Proof

Disconnected DevTools overhead

Baseline and candidate measurements for the diagnostic bridge while no DevTools listener is connected.

v0.1.0-devAdvanced

Disconnected DevTools overhead

One focused experiment compared identical core copies before and after the automatic diagnostic bridge, with no extension listener connected. Each value is median nanoseconds per Command; ten full-profile samples ran on the same Apple M1 system.

ModeScenarioBaselineBridge candidateDifference
JITone Command and one State2,760.172,756.35-0.14%
JIT10,000 sequential updates2,745.392,765.11+0.72%
JIT100,000 sequential updates2,770.922,765.35-0.20%
AOTone Command and one State2,785.792,765.90-0.71%
AOT10,000 sequential updates2,792.922,767.87-0.90%
AOT100,000 sequential updates2,777.512,781.45+0.14%

The largest measured regression was 0.72%, below the experiment's 3% acceptance threshold. Changes point in both directions and do not show a persistent disconnected slowdown.

This result does not say connected DevTools is free. Active event encoding, serialization, transport, and UI work add cost and were outside this experiment. Product VM and web builds compile the bridge out.

ARKTELOS

Purpose-built engineering systems for software that has to hold.