Interlude VI-A: The Predicate That Was Refused
Aa
This interlude serves as a worked example within The Proofs, demonstrating the A29 (Predicate Acceptance) machinery in its failure mode. Rather than introducing new formal definitions, it traces a rejection and a partial remediation for a concrete predicate proposal, showing that the acceptance gate produces structured, actionable artifacts when a predicate fails. The corresponding narrative context for predicate discipline appears in Volume I, Chapter 8.
This hypothetical example follows a proposal whose treatment of a boundary case is unspecified. Its numbers and records are illustrative. The refusal concerns that missing admission condition, not a proof that the concept is contradictory.
The Proposal
A fashion catalog team proposes a new predicate: "trendy."
PROPOSAL: trendy
signature: Item → Score[0,1]
intension: "Items that reflect current fashion trends"
scope: catalog_spring_2026
positive_exemplars:
- dress_A (volume-sleeve midi)
- dress_B (butter yellow shift)
- dress_C (cutout bodycon)
negative_exemplars:
- dress_X (classic black sheath)
- dress_Y (vintage 1950s silhouette)
invariants: []
provenance: { source: "merchandising_team", method: "manual_curation" }
The proposal looks reasonable. Five exemplars, clear scope, valid signature. The proposal enters the acceptance pipeline.
Stage 1: Exemplar Validation
The system checks whether the provided exemplars are well-formed and exist in the declared scope.
VALIDATION:
✓ dress_A exists in catalog_spring_2026
✓ dress_B exists in catalog_spring_2026
✓ dress_C exists in catalog_spring_2026
✓ dress_X exists in catalog_spring_2026
✓ dress_Y exists in catalog_spring_2026
exemplar_validation: PASS
No problems yet.
Stage 2: Invariant Check
The proposal supplies no additional invariants. That empty list has nothing to check, but it does not discharge the containing context’s hard requirements. The following warning records the omission.
WARNING [INV-001]: Predicate "trendy" declares no invariants.
Consequence: No additional invariants were supplied.
The declared exemplars and the containing context still constrain admission.
Recommendation: Declare at least one invariant that constrains
the predicate's extension (e.g., ¬(trendy ∧ discontinued),
or trendy → in_current_season).
The warning is logged but does not block acceptance. A warning cannot waive a mandatory receiving-context condition. Admission remains subject to the full A29 gate.
Stage 3: Conservative Extension Check
The example next checks whether the proposed evaluator changes the existing predicates on the current records. That is a bounded operational check. Logical conservativity asks whether the theory gains any new old-language consequences and requires a separate argument.
CURRENT_RECORD_COMPATIBILITY_CHECK:
Checking interaction with: style_category
Checking interaction with: silhouette
Checking interaction with: season
Checking interaction with: price_tier
✓ No existing predicate's extension is affected.
current_record_compatibility: PASS
logical_conservativity: NOT_ESTABLISHED_BY_THIS_CHECK
The current-record check has passed; the separate logical obligation remains open.
Stage 4: Boundary Case Evaluation
The system identifies items that are "near" the exemplars in embedding space but were not classified. It presents these to the predicate for classification.
BOUNDARY_CASES:
dress_M: volume-sleeve maxi (similar to dress_A)
dress_N: pale yellow A-line (similar to dress_B)
dress_P: classic black sheath with modern cutouts (similar to both dress_C and dress_X)
The predicate's intension ("items that reflect current fashion trends") is evaluated on these boundary cases.
For dress_P, the system encounters a problem:
BOUNDARY_CASE_CONFLICT:
Item: dress_P (classic black sheath with modern cutouts)
Similarity to positive exemplar dress_C (cutout bodycon): 0.78
Similarity to negative exemplar dress_X (classic black sheath): 0.82
The intension "reflects current fashion trends" is ambiguous:
- Cutouts are a current trend → should be trendy
- Classic black sheath was marked not trendy → should not be trendy
These signals do not determine the intended label of dress_P.
The proposal supplies no rule or abstention contract for this boundary.
The opposed similarities do not make the exemplars inconsistent. They expose a question the proposal has left unanswered: should a classic silhouette with this feature fall inside the category, outside it, or in an explicit abstention region?
Stage 5: The Failure
The example next supplies classifier scores for the boundary cases. This is one part of the admission inquiry; the other required checks remain outstanding.
