Evidence tiers for strain-resolved long-read metagenomics.

Long-read metagenomics links variants, repeats, and mobile elements across individual molecules and can reveal lineage turnover that is obscured at the species level. The strength of the resulting inference, however, depends on the type and genomic span of the recovered linkage. We distinguish four evidence units-strain profiles, local haplotype blocks, genome-scale strain candidates, and host-linkage evidence-and describe the diagnostics appropriate to each. Confidence is reported separately as candidate, supported, or high confidence in context, based on read assignment, allele coherence, coverage continuity, boundary support, and independent confirmation. Ambiguous placement, contamination, consensus collapse, coverage breaks, reference bias, or unresolved linkage should prompt more conservative terminology. Aligning terms with the available evidence reduces overstatement in claims of strain reconstruction, host assignment, persistence, replacement, and engraftment, while making analyses easier to compare and review.