Decision-Boundary Utilities

Tools for justified AI decisions.

Most AI readiness tools ask whether an organisation can adopt AI. NSDM asks a harder question: can the organisation make AI decisions that are justified, governed, contestable, and safe to act on?

View tools Back to institute

Tool layer

These utilities translate NSDM research into practical checks for executives, AI builders, governance teams, and product owners. They are designed to expose decision risk before a model output becomes an operational action.

01

AI Readiness Assessment

Evaluates whether an organisation is ready to use AI in decision workflows, not only whether it has data, cloud access, or automation ambition.

  • Decision ownership
  • Evidence availability
  • Governance maturity
  • Human override path
02

Use-Case Scorer

Ranks AI use cases by value, risk, evidence quality, governance clarity, and actionability.

  • Business value
  • Decision risk
  • Data readiness
  • Boundary complexity
03

RAG Readiness / Data-Rot Audit

Checks whether retrieval systems are relying on current, authoritative, and decision-relevant sources.

  • Source authority
  • Document freshness
  • Contradictory sources
  • Evidence traceability
04

Decision-Boundary Checklist

Tests whether a model output is supported, under-specified, unsupported, contradicted, or governance-ambiguous.

  • Supported claim
  • Missing premise
  • Contradiction
  • Escalation trigger
05

TokenOps Cost Card

Connects token spend to justified decisions rather than raw usage volume.

  • Cost per justified decision
  • Cost per safe refusal
  • Negative-value token rate
  • Boundary stress-test cost
06

NSDM Governance Workbench

The first executable governance kernel for tenant-scoped Case initiation with explicit identity, authority, versioning and audit boundaries.

  • Identity and tenant resolution
  • CASE_INITIATION authority
  • AssessmentVersion v1
  • CASE_CREATED audit receipt
Exportable decision pack

From model output to defensible decision.

NSDM tools are designed to produce a compact decision pack: the claim, evidence, governing rule, boundary state, action recommendation, and escalation logic.

Evidence state supported, unsupported, contradicted, under-specified, ambiguous governance, or reward-aligned unjustified.
Governance state policy allowed, policy forbidden, authority missing, human review required, audit required.
Action state answer, ask for clarification, request more evidence, abstain, refuse, escalate, or block.
TokenOps value whether the tokens spent produced a justified decision, a safe refusal, or avoidable noise.
Product status - 12 August 2026: The tool layer now includes the first executable NSDM Governance Workbench kernel. M1 has passed fresh-clone migration, tenant-isolation, authority, atomic-transaction, application and CI acceptance. The remaining utilities and later Workbench layers are active research and product directions; they should be treated as decision-support and assurance infrastructure, not automated certification.

Build evidence around the tools.

ACE integrates external runtimes, profilers and schedulers as implementations under test. The emerging asset is a reproducible workload, reference, manifest, trace and adjudication.

Gate 0 and provenance

Establish exact platform compatibility before installing a comparator. Record actual backend, hardware, toolchain and unsupported features.

Optimization admissibility

Equivalence and lifecycle checks precede acceptance of speedups. These are proposed evidence methods, not downloadable completed tools.

Evidence requirements → · Architecture →