> ## Documentation Index
> Fetch the complete documentation index at: https://docs.blackquantlabs.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Signal Engine
> The live feed, the measured statistics behind each strategy, and how to read them without fooling yourself.
**Apps → Signal Engine.** The live trade feed and the measured performance of every tracked strategy.
**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](/security/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:
| Field | What it is |
| ------------------------ | ---------------------------------------- |
| Chain and DEX | Where the pair trades |
| Pair address | The contract, linkable to an explorer |
| Price / liquidity / FDV | Market context at the time of the signal |
| Buys / sells / buy ratio | Flow behind the observation |
| Age | How long the pair has existed |
| Origin | `tx`, `trade-id`, or `simulated` |
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
| Basis | Meaning |
| ------------- | ---------------------------------------------------------------------------------------------------------------------- |
| `calibration` | The fullest record — every resolved outcome, including ones restored from previous runs and ones reported over the API |
| `stats` | This run only |
| `declared` | **A claim, never a result** |
| `none` | Nothing measured yet |
**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.
The calibrated figure leads throughout. The raw score is what the engine *claimed*; the calibrated one is what that claim has *historically been worth*, and it is the one to act on.
### `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.
| Displayed | Resolved samples | What it tells you |
| -------------- | ---------------- | -------------------------------------------------- |
| `75% [45, 93]` | \~4 | Almost nothing. The true rate could be a coin flip |
| `75% [72, 78]` | \~400 | A real measurement worth acting on |
Both are "75%". Only one is worth anything.
If you take one thing from this page: **read the interval, not the headline percentage.** A wide interval means the strategy has barely been measured, regardless of how good the midpoint looks.
### 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:
| Assumption | Why it matters |
| -------------------------------- | -------------------------------------------------------- |
| Starting capital | Scales everything |
| Risk per trade | The single biggest driver of a drawdown figure |
| Fee basis points | Omitting fees flatters every result |
| Max leverage | Changes the shape of the tail |
| Sizing (`calibrated` or `fixed`) | Calibrated sizing uses the measured edge; fixed does not |
A simulated result without its assumptions attached is not interpretable, so they travel together.
## Strategy metrics
| Metric | Note |
| -------------------------- | -------------------------------------------------- |
| Total P\&L % | Net of fees where fees are modelled |
| Gross win % / gross loss % | Separated, so profit factor is reconstructable |
| Profit factor | `null` when there are no losses yet — not infinity |
| Best / worst trade | Tail behaviour |
| Max drawdown % | Peak-to-trough |
| Longest loss streak | What you would actually have had to sit through |
**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](/billing/plans-and-pricing). Daily signal limits by tier:
| Tier | Signals/day | Active positions |
| ------- | ----------- | ---------------- |
| Starter | 10 | 1 |
| Growth | 50 | 5 |
| Elite | Unlimited | Unlimited |
## Related
* [Positions](/platform/positions) — what happened after you acted
* [Key concepts](/introduction/key-concepts#credible-interval)