How our tests are built —
and how honestly they're scored

1 · Where the questions come from

Our core assessments adapt items from open, peer-reviewed pools — the public-domain IPIP inventories for personality traits, the Experiences in Close Relationships (ECR) tradition for attachment, and published academic scales for conflict and communication styles. Each test page cites its sources. We write our own wording for clarity, keep positively- and negatively-keyed items balanced to reduce social-desirability bias, and add attention checks to longer tests.

2 · How scoring works

Your answers aggregate into dimension scores, which we place against population norms as percentiles — “higher than 72% of people” means exactly that, computed from the norm sample, not a compliment. Archetype names (The Tide, The Anchor…) are labels on top of your dimensional profile, never a replacement for it.

3 · Tests taken with the AI Coach score the same way

Taking a test by talking with the AI Coach is a different interview, not a different instrument. Your free-text answers are mapped to the same item bank with confidence thresholds — when the AI isn't sure, it asks again rather than guessing. AI-Coach results are tracked separately while we accumulate evidence that both modes agree, and we'll publish those statistics here.

4 · Reliability, in public

As soon as a test passes 100 completions we compute and publish its internal-consistency coefficient (Cronbach's α), recalculated monthly — including the ones that aren't flattering yet. An instrument you can't inspect is an instrument you shouldn't trust.

status: collecting — first α report expected after public launch