Docs / Set up your org

Roles & permissions

Every member of an org holds exactly one of four roles: Org admin, Tech reviewer, Feature author, or Observer. This is the admin’s reference for precisely what each role can and can’t do, and where you assign and change it.

A role is your single capability level. It travels with you across every project you can reach and decides what you’re allowed to do there: create, comment, approve, or administer. (A separate dimension, your team membership, decides which projects you can see at all; that is covered under roles, teams & project access.)

The cumulative ladder

The four roles form a single ladder, not a grid of independent toggles. Each role can do everything the role below it can, plus more, so a tech reviewer can do everything an author can, and an author can do everything an observer can. You only ever need to know how far up the ladder a member sits.

What each role can do

Read each row as “this, plus everything in the rows beneath it”:

RoleCan
Org adminThe full org. Configure org settings and AI provider/key/model; invite, remove, and re-role members and revoke pending invites; create and manage teams, projects, and services; author norms, safeguards, templates, and personas; delete features. Admins also see every project regardless of team, plus everything the roles below can do.
Tech reviewerEdit, approve, and reject analysis documents and REASONS canvases, deciding at both review gates (and on through partially approved, approved, rejected, and archived). Can also do everything a feature author can, including on features they didn’t author, plus everything below.
Feature authorCreate features and edit, submit, and version their own features; can’t approve any artifact, and can only modify features they authored. Plus everything an observer can do.
ObserverRead every project, feature, analysis, and canvas they have team access to, and comment on any artifact under review. No create, edit, or approve.
Commenting is the one action open to every role: any member with project access, observers included, can comment on the artifact under review. What separates the roles is everything else: who can create, who can edit, and who can approve.

Approval, and rejection, stops at reviewer and admin

Only tech reviewers and admins can approve an analysis or a canvas, the transitions that push a feature past its two human gates. An author can create the feature, write its definition, submit it for analysis, and version it, but the Approve button never appears for them: a separate person has to sign off. The same restriction applies to the two ways a reviewer can decline: Reject and Reject & rework are also reviewer-and-admin-only, so an author or observer can comment on a feature but never end it or send it back to the AI.

This is the no-self-approval consequence in practice: because authoring and approving are different rungs on the ladder, the person who wrote a feature can’t be the one who waves it through. A gate only means something if a second set of eyes clears it.

Assigning and changing roles

Roles are set and changed by an admin from Org settings → Members:

  1. You pick a member’s role when you invite them by email, and they join with that role the first time they sign in with that address.
  2. To change someone later, open Org settings → Members and set their role to any of the four; the change takes effect immediately on their next read or write.
  3. Removing a member from the org revokes their role entirely; revoking a pending invite cancels a role that was never claimed.
The very first admin isn’t assigned by anyone; the person who creates the org is bootstrapped as admin automatically. From there, admins assign every other role.

Enforced server-side

These limits aren’t just UI niceties. Every member’s role lives on their record inside the org, and REEZN’s server-side permission checks read it on each write: feature creation requires admin, reviewer, or author; editing someone else’s feature or approving an artifact requires reviewer or admin; managing org settings, members, projects, norms, safeguards, templates, and personas requires admin. The interface hides what your role can’t do as a convenience, but even a direct request that slipped past the UI would be refused. The rules are the real boundary.

Why this exists
Folding the four roles into one cumulative ladder keeps the model legible: you reason about a single level per person, not a matrix of toggles. Limiting approval to reviewers and admins is what gives the two review gates their teeth, since an author can’t self-approve, and enforcing all of it on the server rather than in the interface means the boundary holds no matter how a request arrives.