01 · Score formula
Explore efficiency and volume
Volume adds at most one fifth of the points left after efficiency.
100 × (0.50 + 0.2 × 0.50 × 0.50) = 55 Satisfactory 02 · Team usage
Useful work and monthly spend
Count sessions only when they produce useful work, such as a merged PR or a delivered non-PR outcome.
Up to 25 engineers. Rows are local to this page and are not saved.
03 · Policy & assumptions
Set team benchmarks and limits
The post does not publish the saturation curve or score thresholds. These are editable assumptions.
Methodology
Published model and local assumptions
From the post
The score is 100 × (E + λ·V·(1 − E)), with λ = 0.2. Volume can contribute no more than one fifth of the points not already earned through efficiency. Count only sessions with useful work, such as a merged PR or a delivered outcome for non-PR work. The policy uses Healthy, Satisfactory, Needs Improvement, and Unknown grades. Healthy and Satisfactory can receive automatic limit increases up to the admin-set ceiling. Requests above that ceiling and requests from other grades go to an admin. The policy does not auto-deny.
Assumptions used here
The E and V curves use 1 − 2^(−x / benchmark), so each benchmark corresponds to a 0.50 rating. Defaults are $40 per productive hour, 1.5 productive hours per workday, 65 / 50 grade thresholds, and at least 5 sessions for a grade. Hours under each cap are estimated linearly per dollar. Spend pending admin review is not counted as approved spend or productive hours. Devin scores each engineer's recent usage, and the post's examples use the most recent 200 ACUs. This calculator treats one month of inputs as that recent window. The auto-approval ceiling is read as the highest total monthly limit Devin can reach automatically, not the size of a single increase. Spend without useful work is estimated as spend × (1 − useful-work rate), which assumes every session costs about the same.