[cschanhniem]

Implements R1 from NEXT_100_PR_MAP — starts Phase R (scientific output readiness).

What

• src/openamp_foundry/evidence/candidate_family_clustering.py: CFC- schema; CandidateFamilyClustering dataclass (15 fields); validate_candidate_family_clustering(); build_candidate_family_clustering() (enforces dry_lab_only=True); format_candidate_family_clustering()
• VALID_NOVELTY_EVIDENCE_TYPES (6): blast_search_no_hits, hmm_profile_no_match, structural_divergence, charge_length_outside_known, expert_assessment, insufficient_evidence
• VALID_CLUSTERING_METHODS (5): sequence_identity, edit_distance, kmer_similarity, embedding_distance, manual_grouping
• VALID_NOVELTY_VERDICTS (4): novel_family, variant_of_known, insufficient_evidence, disputed
• tests/evidence/test_candidate_family_clustering.py: 63 tests in 5 classes
• BASELINE: 8900 → 8963

Disconfirming pass

• dry_lab_only=False blocked (tested) — clustering is computational
• TOY- candidate IDs blocked (tested)
• novel_family requires hard evidence (blast/hmm/structural/charge) — expert_assessment alone insufficient (tested)
• variant_of_known requires known_family_hits (tested)
• n_candidates < 2 blocked (tested) — single-candidate families not allowed
• n_candidates mismatch with candidate_ids length blocked (tested)
• min_pairwise_similarity > max_pairwise_similarity blocked (tested)