> For the complete documentation index, see [llms.txt](https://nfa.gitbook.io/nfa-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://nfa.gitbook.io/nfa-docs/the-forecasting-engine/anchoring.md).

# Anchoring

## Anchoring

Left alone, a language model recalls a probability from training. Anchoring replaces that recall with something firmer: either a historical reference-class base rate, or a current measured value from a data source. An anchored forecast argues from a real number, not a memory.

### Two kinds of anchor

{% tabs %}
{% tab title="Base rate" %}
A reference-class prior for the question, drawn from the historical outcomes of similar events. For price-driven markets the prior comes from a quantitative [price model](/nfa-docs/live-data-and-adapters/price-models.md) instead of an archive.

The base rate is injected **to the judge only**. It is not placed in the shared advocate brief, because letting the advocates argue from the same prior the judge already holds would double-count it. When a base rate is present, the judge is told to anchor on it rather than recall from memory.
{% endtab %}

{% tab title="Live metric" %}
A current measured value fetched by a structured data source: an inflation nowcast, a policy rate, a temperature, a filing figure, a territorial-control status. The live metric is a **fact**, so it goes into the shared brief that both the advocates and the judge see. Facts are inputs for everyone; only a model's own prior output is restricted.

A live metric served by a structured source is what earns a forecast the blue anchored quality tier. A value found by a plain web search does not.
{% endtab %}
{% endtabs %}

### Why the fact is also given to the verifier

When a live metric anchors a forecast, the same fact is handed to [self-verification](/nfa-docs/the-forecasting-engine/self-verification.md) as verified ground truth. This closes a subtle failure: without it, the verifier could grade the engine's anchored, correct divergence as unsupported and veto exactly the best forecasts. The anchor is first-class evidence on both sides of the process.

{% hint style="info" %}
Anchoring is controlled per deployment and per run. Each run records which anchors were active and what value they carried, so a forecast's provenance can always be reconstructed from its own record.
{% endhint %}


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://nfa.gitbook.io/nfa-docs/the-forecasting-engine/anchoring.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.