Apply three-point PERT estimates
Three numbers per task beat one, because they carry the shape of the risk. PERT weights the most likely value four times and folds the spread into a standard deviation you can sum across a chain.
Procedure
For each task collect: optimistic o (everything goes well), most likely m, pessimistic p (realistic worst case, not catastrophe).
Weighted mean: te = (o + 4m + p) / 6.
python3 -c "o,m,p=3,6,15; print('te', round((o+4*m+p)/6,2))"
te 7.0
Standard deviation for a task: sd = (p - o) / 6, and variance is sd**2.
For a chain, sum the expected times and sum the variances (not the standard deviations), then take the square root:
python3 -c "
t=[(3,6,15),(2,3,8),(1,2,4)]
te=[(o+4*m+p)/6 for o,m,p in t]
var=[((p-o)/6)**2 for o,m,p in t]
import math; print('total', round(sum(te),2), 'sigma', round(math.sqrt(sum(var)),2))"
total 11.17 sigma 2.54
A one-sigma band covers ~68%; for an 80% commitment use roughly total + 0.84*sigma.
Do not sum standard deviations — adding the sigmas directly assumes perfect correlation and badly overstates the chain's spread.
Round te to the unit you can defend; do not report 7.03 days.
Pitfalls
- Letting
p become an unbounded catastrophe so sigma explodes and every range is useless.
- Adding standard deviations across a chain instead of adding variances first.
- Using
m alone (the "most likely") as the commitment, which ignores that the pessimistic tail is where deadlines break.
- Forcing a symmetric range when the task is clearly skewed; PERT assumes a shape that may not hold.
- Reusing one task's three points for another in the same chain, which hides that the chain's risk is concentrated.
Verification
python3 -c "
o,m,p=3,6,15; print((o+4*m+p)/6, (p-o)/6)"
# 7.0 2.0 — the weighted mean and per-task sigma; combined sigma uses summed variances
Report each task's o/m/p, the weighted mean, and the chain total with its combined sigma and the confidence band.
1---2name: apply-three-point-pert-estimates3description: Use when you have optimistic, most-likely and pessimistic values for a task and need a defensible single number plus a variance. Uses the PERT weighted average and combines variances across a chain.4metadata:5 source: https://skillmd.com/skills/jiayaoqijia/apply-three-point-pert-estimates6---78# Apply three-point PERT estimates910Three numbers per task beat one, because they carry the shape of the risk. PERT weights the most likely value four times and folds the spread into a standard deviation you can sum across a chain.1112## Procedure13141. For each task collect: optimistic `o` (everything goes well), most likely `m`, pessimistic `p` (realistic worst case, not catastrophe).152. Weighted mean: `te = (o + 4m + p) / 6`.1617 python3 -c "o,m,p=3,6,15; print('te', round((o+4*m+p)/6,2))"18 te 7.019203. Standard deviation for a task: `sd = (p - o) / 6`, and variance is `sd**2`.214. For a chain, sum the expected times and sum the *variances* (not the standard deviations), then take the square root:2223 python3 -c "24 t=[(3,6,15),(2,3,8),(1,2,4)]25 te=[(o+4*m+p)/6 for o,m,p in t]26 var=[((p-o)/6)**2 for o,m,p in t]27 import math; print('total', round(sum(te),2), 'sigma', round(math.sqrt(sum(var)),2))"28 total 11.17 sigma 2.5429305. A one-sigma band covers ~68%; for an 80% commitment use roughly `total + 0.84*sigma`.316. Do not sum standard deviations — adding the sigmas directly assumes perfect correlation and badly overstates the chain's spread.327. Round `te` to the unit you can defend; do not report 7.03 days.3334## Pitfalls3536- Letting `p` become an unbounded catastrophe so `sigma` explodes and every range is useless.37- Adding standard deviations across a chain instead of adding variances first.38- Using `m` alone (the "most likely") as the commitment, which ignores that the pessimistic tail is where deadlines break.39- Forcing a symmetric range when the task is clearly skewed; PERT assumes a shape that may not hold.40- Reusing one task's three points for another in the same chain, which hides that the chain's risk is concentrated.4142## Verification4344 python3 -c "45 o,m,p=3,6,15; print((o+4*m+p)/6, (p-o)/6)"46 # 7.0 2.0 — the weighted mean and per-task sigma; combined sigma uses summed variances4748Report each task's o/m/p, the weighted mean, and the chain total with its combined sigma and the confidence band.