MavsterDocs
AI Features

Cost-saving playbook

Concrete settings that drop the credit cost of a test without sacrificing reliability. Memory on, the cheapest segmentation mode that still works, and a checklist for auditing existing tests.

If your credit usage is climbing faster than your test count, you're probably leaving free wins on the table. This page is a checklist of concrete settings — every recommendation traces back to a specific toggle or mode documented elsewhere in the manual.

The three high-leverage levers

1. Leave Position Memory on for stable taps

Default is on; if you've turned it off, ask why. The toggle's own description spells out the right rule: "Only enable for elements with stable, fixed positions (e.g., navigation buttons)." Most taps in a typical test qualify.

See Position Memory.

2. Leave Assert Memory and List Bounds Memory on

Both default on. The downside of leaving them on for a non-stable target is essentially nil — a memory miss costs the same as having had memory off.

See Assert Memory & List Bounds Memory.

3. Pick the cheapest segmentation mode that still works

The AI Segmentation picker only appears in Tap and Scroll editors. Three modes, from cheapest to most expensive:

  • None — on-device detection only.
  • Auto — reuses a cached result when the screen still matches, refreshes when it doesn't.
  • Force Fresh — full resolution on every run.

Try None first. If on-device detection finds your target, you spend nothing on segmentation for that step. If None misses, switch to Auto. Reserve Force Fresh for debugging — it's the most expensive mode.

See AI Segmentation modes.

Audit checklist for existing tests

Run through any test that's costing more than you'd expect:

  • Are all stable Tap steps' Position Memory toggles on?

  • Are all Visual Assert steps' Assert Memory toggles on (unless the screen genuinely changes between runs)?

  • Are all Scroll steps' List Bounds Memory toggles on?

  • Are any Tap or Scroll steps stuck on Force Fresh when None or Auto would work?

  • Are any Tap steps on Auto when None would work just as well?

  • Is Run Count set higher than necessary on any Run Test step?

  • Is self-healing set to Full Test when Per Step would do?

What about per-test cost analytics?

Per-test cost is available in both the Mavster macOS app and the web app's project analytics view. Use it to identify the highest-cost tests first — focusing the playbook on the top quartile of spend usually returns most of the savings.