Methodology

Key takeaways

• 13 content packs with 139 obligations, each with a version and verification date.
• Scope rule first, then tier rules in a fixed order (first match wins), then obligations for your role with the first matching deadline rule.
• 125 golden test cases run against both the TypeScript and Python engines in CI.
• Language models review documents and draft text; they never decide a tier, an obligation or a deadline.

Rules are data

Each law is a content pack: the questions, a scope rule, ordered tier rules, and the obligations with their citations and deadline rules. A pack has a version and the date it was last checked against the official text.

How a result is computed

• Scope: the pack’s scope rule is evaluated first. For the EU AI Act it checks for an EU link (Art. 2); if it fails, the result is out of scope with the stated reason.
• Tier: tier rules run in the order the pack lists them and the first match sets the tier. For the EU AI Act: Prohibited practice, then High-risk (Annex III), then High-risk (Annex I, Section B product), then High-risk (Annex I), then Potentially not high-risk (Art. 6(3)), then Transparency duties (Art. 50), then Minimal risk. If no rule matches, the engine stops with an error.
• Obligations: included when they apply to the selected role and their condition holds.
• Deadlines: ordered deadline rules per obligation (for example Digital Omnibus deferrals); the first match gives the date, and a missing match is an error rather than a dropped deadline.
• Trace: the IDs of every rule that fired are recorded with the pack version.

A worked example

Fictional company, SaaS vendor outside the EU selling a CV-screening and candidate-ranking tool to EU employers (Role: provider; Established in the EU: no; Placed on the EU market: yes; Output used in the EU: yes; Annex III 4(a): recruitment and candidate selection; Profiles people: yes): High-risk (Annex III) (Art. 6(2), Annex III). Rule trace: scope-eu-nexus, tier-high-risk-annex-iii, obligation:art4-ai-literacy, deadline:art4-ai-literacy:standard, obligation:hr-provider-conformity, deadline:hr-provider-conformity:annex-iii, obligation:authorised-representative, deadline:authorised-representative:annex-iii. Obligations: Support the AI literacy of staff operating the system — Art. 4 (as amended by Reg. (EU) 2026/1744), in force since 2 February 2025 (wording amended 27 July 2026); High-risk provider obligations (risk management, data governance, technical docs, logging, oversight, accuracy, QMS, conformity assessment, CE marking, registration, post-market monitoring) — Arts. 8-21, 43, 47-49, 72-73, 2 December 2027 (Annex III, deferred by the Digital Omnibus); Appoint an authorised representative in the EU — Arts. 22, 54, 2 December 2027 (Annex III, deferred by the Digital Omnibus).

Golden tests

The free browser checks run a TypeScript engine and the app runs a Python engine over the same packs. All 125 golden test cases run against both in CI, and a drift test fails if the bundled copies of the packs differ from the canonical files.

Known limitations

• Results depend on your answers; the engine cannot check that they are accurate.
• EU AI Act Art. 2 exclusions (military or national-security purposes, research and development, pre-market testing, personal use) are not screened.
• The Art. 6(3) derogation is applied only on your own answer (a stated Art. 6(3) condition and no profiling gives "Potentially not high-risk (Art. 6(3))"); the engine does not check it, and you still have to document the assessment and register the system (Arts. 6(4), 49(2)).
• Statutory exceptions to prohibited practices require legal review.
• Each pack reflects the law as of its verification date.
• Compliance guidance, not legal advice.

Report an error

Email support@shieldra.ai with the pack name, your answers, the result you got and the provision you believe applies. Confirmed errors are fixed in a new pack version with a golden test case that covers them.