A number that survives scrutiny

An estimate adjusted for measurement error, delivered in 48 hours — with a certified interval, a certificate a third party can verify, and an honest "what is not verified" section. For pharma, CRO, medtech and analytics teams.

Send two tables Open a sample report
Sample reports use published aggregates and synthetic data. No personal data required.
€950 per report 48 hours delivery 2,160 / 2,160 true value inside the bounds 0 wrong guaranteed decisions 62 machine checks per report

First pilot free. If no certificate can be built from the data, 100 % refund.

In short

  • What it is. A report with an estimate of a proportion or prevalence adjusted for measurement error, plus a proven interval for it.
  • Price. €950 for a standard report, €600 white-label for consultancies. The first three pilots are free.
  • Delivery. 48 hours after the data arrives; one revision within 30 days.
  • How it is checked. A machine-checkable certificate and an independent module that does not import the solver.
  • When we refuse. If no certificate can be built from the data, the answer is "not determined" with a 100 % refund.
  • Limits. Up to 4 variables and 16 joint cells; the target must be an unconditional marginal.

When you need it

Self-report and validation

A proportion is known from self-report, and a validation study says how often self-report is wrong. You need the true proportion.

Imperfect test

The share of positive results is known, plus sensitivity and specificity from another study. You need prevalence.

Several sources

Two or three partial tables describe one population. You need to know what follows from them and whether they contradict each other.

Rounded publication

Numbers are published with two decimals. You need to know whether the conclusion survives that precision or rests on rounding.

What it replaces

2–4 weeks of a consultant

48 hours, a fixed price and a report you can forward to an auditor, a regulator or a board.

"Approximately ±0.5"

A certified interval whose width is set by your data, not by an analyst's confidence. You can see where it comes from.

A formula without bounds

A certificate and an appendix: a third party rechecks the calculation and does not depend on our word.

What you get

  • an adjusted estimate with a proven interval;
  • a recalculation of the conclusion at 2, 3 and 4 decimal places;
  • the contribution of each source to the bounds;
  • a list of assumptions, including transport between populations;
  • a 6–8 page report plus a print version;
  • a reproducible specification and a certificate for third-party verification.

What it changes

You get more than "a number from an analyst": a conclusion with an explicit line between what is proven and what is assumed. If the data is insufficient, the report says so before a decision is made.

Example

Published aggregates: a self-reported share and a validated measurement mechanism.

With exact numbers:      P(Y=1) = 262691/327750 = 0.801498
Rounded to 2 decimals:   P(Y=1) in [0.7859, 0.8171]
Threshold 0.8:           decided with exact numbers, not decided with two decimals.

Open the full sample report — with assumptions, a rounding table and a "what is not verified" section. The report is in Russian; the certificate is language-neutral.

Numbers come from a published example; carrying the measurement mechanism over is a declared assumption, not a result of the calculation.

Price

Standard report

€950

One question, up to three sources, one revision within 30 days. Delivered within 48 hours after the data arrives. For Russian clients: 85 000 ₽. The euro price is approximate at the CBR rate of 89.8597 ₽/€ on 20 Nov 2026.

White-label for consultancies

€600

The same report under your brand, four or more per month. For Russian clients: 54 000 ₽. You deliver it to your client; we do not appear in the correspondence.

The first three pilots are free in exchange for the right to an anonymous case. Payment by invoice, in advance. If no certificate can be built from your data (invalid input), 100 % refund. If the interval is wide, that is the answer, and it is known before the order.

How we work

StepWhat happensTime
1. EmailYou send the tables and the question. I reply whether the question is answerable and which assumptions it needs.1 business day
2. Specification and invoiceWe fix the question, the data, the threshold and the price. Advance payment.1 day
3. Calculation and reportExact calculation, independent verification, report and appendix.48 hours
4. RevisionOne correction within 30 days, for example when a number is refined.on request

What we do not do

FAQ

We have a statistician. Why you?

Then you are buying 48 hours, an independent check and an appendix for a third party, not a calculation. If your statistician is available and the task is routine, do it in-house — that is reasonable.

Isn't this a formula in R?

For one source with exact numbers — yes, and it is free. We matter when there are several sources, numbers are rounded or given as intervals, and the tables are partial: then the formula is either inapplicable or gives a point without an honest bound.

Is the certificate a regulatory document?

No. Regulation requires uncertainty analysis, not a machine-checkable certificate. The certificate is an appendix that lets your auditor recheck the calculation instead of relying on our word.

What if the interval turns out wide?

Then the report says so. We do not promise a narrow interval: it is determined by your data. Narrowness is discussed before the order, not after.

Who checks the calculation?

An independent verifier that does not import the solver and rebuilds the constraints from the tables. The report contains 62 machine checks; the certificate can be rechecked with a command from the appendix.

What data is needed?

Two or three tables: the observed share, sensitivity and specificity or another measurement mechanism, and intervals where available. Exact fractions are preferable to decimals. No personal data is required.

Which language is the report in?

The written report is currently prepared in Russian; the machine-checkable certificate is language-neutral. For international partners we translate the final report on request.

Confidentiality

Only declared tables are needed, not raw records. But honestly: tables can reveal a joint model. We will sign an NDA; we do not promise technical anonymisation.

Method and provenance

The calculation uses exact rational arithmetic: the problem is reduced to a linear program over a simplex, and bounds are confirmed by primal and dual certificates. The report, certificate and specification are reproducible; an independent module recomputes the result without access to the solver.

Last updated 20 November 2026. Sample data is synthetic or public; the sample report and certificate are linked above.

Write to us

One email: what you need to estimate, two or three tables (exact fractions or intervals), the threshold if any, and who is responsible for the meaning of the variables.

Email: udm-veres@ya.ru — replace with a working address before publishing.

Send an email