Bridge
Empirical Witness for Compositional Failure
Can bilateral validity coexist with failure to specify a coherent bridge?
Identification 100%, specification 0% — topology determines what bridges are needed but not how to implement them
The hidden-repair study recovered 14 of 15 repair types exactly and one partially, with no misses; it did not independently test Gate 2, Gate 3, or policy canonization.
Abstract
Demonstrates that bilateral validity can coexist with compositional failure in real systems. Establishes the identification/specification gap: LLMs correctly identify topologically prescribed bridge types but fail to specify them.
The frozen hidden-repair study is a candidate-generation baseline, not an independent Gate 2, Gate 3, or frontier-model canonization test.
Reproduction failure under the disclosed hidden-repair protocol falsifies the retained baseline; the untested Gate 2, Gate 3, and canonization claims receive no positive status.
not requested
- is operationalized by Seam