> For the complete documentation index, see [llms.txt](https://docs.djzs.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.djzs.ai/how-a-verdict-is-made/extraction.md).

# Extraction

A model reads the idea and reports facts. It never scores and never returns a verdict. Source: `server/engine-v2/extraction-layer.ts`; model `claude-sonnet-4-6` at temperature 0 (`djzs-trust-mcp/src/verify-pm-trade.ts`).

## Three readings, one answer per fact

The text is read three times (N=3). Each reading reports every scored fact as `present`, `absent` or `unknown`. The three are merged per fact:

* All three agree: that state stands.
* Any disagreement: the fact becomes `unknown`.

A fact that is `unknown` cannot fire a rule, so disagreement can only push a verdict toward WAIT, never toward a false FAIL. Over MCP, the `disagreements` field lists which facts split.

## Quote gates

`absent` is a strong claim, so it needs evidence:

* **Every absent needs a verbatim quote** from the idea. A reading that says a fact is missing without quoting the text that shows it is demoted to `unknown`.
* **Critical absents need identical quotes.** A fact whose absence would fire a CRITICAL code merges to `absent` only if all three readings carry the same quote. Otherwise it becomes `unknown`, and `disagreements` lists it as `<field>(evidence)`.
* **Engagement is checked against falsification.** A quote offered as engagement with the market's terms that is really the falsification clause is demoted to `unknown`.
* **The basis quote is one span.** For the odds-basis fact, an absent quote must be one contiguous span of the text.

There is no keyword shortcut. Since contract `DJZS-X-v1.1` (2026-09-01), an unsourced probability does not need a `%` or the word "odds" to be caught; "everyone knows" is an unsourced certainty claim like any other. The verbatim quote is the whole evidence requirement.

## Contracts

| Contract         | Used by                                           | What it extracts                                                                                                                             |
| ---------------- | ------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------- |
| `DJZS-X-v1.1`    | prediction markets (`verify_pm_trade`, free `pm`) | `invalidation_condition`, `resolution_engagement`, `probability_basis`, plus the context and consensus facts M04 reads                       |
| `DJZS-X-LF-v1.2` | perp and spot (`verify_perp_trade`, free `perp`)  | The perp facts in [Verdicts](https://docs.djzs.ai/how-a-verdict-is-made/verdicts.md), including `thesis_statement` with a quote-gated absent |

Every response names its contract under `taxonomy.extraction`. A contract change is versioned and measured before it ships ([What is measured](https://docs.djzs.ai/evidence/measured.md)).

## When extraction fails

If every reading fails, the result carries `extraction_failsafe: true` and the facts are `unknown`, which makes a WAIT. On the free route a run that cannot complete returns `503 audit_unavailable` and gives the day's slot back.


---

# 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 by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.djzs.ai/how-a-verdict-is-made/extraction.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

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.
