INVARA 0.2.1 corrects a report wording defect that described search executions as additional generated inputs. Repeated executions and inputs already in the frozen baseline could therefore be mistaken for new coverage.
Reports now distinguish search executions, distinct inputs, overlap with the frozen baseline, new distinct inputs, and repeated executions. Input identity uses the existing exact input and initial-state identity. Counts that cannot be recovered from preserved evidence are shown as UNKNOWN. Korean and English summaries, technical reports, JSON, and newly generated export/inspection surfaces use the same meanings.
This is a reporting patch. Search strategy, comparisons, normalization, verdict rules, and evidence identity protections remain unchanged. NO_DIVERGENCE_FOUND is finite search evidence, not proof. Existing evidence remains bound to the implementation that produced it; report regeneration does not migrate or reexecute an old session.
Install the pinned package:
python -m pip install invara==0.2.1The package retains zero runtime dependencies and the existing six MCP tools.