Don't trust the engine.
Check it.
Every Quantil calculation engine is rebuilt from scratch, outside Excel, and the results are compared. The method, the measured values and the reference figures ship with the tool.
Three steps, every tool.
The standard way to validate a numerical engine: compute the same thing along a second, independent route.
A second engine.
The calculation logic is read line by line and rebuilt in Python with NumPy and SciPy, sharing no code with the workbook. An error in the original would show up differently in the rebuild.
Clear pass criteria.
Components are checked against an analytical formula, a reference library or an expected statistical property; for the schedule engine, the Monte Carlo results must agree within sampling error. Then the full chain is compared on the reference case.
Repeatable by you.
Each tool ships with a validation or verification note listing the checks, measured values, seeds and iteration counts. You can download them below. Several also ship the Python scripts, so a reviewer can run the comparison again.
The results.
From the validation note of each tool, October 2026. Download a note to see the criteria, the measured values, the seeds and the iteration counts for yourself.
| Cost & Schedule Risk Analyzer | 19 of 19 tests pass. Fourteen tests on the engine (Beta-PERT sampling, percentiles, single-factor correlation) and five on the Cholesky correlation mode; the MRG32k3a random generator has its own self-test. Load test at 100 risks, a EUR 250 million base estimate and 50,000 iterations. Validation note (PDF, 15 pages) |
|---|---|
| QRCIM | All tests pass. Fourteen component and chain tests, plus an end-to-end test on the full reference register and a control test on the correlation effect (Gaussian copula via Cholesky decomposition). Validation note (PDF, 16 pages) |
| Schedule Risk Analyzer Pro | Deterministic schedule reproduces exactly: 909 days and 21 critical activities in both tool and rebuild. Monte Carlo P50 identical; P80 and P90 within 2 and 4 days, inside the sampling error. The rebuild samples durations and risks without correlation, so the correlation path is not part of this comparison. Verification script included, plus the P6 import script. Verification note (PDF, 3 pages) |
| RCAT | 18 of 18 tests pass. Generator, distributions, sampling, correlation, reproducibility and percentiles. The rebuild returns the same P80 of EUR 52,963,087 as the workbook. Python suite included. Validation note (PDF, 10 pages) |
| Bid Price Optimizer | 12 of 12 tests pass. Workbook formulas, VBA engine and a Python reference agree to floating-point precision; the recommended price is identical in an independent optimizer. One limit was found, and the input range was narrowed to stay within it. Validation note (PDF, 11 pages) |
What validation does not tell you.
We would rather say this here than have you find out later.
- It is our own validation.The tests were done by Quantil, not by an external auditor. What is independent is the route. Everything is documented so that a third party can repeat it without starting over.
- It tests the arithmetic, not your register.A correct engine on a weak register still gives a weak forecast. The quality of the result depends on your probabilities and three-point estimates.
- Simulations vary within sampling error.Two Monte Carlo runs with different generators differ slightly. With a fixed seed, the same workbook always returns the same numbers.
- Models have limits.Each validation note ends with the modeling choices and where they stop applying. Read them before you rely on a result in a decision.
Check your own installation.
Every tool page lists the figures the unchanged reference case must return. If your copy gives the same numbers, the engine runs correctly on your machine.