Measure code output beyond lines changed

What does an expert hour mean?

An expert hour is an estimate of how much work an expert engineer would need to complete a pull request. On Weave’s scale, a score of 3.0 means an estimated three hours of expert effort. The unit describes the change, not how long its author spent at a keyboard.

Silk 1 analyzes the diff and provides plain-language reasoning. Its assessment includes the intent of the change, complexity, risk, and dependencies across files. Read that reasoning when a short diff seems unusually difficult or a large change contains repetitive work.

An output score does not cover every part of engineering. Planning, mentoring, incident response, and work outside the connected repositories need additional context. Use the score as one measure of code output.

Why do PR counts and lines changed need context?

A documentation correction, a generated-file update, and a difficult bug fix can each produce one merged PR. Lines changed add a volume measure, but a small change can alter behavior across a large system. Neither count describes the effort or importance of the work by itself.

Keep those measures visible for different questions:

The Engineering Intelligence product brings output together with review flow and quality signals. This page explains the output unit; the product page describes the broader system.

How should you inspect a score?

Start with a representative merged PR from a connected repository. Read the diff and the model’s explanation together. Check whether the reasoning reflects the change’s intent, its dependencies, and the work that required judgment.

If the estimate looks inconsistent, inspect the underlying change before treating it as an individual performance signal. Silk 1 supports score overrides and organization-aware calibration. An override supplies feedback; it does not turn a model estimate into a time record.

For a team comparison, hold the repository scope, time period, and inclusion rules consistent. Record changes in staffing, project type, generated code, and ignored PRs. A comparison becomes hard to interpret when the connected work changes between periods.

How does output connect to AI spending?

The AI ROI methodology compares attributed AI spending with merged output for the same measured group and period. Cost per output hour = attributed AI spend ÷ estimated expert hours of output.

The public Weave Index metric uses a matched cohort with both attributed spend and merged output in the same month. Follow that metric’s exclusions and cohort definition when citing it. Dividing unrelated company-wide totals produces a different calculation.

Pair the cost ratio with reverts, code turnover, review activity, and delivery measures. A lower ratio can coincide with more rework. Output per dollar alone does not show that AI caused an improvement.

Can output scores be converted into payroll savings?

An expert-hour score is not an hour saved, a billable hour, or a monetary return. Multiplying output by an hourly salary does not establish avoided cost. A financial ROI calculation needs a baseline, evidence of incremental benefit, and a stated valuation method.

Use a walkthrough to inspect a real PR, its reasoning, the connected repository coverage, and the quality evidence available for your team. That establishes whether the measurement supports the decision you need to make.