Skip to content

Sprint estimation ​

PMs and team leads use ph ryoma estimate to forecast the completion time and AI cost of a batch of tasks before they plan a sprint. You need to be signed in to Tara with ph tara login and have permission to view the project. Any Owner, Admin, Member, or Viewer of the organization can view its projects, so a PM with the Viewer role can see the underlying Flow and AI data in the Projects section of the Portal.

How does Ryoma estimate? ​

Ryoma takes the project's tasks that finished in the last N weeks (8 by default), groups them by level: S|M|L, and computes three things for each level. First, flow time p50 and p85, which is the time from when work starts to when it finishes. p50 is the level that half the tasks beat, and p85 is the level within which 85% of tasks finish. Second, the average and p85 AI cost, computed over tasks that have AI sessions tagged with the task id. Third, the sample count and the confidence.

When you supply a planned batch of tasks, Ryoma runs a Monte Carlo simulation on that same history to produce the total AI cost and the completion time for the whole batch.

p50 is the typical scenario and p85 is the conservative one. When you make commitments to people outside the team, rely on p85. For internal planning, p50 is fine. The wider the gap between the two numbers, the more erratic the history.

View the project's estimation basis ​

bash
ph ryoma estimate acme-shop

Without a batch of tasks, the command prints the estimation basis. For each of the S, M, and L levels, it shows the sample task count, flow time p50 and p85, average and p85 AI cost, and the confidence. Add --weeks N to change the number of weeks of history.

The estimates rest on this same history, so the Flow section of a project page (see Metrics on the Projects pages) shows you the underlying data before you plan a sprint.

Estimate a batch of tasks ​

You can supply the count for each level:

bash
ph ryoma estimate acme-shop --s 4 --m 3 --l 1

Or supply a list of issues:

bash
ph ryoma estimate acme-shop --issues GH-12,GH-15,GH-21 --weeks 12

The result includes the total AI cost and the completion time for the whole batch at p50 and p85. These numbers also appear in the Flow and AI sections of the project in the Portal.

Confidence ​

Confidence depends on the sample task count for each level.

SamplesConfidenceWhat to do
0noneThere is no basis. Don't use the number
1 to 4lowFor reference only. Increase --weeks to get more samples
5 to 9mediumUsable, but leave a safety margin
10 or morehighReliable in the statistical sense

The confidence for the whole batch is the lowest confidence among the levels you supply.

Limits to keep in mind ​

A forecast is only as good as the team's labeling habits. A task without level: doesn't enter the basis, and cost counts only tasks that have an AI session opened with the task id. History represents the future only when the team, the process, and the type of work don't change much, so if something changed a lot, reduce --weeks.

Estimates are per project and per level. Ryoma doesn't estimate per person, and you shouldn't use the results to set quotas for individuals. The cost here is the converted API price that Claude Code computes. It's a unit for quotas, not the amount you actually pay.

Read more ​