AI Answer Ledger

Can I see what the AI says about a ticker today?

No, and that is deliberate. Today's calls are recorded but never displayed. Showing a fresh call would be publishing a recommendation with a model's name attached; showing a matured call beside what the price did is reporting a track record. Only the second is what this page is for.

How is a call scored?

Each call is a direction — higher, lower, or unclear — recorded with the closing price on the day it was asked. Thirty calendar days later we compare against the close on that date. "Unclear" is recorded and excluded from the hit rate rather than counted as a miss, because refusing to guess is a legitimate answer.

Why does the page show nothing yet?

Because nothing has matured. The ledger holds no aggregate until at least 20 calls have been scored — a hit rate computed on a handful of calls is noise wearing a percent sign. The page states what it is waiting for instead of showing a thin number.

Is this investment advice?

No. Nothing here tells anyone to do anything, no output is personalised, every visitor sees the identical page, and it is free. It is a measurement of how a class of software performs at a task people increasingly use it for.

Why is every hit rate printed beside an always-higher baseline?

Because a model that answers higher every time can look skilled through any stretch in which most stocks rose, without knowing anything. The baseline scores the same windows as if every call had said higher. A hit rate means something only to the extent it beats that, and it can differ from it only on the windows where the model said lower, so the page prints how many of those there were.

How independent are these calls of each other?

Much less than their count suggests. Calls are recorded weekly on the same ten large caps and each window runs about 22 trading sessions, so on one ticker windows 1, 2, 3, 4 weeks apart share 77%, 54%, 32%, 8% of their trading days (measured from the windows' own sessions). The tickers also move together: measured over the 250 sessions to Oct 27, 2026, their daily returns correlated 0.24 on average. Treating each window's move as a sum of independent daily moves, with Sheppard's formula for how two direction calls on overlapping windows covary, the 52 decided windows carry about as much information as 9 independent ones, so the 95% interval on the hit rate runs from 22% to 75%. Overlap within each ticker alone would leave about 19. These are derived figures: the model assumes independent, normally distributed daily moves, while real days cluster in volatility, and it takes up and down as equally likely, the cautious case. Read the interval as approximate, and expect it to narrow only as weeks accrue.