Mind · Strategy
DistinguishingTest
Find an observation that rival explanations predict differently before gathering more evidence that fits them all
Definition
Mechanism
When the evidence in hand leaves two or more explanations standing, as in EvidenceUnderdetermination, each explanation's prediction is worked out for observations not yet made, and an observation is chosen on which the predictions differ: if a failure comes from expired credentials, the request is rejected with an authentication error; if it comes from lost connectivity, the request never reaches the server. The predictions, and which outcome would favour which explanation, are recorded before the observation is made. An inconclusive result is a permitted outcome and is recorded as inconclusive, not read as favouring an explanation. Evidence that every explanation predicts equally is not sought as a test, however easy it is to obtain, and an observation is worth making only when the decision at hand depends on which explanation holds.
Invariants
- Two or more explanations are named, each with its predicted outcome for the chosen observation.
- The predicted outcomes differ, and they are recorded, with what each would favour, before the observation is made.
- An inconclusive result is recorded as inconclusive.
Risks
Failure modes
Confirming test
an observation every explanation predicts is gathered and read as support for the favoured one.
After-the-fact prediction
the predictions are stated once the result is known, so any outcome can be read as favouring the favoured explanation.
Forced verdict
an inconclusive result is read as favouring one explanation.
Distinction without consequence
a costly distinguishing observation is made although the decision at hand is the same under every explanation.
Relationships
Builds on
References
The evidence in hand fits two or more explanations about equally, so it cannot decide between them
Nothing in this library builds on this pattern yet.