RIVER, defined.

The framework, on its own terms.

RIVER encompasses the four DORA metrics as the reference instrumentation for the commit-to-deploy segment; there is no RIVER metric that replaces a DORA metric. What RIVER adds is a per-release operating artifact, and measurement scoped to that artifact, for the segments past deploy.

RIVER is philosophy-coupled to the separation of deploy and release. It does not make sense in organizations that ship in a single binary event from commit to user. It is tool-neutral in expression: it can be instrumented on any stack that produces the required data.

The five commitments RIVER is built on.

Descriptive of where the industry's most capable organizations already operate, prescriptive for organizations moving toward that practice. Not negotiable within the framework.

The unit of analysis.

The central abstraction of RIVER is the release delta: a declared, structured artifact created before a release begins.

Release delta is the unit of analysis: team-level, product-line-level, and organization-level rollups aggregate over deltas, not over deploys, flags, tickets, or story points. The act of declaring a delta before shipping is coercive, and the coercion is the point. Declaration forces three questions teams routinely avoid: what we believe will happen, how we will know, and by when.

The baseline, declared.

A success signal is itself a structured component: it declares the metric, the direction and magnitude of expected movement, the time window, and the baseline the movement is judged against. The baseline is the component teams most often leave implicit, and it is the one a declaration cannot survive without. Every delta declares its baseline at declaration time.

RIVER is design-aware and non-prescriptive about how the comparison is made. Evaluation methods span a spectrum, from randomized experimental arms through guarded progressive rollouts to purely observational attribution, and stronger designs produce stronger attribution. The framework names the spectrum; it does not mandate a position on it. Four baseline types cover the spectrum, ordered by the strength of attribution they support.

The ordering carries the claim: attribution is only as strong as the baseline it is judged against. An organization's distribution across baseline types is a fact about its evidentiary practice, not a ranking of its teams.

Six types, each measured against its own standard.

Not every release is supposed to move revenue. The type is assigned at declaration time.

Three signals, not one.

The word "adoption" in common use collapses three distinct phenomena with different time horizons and different evidentiary value.

First-use resolves in hours to days; it is evidence of visibility, not value. Sustained-use resolves in weeks; the feature has found a place in real workflows. Value-realization resolves in weeks to months; the user has completed the action the feature was built to enable, defined per-delta by the declared success signal.

Seven families, mapped to the value chain.

The asymmetry with DORA's four is intentional. Named representative metrics are version-one candidates, subject to empirical refinement. Definitions live in the Glossary.

Five levels. Teams climb into them.

RIVER is a measurement framework and a maturity practice. Teams do not adopt RIVER in a single act; they climb into it over quarters or years through five levels: Deploy, Control, Declare, Prove, Learn. The transition from Control to Declare is the hardest, where measurement stops being instrumentation and starts being ceremony.

The ladder is a language, not a ranking. An organization that can say "we are at Control on most teams and Declare on two pilot teams" has a more tractable conversation about what to invest in next than one that can only say "we are working on our metrics."