How to audit seven settled assessments
A worked weekly audit shows how outcomes, quoted prices and closing prices should be reported without treating seven results as proof of a method.
Consider a worked weekly audit of seven assessments. Three win, four lose, and the week finishes slightly down. Those figures are illustrative, not BAI8's live weekly record. The result is worth almost nothing on its own, which is the reason the rest of this piece exists.
The week, in full
In the example, the three that win are priced at 2.30, 1.85 and 3.40. The
four that lose are at 1.95, 2.05, 2.60 and 1.70. A raw win count ignores
those different probabilities. Exact inference also needs a probability for
each observation, so the sequence cannot honestly be assigned a single chance
from the prices alone.
The number that carries information
Suppose six of the seven are taken at a price better than the one the same bet closes at. That is useful information because it can be measured for every assessment without waiting for a win or loss. It is not a substitute for a proper probability score or a sufficiently large out-of-sample record.
The seventh closes longer than the quoted price, which means the market moves against the assessment after publication. That makes it a sensible candidate for review, but the movement alone does not identify why the market changed.
The argument for weighting it this way is set out in full in why the closing line is the only honest scoreboard.
What this does not tell you
One week is not enough evidence. It cannot distinguish a method with a small edge from one with none, and presenting a single week as proof of either is marketing rather than measurement. A real record needs every published probability, time-stamped quoted price, closing price and outcome, including the weeks its publisher would prefer to forget. It also needs a declared scoring rule and a fixed review window, chosen before anybody sees which assessments won.
Common questions
Is this BAI8's live weekly record?
No. It is a worked example of the information a complete weekly audit should publish.
How should I read a week of seven results?
As a very small sample. Seven outcomes cannot validate a method, so the assigned probabilities, quoted prices and closing prices all need to remain visible.
Does a losing week mean the assessments were wrong?
Not on its own. The chance of a losing sequence depends on the probability assigned to every event, and a process needs many observations before its calibration can be assessed.