Module load 

The host’s load, and how far it stretches a silence budget (ADR-0009).

A machine running several worktrees and agents at once makes every tool slower, and a tool that is slow because it is waiting for a core is not stuck. The silence budget is therefore multiplied by how oversubscribed the host is — the one-minute load average over the core count — capped by amont.idleLoadScale. A false kill costs a parked commit; a slower verdict on a genuine hang costs minutes.

The arithmetic is a calculation with no clock or syscall in it; the reading is one getloadavg on Linux and macOS, and a factor of one everywhere else, which the messages say rather than claim a load they never measured.

Structs

Load

What was read from the host, in thousandths so it travels as atomics.

Functions

factor_text

×3.9, from a factor in thousandths.

read

The host right now, where it can be read.

scaled_budget

The silence budget under load: idle × clamp(load1 / cores, 1, cap), rounded to whole seconds, then no more than the ceiling when there is one. Pure, so it is tested to the second.