TEST_SUITE_EXECUTION:
Classifier trained on 5 exemplars (3 positive, 2 negative)
Test: Classify boundary cases
dress_M (volume-sleeve maxi):
Predicted: trendy (0.71)
Ground truth: UNKNOWN (not in exemplars)
Result: UNDETERMINED
dress_N (pale yellow A-line):
Predicted: trendy (0.65)
Ground truth: UNKNOWN (not in exemplars)
Result: UNDETERMINED
dress_P (classic black sheath with cutouts):
Predicted: trendy (0.52)
Reference label: NOT_SUPPLIED
Result: BOUNDARY_POLICY_UNSPECIFIED
The absent boundary policy triggers review; the score itself proves no instability.
The Rejection Witness
The system produces a structured rejection:
REJECTION_WITNESS:
predicate: trendy
proposal_id: PROP-2026-0142
rejection_reason: BOUNDARY_POLICY_MISSING
evidence:
type: missing_boundary_treatment
item: dress_P
conflict:
- positive_signal: cutout feature matches positive exemplar dress_C
- negative_signal: black sheath silhouette matches negative exemplar dress_X
intension_ambiguity:
The intension "reflects current fashion trends" does not resolve
whether a classic silhouette with a trendy feature is trendy.
minimal_fix_options:
1. REFINE_INTENSION:
Change intension to "items where the dominant feature is a current trend"
This could classify dress_P as NOT trendy if silhouette dominance is established
2. ADD_INVARIANT:
Add invariant: trendy → ¬classic_silhouette
This would make the negative classification of dress_P explicit
3. RECLASSIFY_EXEMPLAR:
Move dress_X from negative to boundary
This acknowledges that classic pieces with modern elements are ambiguous
4. NARROW_SCOPE:
Reduce scope to catalog_spring_2026_contemporary
This excludes items with classic silhouettes from evaluation
supporting_record:
type: unresolved_boundary_case
observation: (dress_P, 0.52, [cutout_similarity, sheath_similarity])
establishes: "Required boundary treatment is absent"
does_not_establish: "Logical contradiction or classifier instability"
remediation_deadline: 72 hours
status: REJECTED_PENDING_REMEDIATION
The Remediation Loop
The merchandising team can refine the intension, add a rule, revise an exemplar’s status or narrow the scope, as the rejection record proposes. None of these changes completes the remaining admission checks.
The team chooses Option 2: add trendy → ¬classic_silhouette and trendy → in_current_season. Given the example's classification of dress_P as a classic silhouette, the first rule excludes it. This is a choice about the predicate's extension, not a discovery that the dress was logically impossible to classify.
The Second Acceptance Attempt
The revised proposal now supplies a rule for the disputed item. That resolves this particular objection. The excerpt has not established every invariant, logical conservativity or all the confounder and operational checks required by A29. Its honest receipt is therefore a remediation record:
REMEDIATION_RECEIPT:
proposal_id: PROP-2026-0142
revision: 2
boundary_objection: RESOLVED_BY_DECLARED_EXCLUSION
item: dress_P
governing_rule: trendy → ¬classic_silhouette
remaining_admission_checks: REQUIRED
status: PENDING_ADMISSION
The proposal has advanced. It has not acquired certification by changing its description of one test.
What the Revised Boundary Excludes
A team could make this same choice using an ordinary classifier workflow and a well-kept decision log. The proposed interface's contribution is to make the boundary treatment a required admission question and bind the decision to its version. It earns no exemption from proving that the rest of the gate works.
The revised definition sacrifices one possible use of “trendy”: a classic silhouette cannot qualify merely by acquiring a fashionable detail. Another team might keep that possibility and choose different examples or a different rule. The record should preserve which decision was made, rather than present the chosen boundary as an unavoidable verdict of the classifier.
The Audit Trail
If this revised predicate is later admitted, a user asking why dress_P was excluded can be shown the declared rule and the decision that introduced it. The example's classic-silhouette classification triggers that rule. Changing the answer requires revising the rule or its factual input, not pretending that the earlier decision never occurred.
Consequence
Here the refusal has produced a question the team can answer and a revision another person can inspect. Not every failed proposal has a viable remedy; a structured record must be able to preserve that result too.
The catalog still has no admitted “trendy” filter. It has a revised proposal, a reason for the changed boundary and a record of the work left undone. Passing that record to another team should save an inquiry into why the rule changed; it must not save the checks the record says remain unfinished.