Compare with each part alone

Claim: a claim that a person with a model does better sets the combination against the person alone and the model alone.

Grade: measured for the finding that a combination does worse, on average, than the better part alone (106 experiments, secondary); convention for the rule that every claim carries the three arms.
Evidence: Vaccaro 2024, in against and state of the art (abstract read); Dell'Acqua 2023, in against (abstract read).
Check: three-arms (heuristic: a table that names one arm names all three).
Reader styles: none.

What it asks

A table or a figure that reports a gain has three arms: the person alone, the model alone, and both.
The task is named as a decision or as content creation, because the result differs by type.
A source that has only the combination is an upper bound on one side, not a test of the claim.

Across 106 experiments, the combination did worse than the best of either alone, Hedges' g of -0.23 (secondary); it lost on decision tasks and gained on content creation (Vaccaro 2024).
The combination gained when the person was the better alone, and lost when the model was (Vaccaro 2024).
Outside the model's reach, workers using it were 19 percentage points less likely to be correct (Dell'Acqua 2023, secondary).
Single-arm gains exist, measured on writing and coding tasks (Noy 2023; Peng 2023). They are upper bounds, since they do not set the combination against the better part.

The authors of the meta-analysis name possible publication bias and varied designs.
The effect is an average; a given task can lie on either side.
The rule that all three arms appear is a convention. The measured part is the average loss, not the rule.
The check reads tables only. It cannot tell a claim of gain from a glossary, so it warns.
The family has no case yet measured in three arms; the cases member holds that work.