How many flotation tests do you really need?

Count the factors before you count the tests

Test counts grow quickly with factors. Four factors at three levels each is 81 combinations for a full grid, before any repeats. That is why few programmes run a full grid, and why the real question is which tests to skip.

Screening one factor at a time keeps the count low and easy to follow, but it can miss interactions, such as a collector dose that only works at a particular depressant dose. Designed experiments handle interactions with far fewer runs than a grid. Model-guided selection goes further and chooses each test using the results so far.

Repeats tell you what a difference means

Without repeats you can't tell a real improvement from test noise. A small set of duplicates at a central recipe gives an estimate of your repeatability. If two recipes differ by less than that, the next test should not be chosen on that difference.

Decide in advance when to stop

Programmes often end when the budget does. A stopping rule written down at the start is better. For example:

• A recipe meets the grade and impurity limits on repeat tests
• The next few proposed tests are not expected to beat the current best by more than your test noise
• Further testing wouldn't change the decision the work supports

Your old tests may already answer part of this

Historical results on similar ore can narrow the starting range and show which reagents never helped. They can also fit a surrogate model that proposes the next test, if the records are consistent enough and the model beats a simple baseline on later campaigns. Our demo shows this on a synthetic archive and runs one-factor-at-a-time screening alongside it. It is not a measured saving, and only a test on your own archive can show whether it holds for your ore.