Marivo 0.4.13

Marivo 0.4.13 makes multi-axis attribution more useful with typed hierarchy and Top-K grouping, improves native Trino percentile execution, and hardens agent-facing help and artifact recovery.

Highlights

session.attribute(..., top_k=K) now supports fixed, governed Top-K selection across single-axis, joint, and hierarchy attribution. It ranks over the combined current and baseline scope and represents the remainder as an explicit Other player; null values and a real member named Other remain distinguishable.
Multi-axis attribution now uses typed hierarchy semantics. mode="hierarchy" preserves ordered prefixes and per-level reconciliation, including separate resolution semantics for additive/component results and distinct/quantile results.
Trino percentile attribution now plans deterministic Shapley coalitions and evaluates native approx_percentile in bounded batches, reusing independently observed endpoints while preserving reconciliation and approximation-error evidence.
Semi-additive percentile folds now resolve from measure-owned temporal contracts consistently across catalog and runtime planning. Trino sampled percentile folds use APPROX_PERCENTILE, and readiness blocks unobservable fold definitions with repair guidance.

Changed

marivo.help("targets") lists the complete canonical string targets from the installed live registries, including surface, capability, type, and error targets. Unknown-target errors point to this inventory.
Attribution results retain typed Top-K selection, hierarchy paths, Other masks, resolution semantics, and reconciliation metadata so downstream reads can distinguish selection and aggregation behavior.
Artifact persistence publication and recovery are session-aware. Missing session index entries can be recovered only from a committed schema-v4 evidence marker after path, content-hash, and canonical-integrity checks; SQLite lock and busy conditions produce structured lock errors.

Fixed

Interrupted Artifact publication no longer restores an unvalidated registration or loses artifact pagination identity during recovery.
Trino native percentile coalition evaluation avoids unbounded per-coalition execution while retaining endpoint and error semantics.
Percentile fold planning no longer drops measure-owned temporal semantics in nested, derived, or runtime expressions; conflicting or unobservable status axes fail closed.

Documentation

Updated the analysis workflow, semantic layer, evidence persistence and recovery, help-target, and quantile attribution documentation.

Installation

curl -fsSL https://marivo.io/install.sh | bash

In mainland China, use:

curl -fsSL https://marivo.io/install-marivo-cn.sh | bash

For manual installation:

pip install marivo

Marivo 0.4.13 requires Python 3.10 or newer.