Choice between repleting/depleting patches: A concurrent-schedule procedure.

Pigeons pecked two keys in a chamber. One key gave grain fast at first, then slowed as the 'patch' emptied. The other key refilled while the bird worked. This cycle repeated many times.

The setup let researchers watch how animals split time and pecks between a fading reward and a growing one. They tracked both response counts and how long birds stayed on each key.

Birds followed the generalized matching law: the ratio of their pecks matched the ratio of grain deliveries. That part held steady even while patch rates changed.

But melioration theory missed the mark. It predicted birds would stay longer on the richer key than they actually did. Molar maximization also fit poorly. Only matching survived the test.

Eisler (1984) saw pigeons shift toward richer schedules when time limits were added, a result hailed as a win for melioration. The new study keeps the melioration lens but shows the same idea fails when you measure dwell time under repleting and depleting patches. The difference is the patch dynamic, not the birds.

Locurto et al. (1976) and Smith et al. (1975) already noted undermatching on plain concurrent VI schedules. Davison (1992) agrees: matching holds, but only in its looser 'generalized' form once patches cycle.

Burack et al. (2004) adds that matching itself is learned through reinforced switching. Taken together, the patch procedure does not overturn matching; it simply reveals where melioration's moment-by-moment story breaks down.

If you use concurrent schedules in skill-acquisition or preference assessments, track both responses and time spent. A child may distribute responses in line with reinforcement rates yet still leave a rich option sooner than a maximization rule predicts. Watch for those early exits; they signal the patch is 'depleting' for that learner, and a switch may be optimal even if the raw count looks good.