Inverse calibration

Enter dividends (one per line, or comma-separated). The calibrator runs three iterations of monotone-interpolated inversion against the field, and we round-trip the recovered abilities back to prices/dividends to show how tight the closure is.

Reading the table. Input dividends rarely sum to a fair book — $\sum 1/d_i$ usually exceeds 1 by the bookmaker's overround. The calibrator first normalises the prices to sum to 1, which is equivalent to working with the “fair dividends” in the second column. The round-trip price should agree with the input price to ~$10^{-4}$ — that's the column to watch. Dividend differences between input and fair are the overround, not a calibration error.

Recovered abilities are median-centred (only differences are identified). For very small prices the ability transform is well-defined but flat — small price errors translate to large ability errors, so don't read too much into the absolute number for hopeless outsiders.