// every call scored · nothing edited
AI market calls,with the receipts.
Signal reads price history, indicators and the day's news across US stocks and major crypto, then makes a specific call: buy, sell or hold, with a target and a stop. Every one is scored later against what the market actually did, and the whole record is available after sign-in.
This is not financial advice.
Nothing here is a recommendation to buy or sell anything. A language model cannot predict prices, and no track record — good or bad — makes the next call more likely to be right. What this app offers is a stated hypothesis and an honest account of how such hypotheses have turned out. Treat it as a way to study how a model reasons about markets, not as a reason to move money. You can lose it.
- BTC · buy · target +6% · stop −3% · 7d
- fills at the next open, never the model’s price
- a bar touching both levels → the stop
- hold · scored, excluded from the hit rate
- 82 assets · one call each per UTC day
- as_of 09:31 UTC · says when it is stale
- append-only · a record you cannot edit
- not financial advice
- 82
- assets watched
- US stocks and major crypto, one call each per UTC day
- 1,000,000
- AI tokens a day for the cron
- checked before every call in the run, not once per run
- 40,000
- AI tokens a day, each
- a public signup can never run up the bill
- 5,000
- tokens an analysis is budgeted
- pre-charged before the call, so an overshoot is one call wide
// watch
Eighty seconds with Ticker.
A watchlist of crypto and U.S. stocks, the daily brief, a call with its target, stop and confidence, how every call is scored, the track record and the headlines a call cites.
Signal in eighty seconds, with Ticker · 1:19
// why
Anyone can have an opinion about a chart.
What makes one worth reading is a published account of how the last hundred turned out. Most market commentary never goes back to check; the ones that do check their own homework.
Here the scoring is one function in the database, written to be unfakeable — and where a rule could favour the model or the reader, it favours the reader. The record is append-only. A call, once made, cannot be edited into a win.
// how it works
Read, call, then let the market mark it.
step 1
It reads, then it decides
Price history, computed indicators, the last 48 hours of headlines and the crypto fear-and-greed reading go into one brief. The cited headlines are stored with the call, so a year later you can still see what it was looking at.
step 2
The market scores it
Fills at the next open, never at a price the model chose. A bar that touches both target and stop counts as the stop. A gap through a level fills at the gap. When the rules could favour the model or the reader, they favour the reader.
step 3
The whole record is available
Win rate, target-hit rate, median return, worst drawdown, average days to resolve. Broken down by asset, by month, by how confident the model claimed to be — which is how you find out whether its confidence means anything.
// the record
Three numbers, none of them chosen.
An example of what the record page shows — these figures are illustrative, not a live track record. Holds are scored for information and excluded from the hit rate, and the equity curve is one unit per call, never compounded.
54%
hit rate
target hit before stop
2%
median return
per resolved call
−11%
worst drawdown
on the equity curve
example figures · the real record lives in the app
// what you get
Built to be checked, not believed.
// get started
Read the calls. Then read the record.
An account shows you today's calls and the whole history behind them. It moves no money and never will.