How obligation completion is scored
The EU AI Act completion figure is stage-weighted — Build & document carries 45 of 100 points — with priority weighting inside each stage, GDPR counted apart, and a hard zero for prohibited systems.
A system's EU AI Act completion figure is not a flat percentage of completed obligations. It is stage-weighted, because finishing classification paperwork is not the same as finishing the heavy build work.
Stage weights
Each stage contributes a fixed share of the 100 points:
- Classify — 10
- Scope & literacy — 5
- Build & document — 45
- Conformity assessment — 20
- Register — 10
- Operate & monitor — 10
So a system that has completed the first two stages but not started Build & document reads around 15, not 33 — the figure reflects where the real work is.
Empty stages are excluded
If a stage has no applicable obligations for your system (common for deployers), it is dropped and its weight is redistributed proportionally across the remaining stages. Role-specific journeys can still reach 100.
Priority weighting within a stage
Inside each stage, obligations are weighted by priority: high-priority items count 3, medium 2, low 1. Completing a high-priority obligation moves the stage — and the overall figure — three times as much as a low-priority one.
What does not count
- Obligations marked not applicable are excluded entirely.
- Incomplete obligations contribute nothing, whether overdue or simply not started — overdue items are additionally flagged and counted.
- GDPR duties are not in this figure at all. A system's GDPR duties are counted apart — its second readiness figure, done over total — so the two regimes are never added together.
Prohibited systems score zero
A system whose active classification is in the Prohibited risk tier is forced to zero, regardless of completed obligations — a banned practice cannot be compliant. Any prohibited system also zeroes the portfolio figure, and the Dashboard shows a banner saying the system must be withdrawn or reclassified before anything else means much.
Where the figure shows
The figure is recomputed on every save and stored on the system, where the per-system audit bundle and the machine API read it. The screens deliberately show counts with denominators instead: the AI systems register's EU AI Act column (obligations done over total), the system's EU AI Act journey card, and the Dashboard's Obligations done tile — because "14 of 22" is what a person can act on, and a single number is not.