Skip to content

v0.19.0

Release Date: 2026-08-22

Highlights

  • New assessment package: an evidence-backed opportunity prioritization system — a judge answers bounded Y/N rubric questions with cited evidence, and deterministic code (never the judge) resolves those answers into a MoSCoW tier, a RICE score, a portfolio classification (Kano, Market Investment Horizon), and a final rank. This is Phases 1-3 (of 6) of initiative INIT-PRISMROADMAP-001; the remaining phases (persistence, compile/review/rank, rendering) land in omniroadmap.
  • JSON Schema generation for the new IR types — OpportunityAssessment, Evidence, DimensionDefinition, and OpportunityRank are now available as generated, embedded JSON Schema, matching the existing PRD schema generation pattern.
  • Pinned to the tagged prism-core v0.5.0 — picks up CapabilityRef/CapabilityRelation, the shared primitive assessment.CapabilityReference is now built on.

Overview

This release lands the IR (intermediate representation) half of a cross-repo initiative spanning prism-roadmap, structured-evaluation, prism-core, and omniroadmap: a prioritization system where every score and category is resolved deterministically from cited evidence rather than assigned or invented — including by an LLM judge, which only ever produces bounded Y/N answers with citations.

The design principle carries through every layer: a Ladder threshold classifier (shared by MoSCoW and RICE Impact/Confidence) never counts a "yes" answer without at least one evidence citation; a RankingPolicy never sorts an unrankable opportunity to the bottom with a fabricated score — it's excluded with a reason instead; and a ReportDataset-driven report never has a renderer compute a fact or an LLM narrative pass invent one — every generated sentence is anchored to a computed fact or cited evidence.

See the new Opportunity Prioritization and Report Contracts guides for the full pipeline, and docs/specs/initiatives/INIT-PRISMROADMAP-001/ for the PRD/TRD driving the design.

Changes

Evidence

Evidence wraps structured-evaluation's claims.Claim with capture metadata (source system, excerpt, capture time, sensitivity) and reverse assessment refs (EvidenceIndex). Sensitivity-gated so restricted excerpts never render verbatim into a report — a renderer cites the evidence ID with an access note instead.

Ladder, MoSCoW, and RICE resolvers

A generic Ladder/ThresholdAnswer top-down threshold classifier, reused by MoSCoW tier and RICE Impact/Confidence resolution. Both resolve into the existing prioritization.MoSCoWPriority/ImpactLevel/ConfidenceLevel types rather than new ones — this package changes how a value is derived, not what it is. RICE effort estimates are gated by a five-check EstimabilityGate so an under-specified plan returns "needs more detail," not a fabricated Person-Days number.

Portfolio dimensions

DimensionDefinition (Category or Tags kind), with built-in Kano (bespoke pattern-matching classifier) and Market Investment Horizon (generic per-option resolver, prism-roadmap's own KTLO/SAM+SOM/TAM-Expansion framework) dimensions. Descriptive only — dimensions never enter RankingPolicy.Rank.

OKRContribution links an opportunity to a goals/okr objective/key result with an evidence-backed strength. CapabilityReference is a type alias for prism-core's CapabilityRef, letting an opportunity reference a prism-capability capability without depending on its full domain model. Neither is a ranking input.

OpportunityAssessment and RankingPolicy

OpportunityAssessment ties Evidence, MoSCoW/RICE, dimensions, OKR contributions, and capability references into one per-cycle judgment record — corrections go through NextCycle(), never mutating a past assessment. RankingPolicy.Rank orders by MoSCoW tier then RICE score with a ±5% tie band; RankOverride records an explicit, auditable governance decision to move an opportunity, never a quiet reweighting of its inputs.

Report contracts

ReportDataset (computed facts, % Person-Day distributions, deltas since the previous review, override log), OpportunityReport (the per-opportunity six-pager with typed, provenance-carrying narrative slots), and PortfolioReview (the whole-portfolio agenda, plus a PresentationProjection() into slide form). All three are pure functions of their inputs — no fact is computed at render time.

Compatibility

Fully additive. The new assessment package and schema additions introduce no changes to any existing Go type, schema, or API. go.mod now requires the tagged github.com/grokify/prism-core v0.5.0 instead of an older version — no breaking changes there either. Existing documents and consumers are unaffected.

Installation

go get github.com/grokify/prism-roadmap@v0.19.0