Tensions in Practice: A useful provisional inference in tension with a stable fact-only contract¶
A component class and one later inspection result
Exactly 90% of a known component class pass inspection. A particular unit belongs to that class, with no known reason initially to treat it as atypical. A defeasible inference supports “this unit probably passes.” A fact-only store does not assert that individual conclusion. Later, a reliable case result establishes that the unit fails. The class rate and membership stay true, but the earlier supported conclusion must be withdrawn.
Use qualified case-level support
Make a provisional inference before decisive case evidence arrives.
Store only entailed case facts
Avoid treating a class proportion as an individual fact.
Why these aims pull against each other
A useful default inference is not a deduction. New case evidence can defeat the conclusion without contradicting the class-rate premise, so a system that uses such conclusions must track their provisional force.
Choose an arrangement to see what changes and what remains difficult.
Supported means defeasible case-level support, not an entailed fact. Withdrawn means the provisional conclusion is no longer supported after the reliable case result. The class rate and membership remain unchanged.
What this choice protects
What it costs
When it fits
Compare the arrangements
Maintain a defeasible conclusion
Support the pass conclusion from the class rate and membership, explicitly subject to defeating information. Withdraw it after the reliable failure result.
| Before | After | |
|---|---|---|
| Class rate | 90% pass | 90% pass |
| Unit in class | Given | Given |
| Unit passes | Supported | Withdrawn |
| Unit fails | Not known | Given |
- What it protects
- The unit can have a qualified forecast before a decisive case fact is available.
- What it costs
- The conclusion needs provenance and a retraction path; consumers must not mistake support for certainty.
- When it fits
- Fits inquiry or forecasting that can represent provisional support and respond to defeaters.
Illustration note: The earlier support was not a deductive guarantee. Its withdrawal does not require deleting either true premise.
Keep an entailed-fact store
Record the rate and membership, but do not add the unit-level pass claim as a fact. Add the failure fact when supplied.
| Before | After | |
|---|---|---|
| Class rate | 90% pass | 90% pass |
| Unit in class | Given | Given |
| Unit passes | Not entailed | Refuted |
| Unit fails | Not known | Given |
- What it protects
- The store never needs to retract a pass fact it did not assert.
- What it costs
- It provides no provisional individual forecast without a separate inference layer.
- When it fits
- Fits a fact record whose contract excludes defeasible conclusions.
Illustration note: Not entailed does not mean improbable or false. The contract deliberately declines to turn statistical support into a stored individual fact.
What this illustration does—and does not—establish
The source supplies the structural tension. This bounded example makes a particular relation inspectable; the aims, conditions and residual costs are part of the comparison.
- The class proportion and later reliable result are stipulated premises, not estimates or empirical product claims.
- The unit is initially treated as an ordinary class member, with no additional relevant defeating information. A different selection mechanism could change support.
- The example concerns inference, not authorization to act. Costs, rights and decision thresholds would require separate premises.
Source entries
Statistical syllogism
The canonical tension motivates this comparison. The setting and arrangements are declared editorial illustrations, not observed findings.
Default persistence vs. revision
A default is useful only if it yields when a genuine defeater arrives.
The source operation
A statistical syllogism is the defeasible inference from a qualified generalization about a class to a claim about a particular member. In its simplest form, a proportion p of members of reference class F have attribute G; individual i is an F; therefore, to strength related to p and subject to available defeating information, i is supported as G. The conclusion is not deductively entailed unless the proportion and accompanying premises eliminate every exception.