Public, high-volume, and where a wrong answer has a clock.
Recruitment, eligibility, procedures — the answers citizens act on fastest, and the ones most likely to be broken. Everything a ministry workspace has, plus releasability controls and an incident clock for when speed is the deliverable.
Empanelment, tender or purchase order — we'll map to your process.
Asked the way your audience asks
Diagnose
one root cause
Correct
from your facts
Prove
a crawler fetched it
The problem
When the answer is wrong and the window is closing.
An applicant no longer reads the notification end to end — they ask an AI assistant and act on the one answer it returns. A recruitment eligibility rule, an application cut-off, a document list — AI answers these constantly, and a wrong answer costs an applicant a career or a citizen a right. Some facts are restricted and must never enter a draft. And after a triggering event, response speed itself is the deliverable.
In the wild
Where a wrong answer turns into a real consequence
An eligibility rule wrong at intake
An applicant asks whether they clear the bar for an open recruitment, and the AI assistant quotes a cut-off from a superseded notification. They self-select out of a process they were entitled to enter — and neither they nor the desk ever learns a decision rode on a stale fact.
A document list with a deadline attached
Days before a window closes, an AI assistant lists a form that's no longer required, or omits one that is. Applicants arrive at the counter incomplete, against a clock that doesn't move for anyone.
A surge of wrong answers after an event
After an announcement or policy change, the same incorrect claim is returned across several AI assistants within hours, each reading the same outdated source. Response speed, not deliberation, becomes the deliverable — and the workspace reflects that.
Releasable and restricted facts, mixed
A candidate asks about a service entry scheme and the AI assistant blends a public criterion with a detail not for open release. We prepare the releasable answer from the verified record while restricted fields stay locked and visible only as a lock — so an officer can see plainly that something was withheld.
A clarification from a superseded notice
Someone asks about a procedure or the terms in a published notice, and the model answers from a version a later corrigendum replaced — the old page left live beside the new one. We identify the superseded source, correct it against the current record, and confirm the AI assistant reaches the version in force.
A misstatement in the citizen's language
A high-volume question is answered wrongly in a regional language while the authoritative wording exists only in English — so the citizens most reliant on the AI assistant get the worst answer. We flag the language gap, prepare the wording natively for sign-off, and re-ask in that language to confirm.
Impersonation surging after an event
After an announcement, several AI assistants answer from an impersonating site spreading faster than the official page. The incident clock starts; we name each source, prepare the correction with restricted facts locked out, and file a report-request — elapsed time visible until the answer is fixed and confirmed fetched.
How it works for you
Find the one cause. Fix it from your facts. Prove it landed.
- 1
Measure
Capture the high-volume public questions verbatim, with sources, model version and timestamp — evidence first.
- 2
Diagnose
One root cause per wrong answer, ranked by consequence and irreversibility — a deadline outranks an opinion.
- 3
Correct
Publish-ready corrections from your verified record, with restricted facts locked out of drafts entirely and the lock shown, never silently omitted.
- 4
Prove
Confirm the fetch, re-measure, and in incident mode watch the clock on every screen until the answer is right.
The substance
How the loop holds up under public volume and a clock
Capturing high-volume public questions as evidence
The questions citizens ask most cluster around a few consequential topics — eligibility, procedures, documents, dates, entitlements. Each is put to the major AI assistants on a schedule and captured verbatim with its sources, as a dated record rather than a screenshot — so a wrong answer is caught as a change in the evidence, not an anecdote someone happened to notice.
Ranking by consequence and irreversibility, not by volume
Not every wrong answer costs the same, so the workspace orders them by what happens to a citizen who acts on it — and whether that can be undone. A wrong eligibility cut-off before a deadline outranks a dated tone or a loose summary, because one closes a door permanently. The smallest team works the few corrections that protect a right or a career first, not the longest list.
Releasability controls that lock restricted facts out of drafts
Some facts in the record are not for release, and the correction process makes it impossible for them to enter a published draft. Restricted fields are excluded from every generated correction by construction — not by a reviewer remembering — and shown plainly as a lock, so the officer signing off sees what was withheld. A correction is prepared entirely from the releasable record while restricted material stays sealed.
