Group charts — semantic similarity motifs (B14)¶

Scope¶

Cross-session recurring motifs and drift for aggregation id semantic_similarity (prefer member semantic_similarity payloads).

session_rows — repetition scalars plus motif_count, recurring_motif_count, drift_score (nullable), included_in_comparison, provenance key
content_rows / repetition_rows — unchanged pair concat contract
motif_rows — group-stable motifs with appearances and trajectory slopes
semantic_similarity_pooled (schema_version: semantic_similarity_pooled.1) — strength/share/presence matrices keyed by order_index, truncation metadata, primary_cohort_key, excluded members, warnings

Single registry entry: SemanticSimilarityGroupChartGenerator

Session bars for total_repetitions / unique_patterns (generic)
Null-safe bars for motif_count / recurring_motif_count / drift_score (skip null; never coerce unsupported to 0)
Motif prevalence (group.semantic_similarity.motif_prevalence.global) only when:
- ≥2 comparable sessions in primary cohort
- ≥1 recurring motif
- Strength = cluster size (not match cosine)

Compare + centroid match within a comparable provenance cohort
Include valid-zero motif sessions in the cohort
TF-IDF backends are incomparable (export-only; never cross-match)
Reject vector-dimension mismatches
Session drift_score = 1 - Jaccard of group-motif presence vs previous comparable session; first session is null
Per-motif presence_slope is a separate trajectory metric
group_motif_id: sha1(f"{creating_transcript_id}:{local_motif_id}").hexdigest()[:16] where creating_transcript_id is the first session that introduced the motif
Chart strength uses cluster size (and optionally eligible-segment share); match cosine is evidence-only. Absent motif in a session is strength 0 on the prevalence matrix (not a null motif_count coercion)