> For the complete documentation index, see [llms.txt](https://docs.glbnxt.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.glbnxt.com/platform/mcp-servers/dictum/quality.md).

# Quality

DICTUM measures itself every night: a set of questions with verified answers runs against the live search, and the result is stored with the deployed commit. The numbers below are the latest nightly run, put on this page every morning. They are the state of the method, not a marketing claim; the question set is a draft until a lawyer validates every answer key (roadmap P1.1a), and we will say so here until then.

### What is measured

* **Recall\@k**: is the expected ruling (or article) among the top k results? Reported for k = 1, 3 and 10, overall and per source (Hoge Raad, Raad van State, gerechtshoven, rechtbanken, tuchtrecht, Court of Justice, laws).
* **MRR**: the mean reciprocal rank of the expected result.
* **Citation accuracy**: of the questions found whose key names the decisive passage, how many return that passage verbatim among the quoted paragraphs. This is the number that says whether a quote can be checked, not only whether the right file was found.
* **Latency**: p50 and p95 of a search, and of the lookups an assistant calls next.

### Numbers

From the nightly run of 2026-11-22 (commit 93f7f78), 200 questions, draft keys.

| Set                                     | Questions | Recall\@1 | Recall\@3 | Recall\@10 | MRR  | Citation accuracy |
| --------------------------------------- | --------- | --------- | --------- | ---------- | ---- | ----------------- |
| Rechtspraak (eight courts and colleges) | 120       | 0.44      | 0.59      | 0.78       | 0.54 | 52 of 67 (0.78)   |
| Wetten (BWB articles)                   | 40        | 0.43      | 0.78      | 0.78       | 0.58 | 19 of 19          |
| Lokale regelgeving (CVDR)               | 20        | 0.60      | 0.90      | 0.90       | 0.73 | 9 of 9            |
| Kamerstukken                            | 20        | 0.75      | 0.90      | 0.90       | 0.83 | 14 of 18 (0.78)   |

Per source, recall\@10 on the rulings set: Hoge Raad 0.72 (47 questions), tuchtrecht 0.65 (20), Raad van State 0.67 (12), Court of Justice 1.00 (10), rechtbanken 1.00 (10), Centrale Raad van Beroep 0.75 (8), gerechtshoven 1.00 (8), College van Beroep voor het bedrijfsleven 0.80 (5). Search p50 about 1.9 s, p95 about 3.4 s.

Hybrid search over the whole corpus of about 1.1 million rulings. A miss is a question whose expected document is not among the top ten; the misses are the material for the ranking work on the roadmap. Every question was written from the document's own text and every key carries a verbatim phrase of its decisive paragraph that was checked against the database; what the set still lacks is a lawyer's review of the keys.

### What it does not measure yet

Whether the assistant's memo is correct. That needs a validated answer key and a judge, and comes after the 200-question set is reviewed (roadmap P1.1d). Until then, the [Research method](https://docs.glbnxt.com/platform/mcp-servers/dictum/method) is what keeps answers checkable: every claim carries an identifier and a verbatim quote.


---

# 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.glbnxt.com/platform/mcp-servers/dictum/quality.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.
