Skip to content
Marivo

Marivo 0.4.1

Marivo 0.4.1 expands typed analysis for cumulative and non-additive metrics, adds ontology-guided semantic discovery, and tightens the public contracts that make agent-facing analysis deterministic and verifiable.

Highlights

Section titled “Highlights”
  • Cumulative observations now retain evaluation cutoffs and comparable-period evidence. Analysis supports cumulative level comparisons, all-history level changes, and cumulative attribution with explicit scope, cutoff, and reconciliation validation.
  • Attribution supports count-distinct and quantile measures with independent multiresolution execution, typed evidence, and fail-closed artifact checks.
  • Ontology authoring and catalog loading support one-edge semantic hypothesis discovery, stable candidate selection, and candidate identity and lineage.
  • Typed measure bindings now own metric-axis resolution. Analysis aligns only on declared axes, preserves the relevant unit state, and exposes a closed set of funnel affordances.
  • The installer uses uv to manage Python environments across Unix and Windows Bash layouts. The default installation covers DuckDB, Trino, and ClickHouse; MySQL remains opt-in.

Breaking changes

Section titled “Breaking changes”
  • Public LineageStep, Lineage, FrameSummaryEntry, JobSummary, and SessionSummary data classes are keyword-only.
  • The unbounded session.list operation has been removed. Use bounded session.recent(...) and session.inspect(...) operations instead.
  • Analysis artifact identity is ref-only; removed identifier aliases must be replaced with the canonical .ref contract.
  • Legacy session aliases, compatibility shims, and dead errors have been removed. Use the canonical public names and entry points.
  • All-history cumulative comparison is represented as a level change with persisted evaluation-end evidence, not as interval-flow equivalence.

Changed

Section titled “Changed”
  • Analysis rejects dimensions and output columns that collide with reserved result protocol columns, and validates multi-metric axis and coverage contracts before execution.
  • Attribution hierarchy columns use a namespaced layout, and attribution errors provide actionable repair guidance for unsupported inputs.
  • Development dependencies no longer install MySQL by default, and the type checking baseline is updated to mypy 2.3.

Fixed

Section titled “Fixed”
  • Cumulative comparable-period alignment now fails closed on unsupported legacy calendar identifiers and validates persisted evidence on load.
  • Panel forecast axes, folded derived metrics, cross-entity grain requirements, and artifact reuse purposes now receive explicit validation and metadata.
  • Live help targets, public exports, and runtime representations are kept deterministic and aligned with the installed guidance.

Documentation

Section titled “Documentation”
  • Added cumulative comparison, cumulative attribution, non-additive attribution, ontology extension, and agent-facing public-surface specifications.
  • Updated the metric-first analysis workflow, semantic-layer guidance, and cross-platform installation instructions in English and Chinese.

Installation

Section titled “Installation”
Terminal window
pip install marivo

Install the backend extra for each datasource you plan to use. Marivo 0.4.1 requires Python 3.12 or newer.

Section titled “Documentation links”