How the AI incident report is built from your own evidence, reviewed claim by claim with citations, and published only after two deliberate approvals.
Loading…
What this does not do
**nothing is ever auto-published** — draft to approved to published is human-only at both gates, enforced in the database and not just in the interface. A draft, not a filing; not an NFIRS report. Every release runs a safety gate that fails if a mayday goes missing from a report. No AI-vendor names in customer copy.
Writing the narrative is the part of the paperwork that eats an officer's evening. Coffee Table Tactics drafts a first pass from the incident record you already captured, and marks every sentence with where it came from — so reviewing the draft is checking citations against sources, not proofreading a stranger's guesswork.
Where to start
Open the AAR panel and click AI Incident Report. The button is available to any user-level account and above; a viewer or a member sees it disabled with an explanation. The report is generated from the incident itself — the tactical timeline, radio traffic, and curated responder input — so you can draft it whether or not an AAR exists yet, and on training incidents as well as real ones.
What the draft is built from
The draft fuses four sources: the logged tactical events, the radio transcript, officer-curated responder inputs from a contribution link, and the incident narrative. Every claim in the draft carries a citation back to one of these sources. If a section has no evidence to draw on, it says so — Insufficient data — instead of inventing something plausible. Nothing in the draft is left unmarked.
Redacted radio transmissions arrive as [redacted]. If an officer has flagged a transmission off-department, its transcript text is substituted before the draft is ever generated, so a redacted exchange never surfaces in report language.
Reviewing the draft
Each section carries one of three markers:
AI DRAFT — untouched AI output. Review it against its citation chips before you trust it.
AI · edited — you've edited this section. The AI-origin marking stays visible even after you've revised the wording, so the record of what changed is never hidden.
Insufficient data — no source signal supported this section. Fill it in yourself, or leave it as an honest gap.
The citation chips under a section name exactly which event, transcript, or curated input the claim traces back to. Click one to check it against the source.
Editing, then approving
Tap any section to edit it directly. Approving is a deliberate act, available to a department admin and above: it attests that you've reviewed every AI-marked claim and its source, and it unlocks sharing and export.
Editing after approval reopens the report to draft. Type into an approved report and it reverts to draft automatically, so an approved report can never quietly drift from what was actually reviewed — you edit, then re-approve. Reopen to edit does the same thing deliberately, without an edit. A published report is different: it cannot be edited directly, and the app tells you so rather than silently discarding your change.
Publishing and sharing
Approving unlocks Share and Export PDF; a draft is never shareable, and the server refuses one too. Share mints a public, view-only link that anyone with the URL can open, and you can revoke it later. The public page carries the same citation marking as your review screen: AI-generated content stays labeled, and the reader is told the draft was reviewed and approved by a fire officer.
Publish is the second deliberate act, separate from approval and also a department-admin action. It advances the report to published, and a published report can no longer be edited in place.
The safety check behind every release
Before any release of this feature ships, an evaluation gate runs the draft synthesis against a reference set of incidents and fails if a safety-critical event — a mayday, for example — goes missing without a citation. That check happens before the feature reaches you, not something you run yourself.