Revising a system of beliefs › What an unexpected result actually tells us

Return to the imaginary sensor. Suppose our prediction depends jointly on three assumptions: the sensor is calibrated, the heating is off, and no other source is warming the room. Suppose also that, given those assumptions, our model predicts a stable reading. An unexpected rise conflicts with that package. It does not deductively identify which component failed.

This conclusion has a precise logical limit. If assumptions A, B, and C together entail prediction P, and P is false, then at least one assumption in the package must be rejected. It does not follow that A specifically is false. Nor does it follow that A, B, and C are all false. The same structure can be understood without symbols: the failure of a joint guarantee does not identify the failed contributor.

Further inquiry can narrow the options. A second independently checked instrument might weaken the calibration explanation. Inspecting the heating controls might reveal an unnoticed setting. These are additional tests with their own assumptions. Nothing in this reasoning says inquiry cannot make progress or that every proposed explanation deserves equal confidence.

The example illustrates a reason to take holism seriously, but it establishes less than the whole of Quine's program. A familiar dependence on auxiliary assumptions does not by itself prove that every scientific belief participates equally in every test. The extent of the relevant system remains a substantive issue. Readers can accept the limited lesson about joint testing while questioning a more global thesis.