# forge.indicator.Indicator

```python
forge.indicator.Indicator(**inputs: Any) -> None
```

The one base class of an indicator (IND-001: declared by subclassing, never inferred).

Non-time bars: an indicator runs on whatever bars it is given. ``bars.kind`` and
``bars.params`` say which (``bars.kind == "renko"``); ``bars.close`` is the drawn price and
``bars.real.close`` the real one. Sessions, daily levels and anchored averages read
``bars.real.time`` and ``bars.volume`` (the real volume inside each bar). The forming brick is
never part of ``bars`` in a test. How candles are drawn (a Heikin-Ashi style) is a chart setting
and never changes ``bars``; an indicator reads the data's bar type, not a display style.

Bar form: a negative offset is refused when the file is checked and when it runs.

Warm-up: recursive indicators (exponential averages, Wilder smoothing) keep a memory of where
their data started. Each built-in's settle length in the registry is the history after which
more bars change its value by less than 0.1% of its range. An anchored indicator (a running
total) depends on its first bar.

Repainting: set ``repaints = True`` when your indicator changes values it already showed (a
swing marked once it is confirmed, for example). Forge then feeds strategies the value as it
was at each bar's close. Mark an output that deliberately reads later bars, such as a training
label, with ``causality_exceptions = {"target": "ml_label"}``: the check reports it, and
strategies cannot read it.

Since Forge API 1.