ROI by team size

Measure it. Do not market it.

87ms hot reload and 359ms hot restart are measured on a 3,692-library Flutter app with three simulators. Everything else is a modeled scenario. Adjust it.

Weekly time in scope0.8–1.5 hrswaiting on restarts and boots
Token cut per verify cycle60–80%vs scraping logs
Run-button interruptsrarely askedwho can self-serve

Sheets

Four arguments you can recompute.

If a number is not measured, it is labeled as an estimate.

Hot reload vs restart

Measured: 87ms reload, 359ms restart. Estimate: 8–15s for an agent to kill, relaunch, and wait. At 50 reloads × 5 days, that is roughly 33–62 minutes a week for one Flutter developer.

Three simulators

Measured: reload-all under 100ms across three sessions. Estimate: 30–60s to boot or switch devices by hand. Ten cross-device checks × three features is 15–30 minutes a week.

Agent tokens

Baton: 2–3 tool calls and a short structured payload. Estimate without it: 4–8 shell commands, 2k–5k tokens of logs, 3–10 readiness polls. Modeled reduction 60–80% per verify cycle.

Clone-to-running

Estimate: 10–30 minutes of developer time per “help me run this,” 2–5 times a week, 3–10 non-devs. Studio-wide that can be 1–25 hours if a Run button is actually safe to hand over.

Track your own: tokens per task, wall time from edit to verified screen, developer interrupts, and time-to-diagnosis when reload fails. Share results via the feedback issue.