The incident clock that reorders the work after a triggering event
When an event makes response speed the deliverable, the incident clock starts and the workspace reorders around the affected questions. The touched topics move to the front, corrections from the verified record are prepared for sign-off first, and elapsed time since the trigger is shown — so the response is measured by how fast the correct fact is published and confirmed fetched, not estimated.
An audit-grade immutable trail, with a sovereign and air-gapped option
Every measurement, ranking decision, released correction and proof of fetch is retained in an immutable trail — any published answer can be reconstructed with who approved it and when. For environments that can't send data to a shared service, the identical Measure–Diagnose–Correct–Prove loop runs inside your boundary, models serving locally and the record never leaving the perimeter, satisfying data-sovereignty and classification rules that rule out a hosted tool.
Diagnosing the cause by reading the record as a machine would
A wrong answer isn't met with a list of possibilities — the workspace computes the cause. It fetches your official page as a plain crawler does, no scripts, and checks whether the current fact is present: unreadable and needs restructuring; missing in the language the citizen asked; a non-official or superseded page reached instead; or the record is right and one engine is just behind. Computed from evidence, so the correction addresses the real reason and doesn't recur.
Corrections that carry an effective date, so the current rule is unambiguous
For a public body, the difference between the rule in force and last quarter's rule is the whole point — so every correction publishes with its value and exact effective date, as structured data alongside a plain-text fact block and an llms.txt entry. An AI assistant can tell which version is current and cite it, and a superseded page left live no longer competes. The officer approves wording drawn only from the verified record — a dated, extractable fact, not a notice locked in a scanned attachment.
Capabilities
Built for how you work
Releasability controls
Facts marked restricted are locked out of drafts. The interface shows the lock rather than hiding it — an officer can see that something was withheld.
Reputation Radar — impersonation & deepfake watch
Continuous monitoring across news, social and video for impersonation, forged notices and deepfakes in the organisation's name — classified and cross-checked against the verified record, with an immediate alert on anything critical so it's caught before it spreads. Sovereign-deployable, inside your boundary.
Incident clock
After a triggering event the workspace reorders around the incident and elapsed time is visible everywhere. Response speed is the deliverable.
Audit-grade trail
Every state transition appended immutably, with the acting officer, the diff and the timestamp — built to survive scrutiny.
Sovereign option
In-region managed cloud, or an air-gapped build with local model serving where the mandate requires it.
Machine-readable, dated corrections
Every correction publishes as structured data, a plain-text HTML fact block and an llms.txt entry, each carrying the effective-from date — so an AI assistant can extract the current rule unambiguously instead of reading a scanned notification it cannot parse.
Native-language accuracy
Answers are measured and corrected in every language the organisation serves, using the questions citizens actually type. A rule readable only in English is diagnosed as a language gap and fixed natively, never machine-translated.
Misinformation & impersonation defence
When a false claim or a page impersonating the organisation is the source an AI assistant answers from — often surging after a triggering event — the workspace names the exact source, prepares the authoritative correction from the verified record with restricted facts locked out, and generates a report-request for the impersonating page. Runs under the incident clock when speed is the deliverable, and never as a ranking or reputation contest.
No payment form, ever. Government accounts buy through empanelment, tender and purchase order, billed against a compliant invoice with contract-end warnings. The workspace measures only whether a citizen's answer is accurate against the verified record — it never ranks the organisation against anyone and carries no competitor or share-of-voice framing of any kind.
Data sovereignty
Your data. Your infrastructure. Our engine runs the service.
For a public institution the record cannot leave your control. The measurement, the verified record, the captured answers and the immutable audit trail can live entirely inside your boundary — we run the diagnose, correct and prove engine against your data where it sits, and you own every byte of it.
Managed, in-region
Operated by us in a sovereign region inside your jurisdiction, so data never leaves the country while your team gets the full service.
Private tenant in your cloud
Deployed into your own cloud account or VPC on your storage and your keys — we run the engine, the data plane is entirely yours.
