Marivo 0.4.8
Marivo 0.4.8 expands typed runtime composition and agent-facing discovery, adds richer JSON datasource projections, and makes temporal and quality evidence more explicit.
Highlights
Section titled “Highlights”mv.runtime_metric.linear(add=[...], subtract=[...], label=...)constructs a bounded sum or difference over governed metrics. The expression supports replay, nested composition, component metadata, readiness, and observation; numeric child values are normalized at evaluation so Decimal- and float-backed metrics can participate together.md.json(...)accepts list-valued query, body, and runtime source parameters and persists them through authoring and snapshots. Schema fields can select nested values with dotted or indexed paths such asa.b,a[0].b, anda[].b; declared types are validated as Ibis type strings before execution.- Analysis discovery now starts with compact root help and a guarded first observation path. Semantic catalog browsing can move from typed collection lookup to details and typed refs without manually reconstructing registry keys.
- Multi-axis attribution defaults to
joint.decomposeand funnel modes remain explicit, and returned contracts expose the effective choice.
Changed
Section titled “Changed”- Semantic-period observations now distinguish scope coverage from data arrival. Frames expose the observed data extent and full-data signal, and evidence marks a truncated tail bucket instead of treating it as complete.
- Fused multi-metric observations preserve the semantic period metadata carried by the materialized frame, including output period keys when available.
- Period lookup repair stays procedural through calendar details, enumeration, and readiness rather than sampled key candidates.
- Quality assessment validates admissible frame shapes and can flag low value density in time-series and panel frames with structured repair guidance.
- Readiness aggregates missing guardrails into one warning with the affected refs in its details, keeping actual blockers easier to see.
- Catalog lookups now round-trip displayed same-kind typed keys while retaining kind and scope validation, with clear failures for missing, malformed, or wrong-kind keys.
Installation
Section titled “Installation”curl -fsSL https://marivo.io/install.sh | bashIn mainland China, use:
curl -fsSL https://marivo.io/install-marivo-cn.sh | bashFor manual installation:
pip install marivoInstall the backend extra for each datasource you plan to use. Marivo 0.4.8 requires Python 3.10 or newer.