How Complipath works · Written by Yobel Tzegai · Last checked 9 October 2026

How are Complipath's EU AI Act rules tested and kept up to date?

Complipath (complipath.io) holds 509 test files with 5,840 test cases in its code, counted on 13 October 2026; 89 of those files test the rule engine or the pinned law texts. A change in the law reaches the rules only through a code release, and nothing monitors EUR-Lex automatically.

Check your AI systems, free

Which tests exist today?

509 test files with 5,840 test cases written in them, counted in the app's code at commit 5f85bc5 on 13 October 2026. Cases written as a table or inside a loop run once per row, so more run than are written.
44 files, 725 cases, read the pinned law texts: for example, that every quotation stands word for word in the pinned file, and that every point of Annex III has a path through the questions that is run against the engine.
45 files, 693 cases, run the engine or its rules without reading the law texts. The largest is the engine's own suite, with 85 cases.
420 files, 4,422 cases, test the rest of the app: screens, exports, plans and the paths that call an AI provider.
Where the line is drawn: a law test reads a pinned law text itself; an engine test imports the engine or its rules and reads no law text. A test that reaches the law only through another module counts as the rest.
Two of the 509 drive a real browser and run by name; the suite command runs the other 507. The app's deploy step runs three named test files, none of them an engine or law test.
One fixed case also runs through the engine a deployment is serving, on a page in the app, and shows whether the result matches the expected one.

How does a change in the law reach the rules?

Through a release of the app. The rules are code, and a change to them ships with a new engine version; today's is on the bar at the top of this page. A test pins the version, so changing it is deliberate.
The pin cannot force a new version when a rule changes. That is a practice, and the code says so: on 7 August 2026 a stored explanation changed without one.
The law texts are pinned files. A test fails if either changes by one byte, until a person re-derives the rules' coverage against the text now in force.
Amending acts are declared in code. One is declared: Regulation (EU) 2026/1744. A generator reads its 43 numbered points of change from the act's own Article 1, and a test re-derives that list from the pinned file.
A second act needs a code release, and a test fails if one is declared before the code can tell two acts' point numbers apart.
Once a day a scheduled job checks the provisions your confirmed records cite against the declared acts, and emails the workspace once per act for each affected system.
The email waits 14 days after an act's publication, unless your workspace asked to hear earlier, and it can be turned off.
On each record, the app runs your stored answers through today's rules and says when the risk level or the citations would differ. It does not rewrite the record or show the new result.

What is not done yet?

Nothing monitors EUR-Lex automatically. None of the app's four scheduled jobs reads it; the legal watch compares against acts already declared in the code.
Planned: an independent legal review of the rules. It has not been done. The passages the classification rests on were checked against EUR-Lex by the founder on 5 and 6 August 2026, which is not an independent review.
Planned: a test suite that runs the rules against the European Commission's checker. It does not exist, and no test in the app calls the Commission's checker.

Where can you see it?

On each classification: the acts it was derived from and the engine version that wrote it. How the class itself is decided: how Complipath decides a risk class.
In every export: the engine versions in it and each pinned law file with its full sha256, so a reviewer can check the text without asking us.
Who receives data, including the AI providers: the subprocessors page. What the AI switch stops and does not: Turn off all AI processing.
Each guide's "Last reviewed" date is the last day its text changed. Every change is checked by the site's own build against the pinned law texts — quotations, citation forms and dates — before it ships.

Starter

€499 a month

Prices exclude VAT.

Up to 20 AI systems
Your first system is free, with no end date

Check your AI systems, free

All plans

Questions

How many tests does Complipath have?

509 test files with 5,840 test cases written in them, counted in the app's code on 13 October 2026. 44 files read the pinned law texts and 45 run the engine or its rules; the rest test the app around them. Cases written as a table run once per row, so more cases run than are written.

Does Complipath watch EUR-Lex for new acts?

No. Nothing in the app reads EUR-Lex on a schedule. Amending acts are declared in the code, one today, Regulation (EU) 2026/1744, and a new act reaches the rules only in a release. Once an act is declared, a daily job checks what your confirmed records cite against it and emails you per affected system.

Has a lawyer reviewed the rules?

Not independently. An independent legal review of the rules is planned and has not been done. The passages the classification rests on were checked against EUR-Lex by the founder on 5 and 6 August 2026. Where a prohibition turns on a legal judgement, the engine returns Legal review required instead of deciding.

Which plan includes it?

Every plan, the free one included: the same engine, pinned texts and tests stand behind every result, and the legal watch email has no plan limit. Free holds one AI system. Starter is €499 a month for up to 20 AI systems. Prices exclude VAT. Growth and Scale are not available yet, and nothing here waits for them.