What we collect
We collect
- A Skill Story: headline, up to 8 ESCO skills with evidence
- Languages, start date, work mode, permit need
- A random alias instead of a name
- Case sign-ups (2 at most) and team
- Observation cards from the company
- Each side’s private “continue?” answer
- Optional fairness questions, counted only in groups of 5+
We never collect
- Name, photo, age
- Gender, nationality, ethnicity, religion
- Address, social media, CV file
- Video or audio of the sprint
- Grades, personality traits, scores or rankings
Live work, not a CV
CVs are often written and read by AI, and carry proxies like a photo or a name. A company here sees 90 minutes of real work instead.
Humans decide
- People choose their cases. Nothing ranks a person.
- The organizer checks every team and can move anyone.
- Companies describe what they saw. No grades.
- A next step opens only with a mutual yes. A “no” stays private.
How teams are formed
- Sign-ups per case split into teams of 3 to 4.
- People with a must-have skill go to different teams.
- Only skills are read. Never language group or career path.
- The organizer can move anyone. Hand changes are marked.
Observation cards describe one sprint
Seen skills, ticked behaviours and a note of 200 characters. The person reads every card.
- Broke the problem into parts
- Asked for data or facts before deciding
- Explained an idea so others could follow
- Built on a teammate's idea
- Checked an idea against the case material
- Kept the team on time
The Proof Card belongs to the person
- The case, the company and the team
- The role the person chose
- The skills and behaviours the company saw
- The next step, after a mutual yes
Everyone gets one, whatever the outcome. It outlives the event data.
EU AI Act: recruitment is high-risk
Annex III, point 4. Our rules are simple, but we design as if every duty applies.
| Human oversight (art. 14) | People choose cases. The organizer can move anyone. A next step needs both sides. |
|---|---|
| Transparency (art. 13) | A one-sentence team rule. Fit shown as skill overlap, never a score. |
| Record-keeping (art. 12) | Every sign-up, team move, card and decision is recorded. |
| Data and bias (art. 10) | Cases are checked for proxy requirements. The fairness monitor hides groups under 5. |
2 cases each, against mass applications
At most 2 sign-ups per person, so every team is real. Pool previews hide groups under 5.
How to run it for real
- Suomi.fi sign-in for organizers, email links for others
- Data in the EU, the City of Espoo as controller
- Event data deleted after 30 days, Proof Cards kept
- DPIA and AI Act documents before the first pilot
- A 10-minute briefing for company staff
- Event pass works offline
Limits we know about
- The skill list is a small part of ESCO.
- Ninety minutes shows one setting, not everything.
- Teams can be uneven. The organizer can move people.
- Small events hide most groups in the fairness monitor.
- This demo uses synthetic data only.