We explain our math: honest analytics, no black boxes
A useful result needs context: the data behind it, assumptions, freshness, and confidence. Here is what we require, with coverage in older tools still under review.
The problem with an unexplained number
A number such as "85% win probability" is hard to evaluate without context. What inputs? What assumptions? Data from when? How confident? Those questions apply to Pick Pilot as well as any other tool.
What we surface on every number
The Pick Pilot analytics contract requires five fields alongside the value; coverage remains a release-audit item on legacy surfaces:
- inputs — the exact data rows the number was computed from (player ids, date ranges, league id).
- assumptions — the modeling choices (for example, the run fidelity, seed, rolling window, and availability treatment).
- freshness — how stale the inputs are (for example, a source timestamp and status).
- confidence — a model-specific quality score (e.g., sample size, projection variance).
- explanation — plain-language why the available inputs moved the result.
Why it matters
Your next move depends on what you believe about player form, availability, and the matchup. We want you to see those starting assumptions, challenge them when needed, and know when the model does not fit your league.
What it looks like in practice
An illustrative matchup result should read like this:
- Value: a modeled win probability for Team A.
- Inputs: disclosed run fidelity and seed, using a named rolling-form window with availability adjustments.
- Freshness: the actual source timestamps.
- Confidence: the interval/method supported by that engine.
- Explanation: the named advantages, risks, and missing evidence in plain language.
Closing
You should be able to check the math before you make the move.