Four published methodologies, run over your question, returning a document you can put in front of a board — with the method disclosed before delivery and every claim traceable to its source.
Status as of 18 October 2026 The platform has not yet delivered to a production client. The Studies methodology, version 3.4, is written and will be published before the first delivery. The PRISM, Surveys and Crucible methodology documents are not yet written.
Every deliverable carries this page. The weakest figure is shown, not hidden.
The usual order is reversed. You read the method first and approve the plan second, and only then is anything produced. Each stage ends on a named exit condition, and the ones that matter are yours.
You receive the methodology in writing, named and versioned, before anything is produced.
Exit: you accept the method
For the first stretch the engine only asks questions. It uses your vocabulary, and a correction overrides anything said before it.
Exit: you confirm the scope
The chapter plan is a governance document you approve, not a search plan the engine edits.
Exit: your signature
The citation gate runs last, on the text that ships. Every claim is checked against its source, and every source is fetched again.
Exit: every citation passes both checks or is marked as judgment
The document carries its evidence summary, its full bibliography and a section naming what it could not establish.
Exit: you accept, or reject within 14 days
Each engine answers one question. Read in order, they close a loop: diagnose, listen, investigate, validate, then diagnose again with the evidence the earlier engines recorded.
| Engine | The question it answers | Price per deliverable |
|---|---|---|
| PRISMInward | How good is what we already have? | $3,400 |
| SurveysAround | What do our people actually think? | $4,900 |
| StudiesOutward | What should we do about it? | $3,900 |
| CrucibleAhead | Will this survive contact with reality? | $600 |
Prices in US dollars, per named deliverable, as of 18 October 2026
A fixed price for each named deliverable. No seats, no token counts and no overage billing. PRISM always reports two numbers: the weighted index and the conservative staged level, never one without the other.
Every claim in a finished document resolves to a source, or says plainly that it does not. The legend is printed in every deliverable, in the same place.
[ref: key]
Supported by a recorded source that passed both gate checks. The key opens the source in the bibliography.
[agency judgment]
Stated on the analyst's reasoning, not on a source. Not an error. The engine counts these marks rather than hiding them.
[INSECURE]
Reached over an invalid certificate and kept on purpose. Some national statistics offices have broken certificates, and dropping them would lose the primary source.
A failed citation is rewritten, never deleted. If a claim fails either check, the sentence is marked as judgment. A broken citation is worse than none, because it looks like evidence.
A blocked source is not a dead one. Paywalled and bot-blocked sources are real and stay in the bibliography. Only sources that are gone or unreachable are counted as dead.
These documents exist to travel outward, to a board, a donor, a ministry or an accreditor. The terms are written for that.
The document is yours. Put it in the annual report or the donor file.
The credit is refunded if the adversarial critique scores below threshold, or if you reject the deliverable within 14 days with a citation-level reason.
Arabic output is written in Arabic, with tables, scorecards and charts mirrored for right-to-left reading. It is a paid capability, not a translation pass.
Where data cannot leave the country, the platform runs in-region, sold as a fixed annual capacity a tender committee can approve. You supply the model endpoint.
You do not have to take our word for it. Every claim resolves to a source you can open, the link check runs on the version that ships, and any sentence we could not back is marked as judgment rather than dressed as evidence. We print the proportion. And if you find a citation that does not hold, the deliverable does not consume a credit.
The rejection has to come within 14 days and name the citation.