Public scorecardStill being tested

Our calls, graded in public.

Each call is locked before the market reacts, then judged on real closing prices once its time is up. Hits and misses stay here for good. Updated every trading day.

Not proven yet. Foghorn won't call itself proven until it has 300 graded five-day calls that beat every simple guess below. Please don't trade on these.

43 of 300 five-day calls

Right, next 5 days25 of 4260% of calls where the stock moved · 1 flat

vs just guessing up−4 ptsit scored 64% on the same 42 calls

vs a coin flip+15 ptsit scored 45% on the same 42 calls

Simple guesses beaten1 of 5behind 2, level with 2: all shown below

Against simple guesses, on the same calls

On calls for the next 5 trading days, Foghorn is ahead of 1 of 5 simple guesses, behind 2, level with 2. If it can't beat these, it isn't worth paying for.

Foghorn
Our calls
60%42 that moved
Just guessing up
Calling every stock up, on the same days
64%42 that movedBehind
Coin flip
Up or down at random, fixed by each call's fingerprint so anyone can redo it
45%42 that movedAhead
Following the last week
Calling the way the stock moved in the 5 days before
60%42 that movedLevel
Following the sector
Calling the way its industry moved in the 5 days before
76%42 that movedBehind
The AI's read on its own
The AI's up or down from the news alone, where it read the story · Foghorn 61% on these calls
61%41 that movedLevel

A tougher test

Foghorn, against the S&P 500
Right only when the stock beat the S&P 500 (for an up call) or trailed it (for a down call)
62%39 that moved

With 42 calls where the stock moved, Foghorn's true hit rate is likely between 44% and 73% (95% range). It narrows as more calls are graded.

Does “how sure” mean anything?

Every call says how likely it is to be right. Each group should come true about as often as it says. Both call lengths together.

When it said “a bit better than even”98 that moved
Said about 54%, right 62% of the time
When it said “about 6 in 10”11 that moved
Said about 58%, right 64% of the time

The bar is how often calls came true; the black mark is what they said. Groups with few calls swing a lot.

Every graded call

Calls for the next 5 trading days, most recently judged first. Open calls appear once they're graded, and once every other call on the same news is graded too, so none gives an open call away.

Judged Oct 28CBChubbCalled Uphow sure: a bit better than evenThe news: Chubb 8-K: Results of operations (Item 2.02)The stock barely moved (+0.2%)Result: FlatLocked Oct 20, 6:18 PM ET , before its first judged close
Fingerprint 0393e9…b3230892
0393e9dfafaee50f3e0217a766c50676e85cd0464fdf0105175d1158b3230892

Chain check: all 271 calls in the ledger match their fingerprints

Each call's fingerprint includes the one before it, so no past call can be changed without breaking every one after it. Open calls are in the ledger too; they appear here once graded.

Latest daily stamp: Oct 30 · in Bitcoin block 973,872

How the proof works

What counts. Calls made from Oct 7, 2026 on (scorer version stage_a-7 and later). The 89 calls made by earlier test versions stay in the ledger, but they aren't counted or listed here.

How a call is judged. From the first closing price after the call was locked to the close one or five trading days later, adjusted for dividends and splits. A move of 0.2% or less (next day) or 0.5% or less (five days) counts as flat and is left out of the hit rate. Each simple guess is scored on the same calls, and only where it has the data it needs.

General market commentary, made the same way for everyone. It isn't investment advice or a recommendation to buy or sell, and past accuracy does not promise future results.