This API places no orders. Nothing reachable from the signal engine can move money. It watches, measures and publishes — execution is a separate layer that you initiate. See The custody model.
When the engine is not connected
If a deployment has no signal engine configured, this screen says it is not connected. It does not fall back to sample data or a plausible-looking chart. That is a deliberate choice: a screen that invents numbers when its data source is down is worse than a blank one, because you cannot tell the difference from the outside.The live feed
Signals stream over a server-sent event connection, so the feed updates without a refresh. Each signal carries:
The origin matters. A signal derived from a real on-chain transaction and one produced by a simulation are both shown, and they are labelled differently.
Reading the numbers honestly
This is the part worth slowing down for. The screen distinguishes three very different things that most platforms collapse into one number.Basis: what a rate actually rests on
A declared rate is never promoted into a measurement. If a strategy asserts it wins 80% of the time, that assertion is displayed as a claim and stays labelled as one no matter how long it sits there. It never quietly becomes a measured figure.
null is not zero
A win rate is published as null until something resolves. The platform never defaults that to 0%.
“Not measured yet” and “measured at 0%” are entirely different claims, and only one of them is an indictment of the strategy. Collapsing them would make every new strategy look like a failing one.
Credible intervals
Every published rate carries a 90% credible interval — the range containing the true rate with 90% probability, given what has been observed.
Both are “75%”. Only one is worth anything.
Reliability bins
Strategies are also broken into bins by claimed confidence, each with its own signal count, win count and measured rate. This answers a sharper question than an overall win rate: when this strategy says it is confident, is it actually right more often? A well-calibrated strategy shows rising measured win rates across ascending bins. One that does not is confident at random.Simulated performance
Where equity curves are shown as simulated, the screen carries the assumptions that produced them:
A simulated result without its assumptions attached is not interpretable, so they travel together.
Strategy metrics
Longest loss streak is the number most people skip and most need. A strategy with an excellent win rate and an eleven-trade losing streak is one most people abandon at trade seven.
Timeouts
Requests to the engine time out after 5 seconds. Long enough for a busy engine, short enough that a dead one is a clear error rather than a hanging page.Access
The live feed requires an active signal plan. Daily signal limits by tier:Related
- Positions — what happened after you acted
- Key concepts