> 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/verdicts.md).

# Verdicts

The engine is a pure function of the extracted facts: no model, no network, no clock, no randomness (`server/engine-v2/deterministic-engine.ts`). Every fact arrives as `present`, `absent` or `unknown`.

## The three answers

| Verdict | MCP `action` | What it means                                                                | What to do                                                                                                              |
| ------- | ------------ | ---------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------- |
| PASS    | `PROCEED`    | Nothing failed and the idea is bounded.                                      | The check does not say the trade wins, and it did not check your facts. Those stay yours.                               |
| WAIT    | `HALT`       | The check could not settle a fact the verdict depends on. It does not guess. | Treat it as a no. Over MCP, `halt_reason` names the fields; write them down and check again.                            |
| FAIL    | `FAIL`       | A rule fired on what the idea says.                                          | Do not place it as stated. Each code maps to a fix: [Rulesets](https://docs.djzs.ai/how-a-verdict-is-made/rulesets.md). |

## The rule for prediction markets

Ruleset `DJZS-PM-v1.0`. In order, first match wins:

1. **FAIL** if any CRITICAL code fired, or if `risk_score` is 25 or more.
2. **PASS** if no code fired and all three scored facts are present: what would prove the bet wrong (`invalidation_condition`), engagement with the market's own terms (`resolution_engagement`), and a basis for the odds (`probability_basis`).
3. **WAIT** if any scored fact is `unknown`.
4. **PASS** otherwise. This is how M04 rides along: alone it weighs 15, under the line, so the verdict is PASS with the flag attached.

M04 is advisory and never changes a verdict. Over all 27 combinations of the three scored facts, the verdict is the same with and without it (`tests/pm-engine-integrity.ts`, property 9). It adds 15 to `risk_score`, so it does change `verdict_hash`.

## The rule for perp and spot

Ruleset `DJZS-LF-v1.2`. In order, first match wins:

1. **FAIL** if any CRITICAL code fired, or if `risk_score` is 50 or more.
2. **PASS** if no code fired, the position is bounded (a stop or an invalidation condition is present), and a reason for the trade is present (`thesis_statement`). Other unknown fields cannot change a no-flag verdict here, so a perp PASS can still list some.
3. **WAIT** if any scored fact is `unknown`.
4. **PASS** otherwise.

The perp scored facts: `leverage`, `position_size`, `stop_loss`, `take_profit`, `invalidation_condition`, `data_sources`, `oracle_source`, `confidence`, `thesis_statement` (`server/engine-v2/audit-input-schema.ts`).

## Why WAIT exists

A fact becomes `unknown` when the check cannot confirm it either way: the three readings of the text disagree, or a reading says it is missing without quoting the text that shows it ([Extraction](https://docs.djzs.ai/how-a-verdict-is-made/extraction.md)). The engine abstains instead of guessing. The way out of a WAIT is to fix the text: the same text checked again usually gets the same answer ([What is measured](https://docs.djzs.ai/evidence/measured.md)).

## The verdict hash

`verdict_hash` is `0x` + sha256 of canonical JSON (keys sorted, no whitespace) over four fields: the verdict, the score, the fired codes sorted, and the unknown fields in engine order. Nothing else is in it: not the text, not the time, not the caller.

```python
import hashlib, json
pre = {"verdict": "FAIL", "risk_score": 40, "flags": ["DJZS-M03", "DJZS-M04"], "unknown_fields": []}
print("0x" + hashlib.sha256(json.dumps(pre, sort_keys=True, separators=(",", ":")).encode()).hexdigest())
```

That prints `0x85918814b3dffa31b00d6892c2e00b2001efd35f7e0044b4cd3789fe1df14937`, the hash of the second [Quickstart](https://docs.djzs.ai/quickstart.md) example. Because the hash covers only those four fields, two different ideas with the same outcome share a hash; the input is bound separately by `intent_sha256` and, on paid checks, by `intent_hash` ([Receipts and certificates](https://docs.djzs.ai/how-a-verdict-is-made/receipts.md)).


---

# 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/verdicts.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.
