Skip to contentComing soon: SprintProof for ChatGPT and Codex

Feature · Estimate

Not one number. A range and the dates behind it.

SprintProof runs the plan 10,000 times over every part’s best, likely and worst days, and gives you the dates you can be 50, 80 and 90% sure of.

The estimate in 22 seconds. One number becomes a range from 10,000 simulated runs. The 80% and 90% dates are picked, the riskiest part is named, and the AI is asked for a second opinion. Example data; the captions are part of the picture.

In short

SprintProof estimates a feature as a range, never one number. It simulates the plan 10,000 times over every part’s best, likely and worst days and gives the dates you finish by with 50, 80 and 90% confidence. It shows how the total adds up and which parts are riskiest. The estimate is maths, not AI, and uses no credits.

Facts you can check

What you get

Estimates with error bars.

  • 50, 80 and 90% dates

    Say “done by this date, with 80% confidence”, and show the 50% and 90% dates next to it. 50% for planning, 80% for a commitment, 90% for a launch date.

  • How the total adds up

    Planned work, work found in the gap review, overhead and margin, each shown on its own line.

  • The riskiest parts

    The parts with the widest ranges, which move the late dates most, so you know what to split or spike first.

  • Every outcome, charted

    A chart of the simulated outcomes and a range bar for each part.

  • Blind estimates

    Everyone votes privately, then the votes are revealed together with the spread, the median and the outlier. No credits.

  • A challenge on request

    For 10 cr, the AI reviews the uncertainty and proposes fixes: add a spike, ask someone to decide, split a part.

How it works

Three numbers per part. Dates you can commit to.

Every AI step shows its credit cost before you click, and nothing in the plan changes until someone accepts it.

  1. 01

    Give each part three estimates.

    Every part in the breakdown has best, likely and worst days. The planner drafts them with the plan; your team corrects them.

  2. 02

    Set the team and the start.

    Enter the number of engineers and the start date. Dates count working days from there.

  3. 03

    Pick how sure you need to be.

    The estimate shows the range and the 50, 80 and 90% dates. Switch the confidence level to see the date you would commit to at each one.

  4. 04

    Find the risk.

    10 cr

    The riskiest parts are listed by how wide their range is. Split them, add a spike, or Challenge the estimate to get the AI’s proposals.

How the dates are worked out

Each part’s best, likely and worst days describe a spread of outcomes. The estimate draws a result for every part, adds them up, adds overhead (15% by default), and repeats that 10,000 times. Parts tend to overrun together, so each simulated run shares part of its overrun across the parts. The 50% date is the one half the runs finish by; the 90% date is the one nine in ten finish by.

Dates assume engineers work in parallel about 70% of the time, so a second engineer adds 0.7 of a person, not a whole one. No run finishes sooner than its longest part, and the work is counted forward in working days from the start date. The dates don’t model which part waits on which.

Estimate together, without anchoring

Start an estimate round on up to 20 parts with up to 12 people. Votes stay hidden until everyone has voted, then they are revealed at once, with the spread, the median and the outlier marked. The team talks about the outlier, not about whoever spoke first. Rounds never use credits.

Challenge the estimate

Ask the AI to challenge the estimate (10 cr). It says whether the plan looks realistic, optimistic or pessimistic, names the risks, and proposes actions: a spike, a decision someone has to make, a part to split, or missing parts and questions. You apply the findings you agree with, one click each.

The numbers come from the estimate engine, not from the AI.

What the estimate doesn’t claim

The estimate shows how uncertain the plan is; it doesn’t promise accuracy. It does not yet learn from how long past tasks really took. Time logged on tasks is recorded, but later estimates don’t use it.

FAQ

Estimates: common questions

Why does SprintProof show a range instead of one date?
Because every part of a plan is uncertain. A single date hides how sure the team is. The range and the 50, 80 and 90% dates show how likely each finish date is, so you can choose the confidence you commit to.
How are the 50, 80 and 90% dates calculated?
The plan is simulated 10,000 times from every part’s best, likely and worst days, plus overhead. Parts tend to overrun together, so each simulated run shares part of its overrun across the parts. Dates assume engineers work in parallel about 70% of the time, and are counted in working days from the start date. The 80% date is the one 8 in 10 runs finish by.
Does the AI decide the estimate?
No. The numbers come from the estimate engine and the days your team entered. The AI drafts first estimates for each part and, when asked, challenges the estimate with proposals you can accept or ignore.
Does the estimate learn from past features?
Not yet. Time logged on tasks is recorded, but it does not change later estimates.
What does the estimate cost?
The estimate, its dates and blind estimate rounds never use credits. Challenging the estimate with the AI costs 10 credits. Credits come with the Team plan; see the pricing page.

Team

Plan the next feature before the sprint starts.

Create your account and name your workspace. Your first feature plan is free. To keep planning, subscribe to Team: the 14-day money-back guarantee applies.

Want to hear when the ChatGPT and Codex plugin ships? Leave your email.

Team size (optional)
One email when it ships. No spam.