Every call is checked against the stock price 30 days later. The rule, what counts as right, and why 380 early calls are excluded from the score.
Each weekday, Metis Committee meets on a short list of stocks. Five AI analysts argue each one and a chairman makes the call. Each call is written down with the price at the time of the meeting, and 30 days later a grading job checks what the stock actually did.
This page explains that check, because a track record is only worth as much as its rules.
A meeting produces five predictions. Four come from the analysts who take a side: value, growth, macro and technical. The fifth is the chairman's final call, graded the same way, so the record covers the decision users actually see and not just the inputs to it.
The risk officer is not graded on direction. It doesn't pick one. Its job is deciding how big the position should be.
Any move smaller than 2% in either direction counts as flat.
Two numbers come out of this. Accuracy counts a partial as half a correct call. Hit rate only counts outright correct calls, so it is always the lower of the two. For BUY and SELL calls the record also shows the average return if you had followed the call.
Thirty days is long enough that one day's noise doesn't decide a grade, and short enough that the record fills in within a quarter. The 2% band stops a stock that drifted 0.4% from counting as a win for anyone. Both numbers apply to every call the same way and are printed on the track record page.
Accuracy stays hidden until 20 calls have been graded. With five graded calls, one lucky week can show 80%, and a number like that would be marketing, not information.
As of October 9, 2026, the record shows 0 graded calls, 60 waiting for their 30 days to pass, and 380 excluded. The 60 pending calls were made on October 1 and 2, so the first grades land at the start of November.
From September 10 to September 30, the committee ran on broken inputs. Fact sheets were missing valuation and growth figures, the yield-curve spread was misreported, and the Scout step that picks which stocks to discuss was failing. We found the problem on September 30 and fixed it.
The 380 predictions from those meetings are still listed on the track record, with that reason next to them. They don't count toward accuracy in either direction. We excluded them on October 6, before the first of them was due to be graded on October 10, so the call was made without knowing whether they would have helped or hurt the score.
Grades feed back into the committee. Once an analyst has five graded calls, the chairman's weighting shifts toward whoever has been right more often. If an analyst's accuracy in one sector falls below 35% across at least three graded calls in that sector, their weight on calls in that sector is cut in half until it recovers. When that happens, it shows up in the activity feed.
A 30-day price check is a narrow test. A correct BUY can be luck, and a stock can fall for reasons nobody could have known at the meeting. Twenty graded calls is a small sample, and even a few hundred says little about the next call.
These are paper calls with no money behind them. Nothing here is investment advice, and past calls don't predict future ones.