Signals
Every fill starts as a signal.
A signal is an agent saying, in words and with a number attached, that the book is wrong right now. What separates a usable signal from noise is everything around it: the conditions that produced it, the limits it has to respect, and the record it leaves behind.
The pipeline
From live tape to logged fill.
01
Watch the book
An agent holds one thesis and reads the live book for it continuously — price, spread and depth from Kalshi, on the cadence its own strategy names. Nothing is polled on a timer that is slower than the decision.
02
Test the conditions
Momentum, mean reversion, breakout, or a rule set you wrote. The agent acts only when its stated conditions hold, so a signal is a claim about the tape, not a hunch rendered as a number.
03
Bind the guardrails
Position size, daily loss limit and venue exposure caps are checked before an order leaves. Breach one and the agent stops instead of asking for forgiveness afterwards.
04
Act and account
Approved signals execute over your own connected session and land in the same ledger as manual trades, attributed to the agent that produced them — so a thesis can be audited weeks later.
Live sample
Signals recently acted on.
Every row below is a real fill from the public Farm ledger in the last day — the action a signal produced, on a market you can open yourself.
Reading the last day of fills…
See a market’s full ledger ›Sample refreshed every minute · times in New York
Two ways to run it
You decide how much of it is automatic.
Paper
The agent runs its strategy against live prices and keeps its own book, so you see what the signals would have done before a dollar moves.
Live with approval
The agent queues the signal instead of firing it. You approve or drop it in the console; a hand in the trade never breaks the audit trail.