MaveraMavera Docs
Operator's Guide
Checking…
Use or share this pageSend the page itself, print it, or open its complete context in your AI workspace.

Review queue & quality

What lands in Review, and how Mavera keeps itself honest.

1 min readDocs build Updated Jul 26, 2026
Interactive product walkthrough

Work the Review queue

Inspect the proposed change, diff, evidence, decision, and audit trail.

Simulation onlyNothing is saved or charged
1
Step 1 of 6Open Review

Choose Review from the sidebar.

Demo workspace / Mavera workspace
Local simulation
Mavera workspace

Good afternoon

Choose a surface from the sidebar to begin.

Demo workspace0 actions completed
WorkspaceDemo companySafe local simulation
Current surfacereviewFollow the highlighted action
Recent workspace activityStatus
Audience comparisonReady to review
Customer interview evidenceGrounded
Campaign pre-flightDraft
  1. 1
    Open ReviewChoose Review from the sidebar.
  2. 2
    Open a proposalOpen the first simulated review item.
  3. 3
    Compare current and proposedInspect the exact field-level change.
  4. 4
    Open evidenceRead the source and confidence behind the proposal.
  5. 5
    Make the decisionApprove or reject the local-only change.
  6. 6
    Read the audit trailOpen the resulting audit entry.

The Review queue

Some outputs want a human before they become "the record" — document updates proposed from research findings, mostly. They land in Review (left sidebar, with a count badge).

Click any item to see exactly what's proposed: the current text and the proposed text side by side, plus the evidence behind the change and its grounding strength. Approve or dismiss — nothing changes without you.

How Mavera stays honest (the short version)

  1. Numbers are computed, never generated. When you see "62% preferred Variant A," code counted 62% of actual synthetic responses. The model writes words; arithmetic comes from arithmetic.
  2. Everything carries provenance. grounded → built from your real data. category-modeled → built from category knowledge. estimate → labeled guess. The badge travels with the result everywhere it goes.
  3. Refusal over invention. Audiences decline questions outside their profile. Mave says "I don't have data for that" instead of manufacturing an answer.
  4. Web claims cite sources. Programs and audits link every external claim to where it came from.

When a number looks off

Click into the run (My Research → the item) — the transcript shows every step, what was asked, what came back, and how the number was computed. If something's still wrong, the conversation is the interface: tell Mave, she'll re-run it.

Workspaces, team & billing

Multiple workspaces, roles, invites, spend controls, and where the money goes.

The prompt library

Copy-paste prompts for the fifty most useful jobs, organized by what you're trying to do.

On this page

The Review queueHow Mavera stays honest (the short version)When a number looks off