Chapter 3 – Reading a distribution: what P80 promises and what it doesn't
The Quantitative Schedule Risk Analysis Handbook · Edition 1, August 2026 · by the team behind [Epoch SRA](https://epochsra.com), an Excel add-in for schedule risk analysis · corrections welcome at contact@epochsra.com
A schedule risk analysis does not produce a date. It produces a distribution of dates – typically from tens of thousands of simulated program futures – and then summarizes it with percentiles. Reading those percentiles correctly is the difference between using the analysis and being decorated by it.
The definitions, precisely
P50: the median. In half of the simulated futures the program finishes by this date; in half it does not. A P50 commitment is a coin flip, knowingly taken.
P80: the 80th percentile. In 80% of simulated futures, the program is done by this date. Committing at P80 means accepting a one-in-five chance of missing – explicitly, on the record, instead of implicitly and by surprise.
P-anything is a statement about the model's futures, not a promise about the world. If the inputs are optimistic, the P80 is optimistic. Percentiles inherit the honesty of what feeds them – a theme Chapter 7 (calibration) treats fully.
The gap is the message
The distance between P50 and P80 measures how much the program does not yet know. A program with P50 in December and P80 in mid-February carries roughly ten weeks of uncertainty in that band alone. Two programs can share a P50 and differ enormously in P80 – the second number is what distinguishes a tight plan from a hopeful one, and it is the number single-date scheduling never produces.
When someone presents one date with no percentile attached, it is fair – and clarifying – to ask which percentile it is. In our experience the honest answer, once computed, is usually nearer P20 than P50: the official date is a date the program will probably miss, presented as a plan.
Distributions have shape, and the shape talks
Two features of a finish-date histogram deserve attention before any percentile is quoted.
The left edge. If a visible fraction of simulations lands exactly on the deterministic finish, that spike is the probability that nothing goes wrong – on a program with discrete risks, it approximately equals the product of the risks' non-occurrence probabilities. A tall spike is not an error; it is the model saying your uncertainty is dominated by discrete events rather than duration noise. It can be checked with one multiplication, and auditable arithmetic is the cheapest trust a model can buy.
The right tail. Its length reflects compounding – risks landing on risks, [merge bias](/guide.html#merge-bias) amplifying both. Deep-tail percentiles (P90, P95) are the least certain numbers the analysis produces, because tails are where model assumptions matter most and calibration evidence is thinnest. A tool that labels which of its percentiles rest on evidence and which are model extrapolation is telling you something most tools do not.
Choosing a commitment percentile
There is no universally correct choice; there is a correct conversation. Planning internally at P50 and committing externally at P80 is a common and defensible convention. Safety-critical or contractually severe milestones may justify P90. What is not defensible is committing at an unknown percentile – which is precisely what single-date scheduling does, every time.
--- In practice: at your next schedule review, ask for two numbers instead of one – the P50 and the P80 – and put the gap between them on the slide. The size of that gap, and whether it shrinks review over review, tells you more about program health than either date alone.