Review Method

Every Stewardship Review begins with the same ten questions.

1. Does this protect human dignity?
2. Does this respect attention?
3. Does this improve understanding?
4. Does this preserve context?
5. Does this preserve rationale?
6. Does it help quiet contributors participate?
7. Does it reduce unnecessary complexity?
8. Does AI remain assistive rather than authoritative?
9. Will someone understand this years from now?
10. Does this leave people more capable than before?

How a review is opened#

Use openReview({ id, subject, purpose, openedAt, openedBy, constitutionalAlignment }). The returned StewardshipReview is seeded with all ten canonical questions and an opened-history entry. Findings and recommendations are appended over time.

How a review is validated#

validateReview(review) runs the eight checks:

• uses-ratified-questions
• canonical-questions-present
• uses-ratified-outcomes
• no-verdict-language
• no-non-stewardship-metrics
• human-reviewer-required-to-conclude
• findings-reference-known-questions
• constitutional-alignment-declared

A concluded review with no named human reviewer, or one that references verdict language (Approved / Rejected / Failed) or non-stewardship metrics (popularity / engagement / retention / growth / virality / attention capture), fails validation.

What is never reviewed#

Popularity · Engagement · Retention · Growth · Time spent · Viral potential · Competition · Attention capture · Conversion · DAU · MAU · Click-through. These are enumerated in NON_STEWARDSHIP_METRICS and any review that references one fails validation.