Evidence you can check
Every finding points at the specific simulated reactions it came from — you can follow any conclusion back to what produced it. A claim that can't be traced to evidence is dropped and counted, never smoothed into the prose.
Meet PRISM · The engine behind Presaga
The engine whose answers can be checked.
PRISM is Presaga's reaction intelligence engine, behind all three products: it builds an AI-simulated audience from your real stakeholders, lets that audience react to the actual material, has a panel of AI reviewers argue about what the reactions mean, attacks the conclusion it reached, tells you plainly how much to trust the result — and writes it down so it can be scored against what really happened.
In plain terms
Ask a chatbot how people are likely to react and you get a confident answer with nothing behind it. PRISM shows its homework: for every conclusion you can see the reactions it was based on and the arguments against it — and afterwards we check whether it was right, and keep the score.
Two rooms
First it establishes what people would do. Then it tries to break its own answer.
Room one · the audience
A population built from your stakeholders.
PRISM reads what you describe — and what you point it at — and builds a simulated audience from it: the customers, the staff, the residents, the journalist on the story, the regulator. It looks up the real public bodies your situation draws in rather than inventing them, fetches and time-stamps the live pages that matter, and then lets the audience react to the actual material, and to each other.
Every exposure is recorded. PRISM can say exactly what each simulated person had seen before it spoke.
Room two · the scrutiny
A panel, a critic, and a ceiling on confidence.
A panel of AI reviewers in named seats reads the reactions — each seat there because without it a specific way of being wrong would go unnoticed. They argue on the record. A reviewer whose only job is to break the draft attacks the conclusion. And a confidence reading, computed from what the evidence supports, can be lowered by any of them and raised by none.
What survives is the report. The disagreements are still in it.
What makes it different
What PRISM does that a chatbot can't.
A general-purpose AI produces fluent advice with the same confidence whether it's right or wrong, keeps nothing on the record, and never finds out. PRISM was built for the opposite job.
A confidence reading that can only be lowered
Confidence is a reading computed from what the evidence supports, and it can only be lowered. A contested situation gets a low reading with the reasons beside it. That's the system telling you the truth, not failing.
Disagreement on the record
A panel of AI reviewers in named seats — the Devil's Advocate and the Evidence Checker in every product, plus the seats that product needs — argues over what the reactions mean. Where they split, you read both sides. A divided room is information.
Every finding says what would prove it wrong
Each conclusion names the observable, datable event that would reverse it. Those become your watch-list for the first days after you act — and the questions we ask you afterwards.
On the record, and scored
A finished report is sealed the moment it's delivered — its fingerprint is recorded, so it can't be quietly rewritten to look right later. Then we ask what actually happened. That builds a track record of where PRISM was right — one nobody can retrofit.
Grounded in the real world
PRISM reads the live pages you point it at and stamps when it read them, looks up the real bodies your situation pulls in — the regulator, the health unit, the council — and records exactly what each simulated person had seen before they reacted.
It writes the statement, then tests it
Arrive without a draft and PRISM proposes one built only from your facts — never an invented promise, discount, or apology. It tests the draft, revises on what it found, and hands you the finished statement with the evidence attached.
Yours, and in Canada
Your account, documents and reports are hosted and stored in Canada on infrastructure we control, encrypted at rest — locked even to someone who stole the files. On the standard service the AI model that processes your content runs outside Canada; if your rules require in-country processing end to end, we run PRISM for you on dedicated Canadian hardware as a private deployment.
Delivered the way you work
Read the briefing in the app, publish it to your own branded portal at your own address, share it by secure link, download the PDF — or listen to the audio briefing on the way to the meeting.
Three environments
Reactions take shape in different places.
A message doesn't meet one public. It meets a fast public feed, a slower community discussion, and a thousand private conversations — and it behaves differently in each. PRISM models all three, and every report draws its findings across them.
The fast public feed — Velocity™
Rapid public reaction: how quickly a message is picked up, amplified, and reframed, and how the story around it takes shape in the first hours.
The slower community discussion — Terrain™
Community discussion: how groups weigh in, deliberate, and organize, and how an issue matures as different stakeholders take positions over time.
Person to person — Echo™
Person-to-person influence: how a message travels — and shifts — as people relay it to one another, away from any feed.
What every report contains
The Decision Report, whichever product produced it.
Each product has its own sections for its own question. Underneath, every report is built the same way — so a reader learns to read one and can read them all.
- The CallThe single recommendation the evidence supports, first — written to be shared as-is with the people who decide.
- Findings with citationsEach finding names the specific simulated reactions it rests on. Follow any of them back.
- Where the reviewers disagreedWhen the AI reviewers read the evidence differently, both readings are printed. A contested situation should produce a contested report.
- What would change the callEvery finding states the observable, datable event that would reverse it — your early-warning list, and the follow-up we ask you later.
- Recommended actionsConcrete steps, each with an owner, a timeframe, and the evidence behind it.
- A confidence readingAn honest reading with its reasons attached. It moves down for what the evidence can't support and never up for anything.
- Your original inputReproduced exactly as submitted — so anyone reading the report can see what the scrutiny examined.
- Delivered your wayIn the app, on your own branded portal, as a PDF, by secure link, or as an audio briefing you can listen to on the way in.
On the record
We are on the record, and we keep score.
Every report is sealed the moment it's delivered — its fingerprint is recorded at delivery, so it can't be quietly rewritten later to look right. Every finding states in advance what would prove it wrong. And afterwards, PRISM asks you what actually happened.
That turns each briefing into a claim that can be checked, and each answer into a track record: where PRISM was right, where it wasn't, and by how much — built from records made before the outcome existed, which is the only kind nobody can retrofit.
What we don't claim
Simulation, not prediction.
PRISM models how audiences are likely to react. It does not foretell events, it is not a poll of real people, and it is not a substitute for your judgment. It is built to inform the call — you still make it. Where the evidence is thin, the report says so, in the number and in the words beside it. We would rather hand you a low confidence reading than a comfortable one.
Plans and prices
Live from the catalog — what the app charges today.
Every name, description and price below is read from the same catalog the app bills from, at the moment you load this page. Change a product in the back office and this page changes with it.