Neverthought
Menu
Validation

Validation &
Claim Boundaries

The bench is designed around a simple principle. The system structures judgement. It does not replace it.

§ 01 / Permanent

Permanent claim boundaries

These are not phase boundaries. They will not be lifted in a later version.

  • ⊘No readiness scores
  • ⊘No diagnostic findings
  • ⊘No risk ratings
  • ⊘No compliance determinations
  • ⊘No legal conclusions
  • ⊘No deployment approval
  • ⊘No liability assessments
§ 02 / v0.2

v0.2 implementation boundaries

These describe the current technical posture of the Mode 2 demo. They may change in later versions, by deliberate design choice and with explicit consent flows.

Local-only
Browser-based
No backend
No lead capture
No data transmission
No telemetry
§ 03 / Inference

No autonomous inference on the customer-facing surface

The Deployment Judgement Snapshot does not use an LLM to interpret participant responses. It structures participant-provided context into a reviewable snapshot.

This is deliberate. The bench is designed to preserve human judgement rather than simulate it. Customer-facing outputs remain bounded by question structure, answer metadata, schema, and participant-provided context.

Agents may support research, scenario generation, and build-time tooling. They do not generate customer-facing conclusions in the current Mode 2 surface.

§ 04 / Checks

Validation checks

The current Mode 2 package includes local validation scripts that check the following properties of any snapshot output:

  • Schema compliance
    Output conforms to the snapshot schema. No fields outside the contract.
  • Prohibited fields
    No score, rating, grade, classification, risk level, or compliance status fields.
  • Claim-language leakage
    No diagnostic, evaluative, or approval language in copy or output.
  • Free-text preservation
    Participant responses are reflected verbatim — not summarised, classified, or reinterpreted.
  • Answer classification metadata
    Classifications are user-assigned, not system-derived.
  • Mode 3 absence
    No Mode 3 evidence-scaffold material is present in Mode 2 output.
  • Local-only operation
    No network requests during snapshot generation.
§ 05 / Artefacts

Artefacts under pilot / NDA

The following artefacts are available in pilot or partner contexts because they contain the working machinery: schemas, claim boundaries, validation reports, scenario logic, and scaffold specifications. The public site explains the method; partner review exposes the instrument.

  • Mode Boundary Contract
  • Snapshot Output Schema
  • Claim Audit
  • Validation Report
  • Manual Demo Test Plan
  • Mode 1 Simulator (private)
  • Mode 3 Evidence Scaffold Specification
  • Wind tunnel pattern library
Boundary

If you are a buyer wondering why this page exists in this much detail: it exists because the distinction between an instrument that helps an organisation see itself and a tool that pretends to see it for them is the only distinction that matters. