a prediction was stored, the Results page still reflects the stored prediction state, not a re-run using the model as it exists now.
Value-only default: a row is counted as a signal only when model win probability exceeds market-implied probability enough to produce positive edge.
NO VALUE means the model still made a directional pick, but the page is not counting that row in signal-based ROI or signal accuracy.
Profit, ROI, and signal accuracy are reported off the signal set, not off every tracked row on the page.
Profit values are shown in units. Unless noted otherwise, signal summaries assume flat staking of 1u per counted signal.
Reinvested profit is shown separately as a compounding illustration. It should not be read as a claim about executable sizing quality, liquidity, or real bankroll management.
This page does not claim out-of-sample proof in the statistical sense. Early samples are small, model edits matter, and boxing schedules are sparse compared with mainstream team sports.
The intended use is auditability: users should be able to inspect stored calls, compare them with strict backtests, and decide for themselves whether the model behavior is coherent.
Use the Backtest page with strict time-safe mode to reproduce the same decision policy on settled fights without using future-state inputs.
Use the model window filters (Since Created and Since Last Edited) to separate baseline performance from later configuration changes.
If you want every eligible directional pick instead of value-only signals, that is available in backtesting as an explicit strategy choice rather than being mixed into Results by default.
Draws and no contests remain visible for transparency but are not treated as signal wins.
Odds availability and source timing still matter. A row can be a valid prediction record while not contributing to signal accounting if no qualifying value signal existed.
Related
