v1.3.0 — distribution-agnostic tower summary

Resolves register C-45 — the tower estimators no longer bake a count-domain magnitude assumption into the domain-agnostic leaf. The max(draws) <= 1.0 "quiet row" rule zeroed every cell of a rate/probability [0,1] target and silently erased low-intensity counts. No change to the frozen estimators (ADR-018); CONFORMANCE_FLOOR stays 1.0.0. pip install --upgrade views-frames.

Changed

• No magnitude-based zeroing by default. tower_point / hdi_tower / summarize_tower / bimodality no longer collapse sub-1 rows to 0. Zero-inflation is handled by the density of the tip_mass floor (a zero-majority row reads 0 naturally) — distribution-agnostic (counts, continuous, normal, beta/probability).
• config['zero_cutoff'] is now an optional, off-by-default opt-in (None). A count consumer that wants "sub-1 ⇒ 0" sets it to a float; it is read live (the prior import-time snapshot that made it non-configurable at runtime is fixed).

Notes

• The modeling choice "should a sub-1 count posterior read 0?" is the consumer's (set zero_cutoff, or apply a downstream mass_at_zero policy) — not a leaf default.
• ADR-019 amended; the Summarize CIC documents the opt-in + the consumer-owns-the-zero-policy note. Register C-45 → Resolved.

Full notes: CHANGELOG.md.