NRR-Phi: Text-to-State Mapping for Ambiguity Preservation in LLM Inference

Ambiguity-bearing inputs often pass through interfaces that favor one resolved response before later context arrives. This creates a text-to-state gap: multiple plausible interpretations may be expressible, but not preserved as one manipulable state. We present NRR-Phi, a formal text-to-state mapping (phi: T -> S) for Non-Resolution Reasoning (NRR). The mapping decomposes into conflict detection, interpretation extraction, and state construction, producing non-collapsing states in which multiple interpretations coexist with context tags, weights, and metadata. We instantiate phi with a hybrid extraction design: rule-based segmentation for explicit conflict markers and LLM-assisted enumeration for implicit ambiguity. On a 68-sentence ambiguity test set, the resulting states preserve interpretive multiplicity, with mean state entropy H = 1.087 bits across categories, compared with H = 0 for single-interpretation collapse baselines used as reference points. We also instantiate the rule-based conflict detector for Japanese markers such as kedo and kamoshirenai, illustrating portability of the conflict-detection stage. Appendix operator validation on 580 test cases shows 0% collapse for principle-satisfying operators versus up to 17.8% for violating operators. The contribution is algorithmic: it supplies the bridge from text to retained NRR state, while treating later interface and operational metrics as complementary layers rather than competing definitions of success.