Coverage

What this check can and cannot decide

The list of what Complipath's check can and cannot decide is never generated per visitor. It is written once, by a person, against the engine's actual coverage, and versioned with it.

Written 2026-08-10 by Yobel Tzegai, against engine version 2026-08-09.1.

The corpus it was derived from

Instrument	Known as	In force	Status
Regulation (EU) 2024/1689	AI Act	—	Law · superseded in part
Regulation (EU) 2026/1744	Digital Omnibus on AI	27 July 2026	Law
Commission Guidelines on the definition of an artificial intelligence system established by Regulation (EU) 2024/1689 (AI Act)	—	—	Guidance, not law

a810dc0785e21127fb55d8d3d64ed7a0c00591b778cfa7f2fa8a60e6a1c21327

One pinned hash, shown in full. Every instrument in the corpus is published with its own. The day the corpus took in Regulation (EU) 2026/1744 and every citing guide was corrected against it is dated in what changed, and when.

What it decides

✓

Whether a system falls under one of the prohibited practices in Article 5(1), points (d), (e), (f), (g) and (h).

These five turn on facts about what the system does, not on a judgement about degree.

✓

Whether it matches an Annex III category, and which point.

Annex III enumerates use cases, and the assessment questions map onto them.

✓

Which transparency obligations under Article 50 attach.

They follow from what the system is and who it interacts with.

✓

Which obligations follow from the role you stated.

Provider and deployer carry different duty sets; once the role is stated, what follows is derivable.

✓

Which version of the legal text the assessment was made against, and which provisions have been amended since.

Every assessment is made against a specific text, and a comparison against the current one shows what has moved.

What it does not decide

This list is the reason to trust the first one. Nothing moves from here to there because the engine mostly handles it — a partial capability stays here, with the part it cannot do named.

Whether Article 5(1), points (a), (b) and (c) apply.

They turn on judgement — materially distorting behaviour, exploiting a vulnerability, a social score that is unjustified or disproportionate. These are flagged for legal review instead of being decided.

Whether you are the provider or the deployer, where you have not said so.

The role turns on facts about what you built, what you changed and under whose name it ships. It is asked, never inferred.

Whether a general-purpose model you build on crosses the systemic-risk threshold.

That is a fact about the model, held by its provider, not something visible from your system.

Whether Article 6(1) applies.

The product route turns on two conditions that have to hold together — a safety component of, or itself, a product covered by the Union harmonisation legislation in Annex I, and that product being required to undergo a third-party conformity assessment. It needs its own assessment.

Anything that depends on facts about training data or intended purpose that you have not supplied.

The engine works from what it was told. An unanswered question is not a permission to guess.

Whether something is an AI system at all comes before any of this — the qualifying test is in Article 3, point (1). The order the classification runs in is here. These lists are about one thing the engine does; what the product covers of the Act article by article is the wider question, and five of its rows say it covers nothing. Why a record rather than an answer from a model is a different question again.