On-premise / air-gapped
The identical loop inside your perimeter with local model serving and no egress, for classified or offline environments that rule out any hosted service.
You own all data — the verified record, every captured answer, the evidence, and the audit trail — with full export, hard delete, contract-scoped retention, and a signed data-ownership and processing agreement. We are the engine; the data is yours.
Inside the team
Who works in it
Public-affairs and PRO desk
Watches how the major AI assistants answer the questions citizens act on, and sees a wrong or dated answer as evidence before it spreads rather than after a complaint arrives.
Recruitment or programme desk
Confirms that eligibility rules, cut-offs, document lists, and dates are being returned correctly, and prepares a correction from the verified notification when they are not.
Security and releasability officer
Sets which facts are restricted, sees the lock enforced on every draft, and signs off corrections knowing sealed material could not have entered the text.
IT and CISO for the sovereign deployment
Runs the loop inside the organisation's boundary with local model serving, and holds the immutable audit trail for classification, data-sovereignty, and review requirements.
Procurement and contracts officer
Acquires the workspace through empanelment, tender, or purchase order against the existing route, with the scope, deliverables, data terms, and conformance statements the file requires and no payment form anywhere in the process.
Spokesperson and control-room duty officer
During an incident, works to the same visible clock as everyone else — sees which public questions are being answered wrongly, and releases a correction from the verified record so the accurate fact is published and confirmed fetched while the event is still live.
The takeaway
What the organisation gets
FAQ
Questions, answered
How is this procured? There is no payment form.
There is no self-serve checkout and no card is ever entered. The workspace is acquired through empanelment, tender, or a purchase order against the organisation's existing procurement route, and it can be scoped and priced to fit that process.
Can it run inside our own boundary with no data leaving?
Yes. A sovereign, air-gapped deployment runs the identical Measure, Diagnose, Correct, Prove loop inside your perimeter with the models served locally, and no measurement or record leaves the boundary. This is intended for environments with data-sovereignty or classification requirements that rule out a hosted service.
How do you guarantee restricted facts never get published?
Fields marked restricted are excluded from every generated correction by construction, not by a reviewer catching them, so sealed material cannot enter a draft. The workspace also shows the lock plainly, so the officer signing off can see exactly what was withheld before anything is released.
What happens after a triggering event, when speed matters?
The incident clock starts and the workspace reorders around the affected questions, moving them to the front and preparing corrections from the verified record for immediate sign-off. Elapsed time since the trigger is shown throughout, so the response is measured by how quickly the correct fact is published and confirmed fetched.
Do you change or influence the AI models themselves?
No. The models are left alone; what gets fixed is the record they read. Corrections are published from your verified source and the loop then proves a crawler fetched the updated fact, so the answer improves because the underlying record is now correct and retrievable.
Is this a competitive or reputation-ranking tool?
No. There is no competitor, no ranking, and no share-of-voice in the workspace, and there is no way to add one. It measures a single thing: whether an AI assistant states your organisation's own facts accurately against your verified record, and whether that accuracy can be proven with captured evidence. The entire vocabulary is accuracy, evidence, and releasability, not marketing.
How does the air-gapped deployment differ from the hosted service?
It runs the identical Measure, Diagnose, Correct, Prove loop, but entirely inside your perimeter: the language models are served locally, the verified record and every captured answer stay on your infrastructure, and nothing egresses the boundary. It is intended for classified or offline environments where a hosted service is ruled out, and it keeps the same immutable audit trail and releasability controls as the managed option.
Can we restrict who can prepare versus who can approve a correction?
Yes. The workspace separates preparation from approval by design — a desk officer drafts a correction from the record and a division head releases it — and every state transition is appended immutably with the acting officer, the change, and the timestamp. Restricted facts are locked out of drafts by construction, and the lock is shown so the approving officer can see exactly what was withheld before signing off.
Public, high-volume, and where a wrong answer has a clock.
Empanelment, tender or purchase order — we'll map to your process.
Not sure which fits? See every kind of reputation we work on.