Forecast

A Monte Carlo cycle forecaster. A cycle is the fixed stretch of time your team plans in, such as a two-week sprint, and you set its length below. Give it ranges for what's left, how much you typically finish per cycle, and how much scope tends to grow. It runs 10,000 simulations in your browser and shows the distribution of completion dates.

Inputs save locally. Nothing leaves this page.

This is the free, in-browser version of the forecasting engine behind Intervals.

Items Remaining

Tickets, stories, units of work. Triangular distribution.

Low Likely High

Throughput / Cycle

Items the team finishes in a typical cycle.

Low High

Scope Creep

Multiplier on items as you discover new work.

Low High

1.0× well-defined, no surprises
1.2× some items will split
1.5× significant unknowns
2.0× prototype, exploration

Weeks per cycle Start date

The start date is the day the remaining work begins. Each finish date is the start date plus the run's cycles times weeks per cycle, which is the day after the run's last cycle ends.

Distribution

Each bar counts the simulated runs, out of 10,000, that finished in that many cycles. The left axis is the run count.

Through P50 cycles After P50, through P85 After P85, through P95 After P95

Completion dates

At least the stated share of simulated runs finish by each date, based on your inputs. Whole cycles can give several percentiles the same date. At P50, at least half the runs finish, so treat it as roughly a coin flip under your inputs. When you need a date to commit to, quote P85.

Percentiles

"P85 = 12 cycles" means at least 85% of simulated runs finished in 12 cycles or fewer. Higher percentiles give the same or a later completion date.

The product version

Explore more with Intervals

Intervals is the full instrument: save forecasts across cycles, track which product bets actually paid off, and see whether your dates hold up over time.

Explore Intervals↗

How the forecast works

What it does

Each run is one plausible future for the project: how big the work turns out to be, and how fast you finish it. We do 10,000 of these runs and sort the cycle counts. Each histogram bar counts the runs that finished in that cycle. The percentiles count completion cumulatively, through the indicated cycle, using the same results.

How items become a forecast

Sample the item count from a triangular distribution over (low, likely, high). The "likely" value is the peak; most samples cluster around it, but some land at the extremes.
Multiply by a uniform scope-creep factor drawn between (low, high). 1.0× = the work is exactly what you counted; 1.5× = expect about half again as much.
At the start of each run, draw two throughput samples uniformly from (low, high). That's the run's throughput pool.
Each cycle, draw one throughput value from the pool and subtract from items remaining. Repeat until items hit zero. The cycle count is the result of that run.

What the percentiles mean

P85 = 12 cycles means at least 85% of those 10,000 runs finished in 12 cycles or fewer. Several runs can finish in the same cycle, so the actual share can exceed 85%, and several percentiles can have the same date. These shares describe the simulated runs under your inputs and the model's assumptions.

How to read the completion dates

Each card gives one completion-by date:

P50 - At least 50% of simulated runs finish by this date.
P70 - At least 70% finish by this date.
P85 - At least 85% finish by this date.
P95 - At least 95% finish by this date; later outcomes remain possible.

The histogram colors group individual completion cycles by the P50, P85, and P95 thresholds. P70 has no color of its own; its cycle falls in whichever band contains it. Hover a bar to see the share of runs that finished in that cycle; each card counts every run finished by its date. The percentages on the cards describe completion by a date.

Assumptions and limits

Each run draws two throughput values from your range, and every cycle in that run uses one of those two. A run that draws two slow values stays slow to the end; a run that draws one slow and one fast value mixes them, cycle by cycle. Because each run only gets two draws, the spread between your throughput low and high matters a lot. A tight range gives a narrow distribution; a wide range fans out fast. The tool uses two draws so the later dates (P85 and P95) stay wide when you have no history to go on; drawing a fresh value every cycle would pull them in.
Runs count whole cycles. A run that finishes partway through a cycle counts that full cycle, so every date falls on the day after a cycle ends.
A run stops at 1,001 cycles, so a result of 1,001 means 1,001 cycles or more, and the later percentiles can understate a very long forecast. This can only happen when items remaining, after scope creep, exceed 1,000 times your throughput low.
Scope creep can't go below 1.0×, so the model can grow the work but never shrink it. Throughput low has to be at least 1 item per cycle, and a cycle at least one week.
Scope creep is uniform between low and high. Real projects usually have fatter tails on the high end; this model won't capture that.
Throughput is assumed stationary. No seasonality, no team changes mid-project, no cycle-over-cycle improvement or decay.