Monte Carlo Risk Calculator
Simulate independent three-point activity durations or cost items with a reproducible integer seed. The browser sums sequential schedule rows or independent cost rows and reports empirical P50, P80 and P90 alongside a trial-count histogram and cumulative probability curve.
By Dr. Hassan Khames EliwaLast updated
Result
Not run
How it works
Define the experiment before drawing samples
Each comma-separated row contains optimistic, most likely and pessimistic estimates in a single common unit. Schedule rows mean a fully sequential chain: every sampled duration is added. Cost rows are additive independent items rather than alternative choices. Neither mode finds a critical path, applies calendars or models resource contention. A cost entered in thousands cannot safely share a list with an amount entered in whole currency units.
Choose the distribution deliberately
Triangular sampling uses a bounded piecewise-linear density whose mode is the middle input. Beta-PERT uses lambda four, with alpha equal to 1 plus four times the relative distance from optimistic to most likely, and beta defined from the remaining distance. Its theoretical mean is (optimistic + 4 × most likely + pessimistic) / 6. This is a distribution choice, not a normal approximation built from the traditional PERT standard deviation.
Understand the numerical approximation
Triangular samples use the inverse distribution directly. Beta-PERT builds 1,025 inverse-beta cumulative-probability knots per nonconstant row using bounded bisection and a bounded continued fraction, then linearly interpolates between knots. Endpoint neighborhoods are therefore approximations, particularly in thin tails. This tradeoff keeps browser work bounded for up to 40 rows and 20,000 trials. Use an independently validated specialist package before making material decisions that depend on extreme percentiles.
Read percentiles as statements about this model
Sorted simulated totals define empirical quantiles by linear interpolation at index (trial count minus one) times probability. P80 means approximately 80 percent of simulated totals are no greater than that figure. It is not an 80 percent guarantee of delivery. The histogram counts trials in twenty equal-width bins across observed outcomes, while the cumulative chart samples 101 empirical quantiles. Repeated totals can create vertical portions of the cumulative curve.
Preserve a reproducible run record
The seed initializes Mulberry32 and never uses Math.random. Identical rows, row order, distribution, trial count and seed reproduce identical output. Run the simulation after checking your inputs. Editing an input leaves the previous result marked as out of date until you run it again; no server simulation request is made. Constant rows produce constant samples. Shared weather, escalation and productivity shocks are excluded because rows are independent; adding trials cannot repair that structural limitation or an implausible estimate range.
Worked example
Two deterministic benchmarks
Two constant rows 4,4,4 and 6,6,6 with 100 trials and seed 42 return P50 = P80 = P90 = 10.000 days in both distributions. All 100 observations occupy the first histogram bin.
A single triangular row 0,0,1 with 10,000 trials and seed 42 is checked against its analytic quantiles: P50 approximately 0.292893, P80 approximately 0.552786 and P90 approximately 0.683772, each within 0.02. A symmetric beta-PERT benchmark 0,0.5,1 has mean 0.5 by symmetry; separate inverse-beta tests use alpha = beta = 1 to verify exact uniform quantiles.
Expert tips
Common mistakes
- Adding parallel activity durations as though they were sequential.
- Mistaking independent rows for correlated project-wide risk.
- Claiming the displayed P90 is a contractual guarantee.
Frequently asked questions
How many trials are accepted?
Only integer trial counts from 100 to 20,000 are accepted. More trials reduce sampling noise but do not validate independence, bounds or the chosen distribution.
Are costs correlated?
No. Every nonconstant row receives separate pseudorandom draws. Common inflation, supplier failure or shared productivity risks require a different model with explicit dependence.
Can activities run in parallel?
Not here. Schedule mode adds every row sequentially. A network with parallel branches needs critical-path evaluation inside every simulation trial, not a sum of all activities.
Why does changing the seed move P80?
Different seeds select different finite samples from the same assumed model. Small percentile movements are ordinary sampling variation, especially for short runs.
Are estimates outside the range possible?
Individual draws stay inside their entered bounds, so their sum stays within summed bounds. Real outcomes can still exceed those assumptions if the bounds omit credible events.
Is PERT exact in the tails?
No. Beta-PERT uses a documented interpolation table. Report central percentiles with suitable rounding and independently check tail-sensitive applications rather than treating displayed decimals as certainty.
Related calculators
Related guides
Construction & ProjectMonte Carlo schedule risk in plain English: what P50 and P80 actually meanRead schedule percentiles without confusing a serial simulation, a network forecast and a commitment.15 November 2026 · 8 min read
Construction & ProjectThree-point estimating: when PERT beats a single guessA practical guide to three-point duration estimates, weighted averages and the limits of turning uncertain activities into a promised date.14 November 2026 · 7 min readSources and standards
About the author
Dr. Hassan Khames Eliwa holds a PhD in Construction & Built Environment and has more than 20 years of international project controls experience. He is Director of Eliwa Project Controls Limited in Auckland and founder of PMMilestone.org and PMMilestone.com.
Buy me a coffee