Module hypothesis 

Hypothesis-loop mismatch classification for the agent runloop.

desirable loop: hypothesis → observation → mismatch → inspect evidence → revise hypothesis. This module is the pure, testable core of that loop: it names the mismatch, but never blocks, retries, or performs I/O. The binary’s deterministic failure diagnosis calls into it to frame the model-facing next_action; the evaluator scores the behavior.

Dimension key for MismatchEvidence (index → field → meaning):

field	meaning when set
exit_code	tool result carried this process exit status
empty_search_no_match	grep-style exit 1 with no output inside the queried scope
patch_mismatch	patch context/deletion lines do not match the current file
verifier_failed	a standalone verifier reported a non-zero status
repeated_evidence	query returned only previously seen lines/ranges

Structs

MismatchEvidence

Structured evidence for one mismatch classification.

Enums

MismatchKind

Named mismatch between the agent’s hypothesis and its observation.

Functions

append_revision_guidance

Append revision guidance to a deterministic next_action.

classify_mismatch

Classify mismatch evidence into its highest-priority kind.

revision_guidance

Model-facing revision guidance for a classified mismatch.