Skip to content
FRA Flow
Menu
SoftwareBS 9792AppPricingAssessorsToolsDocs
Try FRA Flow freeTalk to usLog in

The reviewer queue

How drafts reach a reviewer, what the queue row shows at a glance, and how a reviewer approves or rejects a draft.

By Richard Pryce·Last updated

The reviewer queue is where drafts wait for second-pair-of-eyes attention before they go to the landlord. This page explains how a draft reaches the queue, what the reviewer does there, and what limits apply pre-pilot.

How a draft reaches the queue

The assessor on the report viewer sees a Submit for review button once at least one paragraph has been generated. Clicking it flips the assessment status to "in review" and adds a row to the queue. The button is idempotent; re-submitting a draft that is already in review is a no-op.

The assessor still sees the report viewer in the same shape afterwards (with an "In reviewer queue" banner up top); they can keep editing paragraphs, adding observations, regenerating, and so on while review is in progress. Reviewers commonly ask for more evidence on a finding, and FRA Flow does not lock the workbench against that.

What the reviewer sees

Open Review queue in the sidebar (visible only to users with the reviewer or admin role). The sidebar entry carries a count badge so the reviewer knows at a glance how many drafts are waiting. Each row is one in-review assessment, showing at a glance:

  • The property name.
  • The assessor who submitted it, the building type (block of flats, HMO, etc.), and the active section count.
  • A relative timestamp ("submitted 25m ago").
  • Risk-band counts (high / medium / low observations) so the reviewer can triage which drafts deserve a deeper read first.

Rows are sorted oldest-first, so the draft that has been waiting longest sits at the top.

Click a row to open the report viewer for that assessment. From there:

  • Walk every paragraph top to bottom.
  • Tap Show source on any paragraph to see the observations the AI was given and the observations it cited. Yellow and red flags surface citation gaps and hallucinated citations.
  • Edit a paragraph in place if a sentence needs tightening.
  • Switch to human if the AI draft is unsalvageable.
  • Regenerate if the AI draft needs a fresh swing.
  • Open the source panel on the executive summary and the action plan introduction; these are the ones reviewers most commonly want a second look at.

Source linking and AI / human / edited modes cover the interactive parts of the review pass in detail.

Approving or rejecting

When the reviewer is happy with the draft they click Approve draft: the assessment status flips to signed, the row drops out of the queue, and the assessor gets a bell notification that the report has been signed off.

When the reviewer wants the assessor to make changes first they click Reject with note. A dialog opens for a short note (up to 2,000 characters) explaining what to change. Confirming flips status back to drafting, stamps the rejection, and the assessor sees three signals on next page load: a bell notification, a "Sent back Xh ago" chip on the visit row, and a rose banner on the report viewer with the note in full. See the sign-off page for the full two-action flow including audit-trail behaviour.

Pre-pilot scope

The queue today is intentionally minimal:

  • Lists every in-review assessment in the tenant. There is no "assigned to me" filter yet.
  • Implicit oldest-first sort; no manual sort or filter controls beyond that.
  • No reviewer-side notes that survive sign-off; the reviewer edits paragraphs in place rather than adding margin comments.

Pre-pilot, sign-off is the lightweight Approve action described above. The heavier sign-off pass (named-reviewer signature, hallucination-guard hard gate, PDF and Word export, audit lock) ships in M3.

Permissions

  • Assessor: can submit for review, but does not see the queue page in the sidebar (the entry is gated to reviewer + admin).
  • Reviewer: sees the queue page, can open any in-review assessment in the tenant, can edit paragraphs.
  • Admin: same as reviewer plus broader admin access.

What lands in M3

The basic two-action flow (Approve / Reject with note) ships today. M3 hardens the queue and the sign-off itself:

  • Named-reviewer signature. Approve captures the reviewer's name and a typed confirmation, stamped against the assessment alongside the existing paragraph audit.
  • Hallucination-guard hard gate. The Approve button blocks if any paragraph has high-risk evidence missing from the AI's declared input set, or if the model fabricated a citation.
  • PDF + Word export of the signed report, with the audit metadata embedded.
  • Per-reviewer assignment so each reviewer's queue is filtered to assessments routed to them, with assigned-to-me / all filter chips on top of the row list.
  • Paragraph-level threaded reviewer notes for the audit trail, capturing the reviewer's reasoning when overriding a flag or asking for a tweak below the rejection threshold.

Where to go next

Next

Source linking →