Claims should never outrun evidence.
These rules govern assessment labels, score interpretation, AI-assisted report language and corrections across Inward Atlas.
1. Separate source evidence from product evidence
A strong source instrument does not automatically validate a shortened, translated or newly presented implementation. Every methodology record states which claim belongs to the source and which has been verified for Inward Atlas.
2. Preserve continuous scores and uncertainty
Midpoint and balanced results remain visible. We do not force a type label when the score does not support one, turn response scores into unearned percentiles, or imply that a current pattern is permanent.
3. Keep AI downstream of scoring
AI may expand language around a minimized, deterministic facts package. It cannot change scores, receive item-level answers, diagnose, invent norms or replace the original report. Model output must pass structural and claim checks before it is saved.
4. Version and correct
Definitions have stable identifiers and schema versions. Source-check dates and source versions are shown when they are recorded. Material scoring changes should create a new version rather than silently rewriting an old result. Corrections can be reported through the contact page.
5. Avoid hidden commercial influence
Payment status changes access, not the score or evidence label. Recommendations are curated by assessment identity, not inferred from sensitive answers. Sponsored placements, if ever introduced, must be visibly labelled and kept outside score interpretation.