One scorer's verdict for one matrix cell. value: 0..100, higher is better. evidence: raw measurements behind the value, for debugging +

Geometry of a tappable control, in points (44pt is Apple's HIG minimum). x_pt/y_pt are the target CENTER; used for Fitts movement-time + reachabil

KLM/TLM operator times in seconds + Fitts constants. Literature-grounded defaults; a benchmarking harness may override after empirical calibration

An edge: a single user action moving from `src` state to `dst` state. weight relative likelihood a user in `src` takes this action (unnormali

Server state

Adapter so prepare code can resolve per-image expand levels straight from the app state without copying the whole map into this module.

Client for connecting to a running server

A tiny deterministic LCG so the sweep is reproducible without an RNG dep.

High-frequency sampler that tracks one pid's resource high-water marks.

Result of a single tool run.

Token usage from a single run.

Mined (or defaulted) user-behavior profile for the mobile ActionGraph. raw_counts raw TUI event/verb counts actually mined (audit trail).

The full user-behavior model.

Result of pricing one action: total seconds + per-operator detail.

Current state of the monitor

One point-in-time sample of a process from /proc/<pid>.

Aggregated resource utilization for one spawned process over its window.

A canonical user goal expressed as an ordered list of action ids. The engine sums per-action times to get a task-completion time in seconds, and

A node: a UI context the user can be in (e.g. 'chat', 'settings_sheet').

Minimal server-side WebSocket connection wrapper.