Ask a question in plain language. The engine plans which of its tools to call, runs them against Scrutica’s records, and cites the records they return. How answers are produced

Query Builder retrieves records and writes an answer with citations; a complex question can take several searches.

Save as

A saved conversation becomes a permalink anyone holding the URL can read; search engines do not index it.

55 tools; at most 8 data-tool calls per question

Your subject

Probed: — (resting; nothing probed)

All 55 tools, at rest.

55

tools the public engine may call, counted from the engine’s own definitions, so a tool added to or withdrawn from the server moves this number with it.

180

distinct values those tools accept, declared across 45 parameters that take a fixed list, from logic_fab to eu-ai-act. A subject matching one of them is a subject the engine can name exactly. Values are counted once however many tools declare them and parameters once per tool, so the two figures do not divide.

—

nothing probed yet. Type a subject above, or take one of these:

One mark per tool the engine may call, in catalogue order. A subject entered in the field above lights the tools whose names, descriptions or declared values match it. The match is lexical; the engine’s planner chooses its own tools, and answers subjects the catalogue declares nowhere.

How an answer is assembled

The engine writes a plan first: which tools it will call, in what order, and which calls wait on an earlier result. The page draws the plan as it arrives, and the calls then run, in parallel where they are independent. A question has at most 8 data-tool calls across every round; calls past that budget are not run, and the answer is written from what was gathered.

The answer cites the records the tools returned. The citations panel beneath it lists each citation with the tool, the record identifier, the cited claim and the authority tier, links the source where the tool returned one, and expands to show the provenance behind each row. After the answer has streamed, the server checks that each cited identifier appears in the tool results of that run and marks any it cannot find unverified provenance. This lookup does not check whether the cited record supports the claim. The model failover chain, the tag grammar and the matching rule are at the query-engine methodology.

Scope

The tool reference lists the available searches and analyses. When a requested figure is missing, the model is instructed to say so. Revenue projections, for example, require a source that supplies a forecast. The model can explain regulations and threshold calculations; case-specific legal advice is outside its remit.

The subject box on the figure above matches against the catalogue in the browser and runs no query.