Assistance must behave predictably rather than randomly

What it is

Predictable pointing assistance gives stable, explainable candidates and outcomes for the same input, objects, and state. Assistance may change direct mapping but cannot appear random: people must be able to learn when it snaps, what it will select, and how to escape it.

Why it happens

People calibrate input-result relations through repeated action. If candidate ranking depends on invisible history, network latency, frame rate, algorithmic noise, or unstable object order, equal paths yield different outputs and calibration cannot form. Deterministic precedence, stable thresholds, clear state, and continuous feedback make assistance a learnable control rule rather than intermittent intervention.

Studying it

Repeat identical paths, boundary conditions, and candidate layouts, measuring output consistency, candidate switches, post-learning error, correction, and trust. Reproduce across frame rates, zoom, devices, and loading states; stable behaviour in one ideal static condition does not prove real-world predictability.

Where it stops holding

Personalised or probabilistic prediction can change with context, but change needs stable grounds people can understand or override. Fully fixed output is not always best: visible focus and permission change should update candidates. The requirement is not never changing, but changing observably, explainably, and with task state.

Applying it

• Define deterministic candidate precedence and stable entry and exit thresholds.
• Update visible feedback immediately on state change rather than silently changing hit results in the background.
• Add repeatable tests for identical input across devices, zoom, and loading timing to detect nondeterministic output.

Related

• Same group: C1.14.1 Assistance changes actual hit boundaries and miscalibrates the user's position model · C1.14.2 Assistance causes unintended attraction and wrong drops during drag
• Nearby: C1.12 Target snapping and sticky targets · I1 State, time, and response
• Search terms: predictable assistance · deterministic input · interaction consistency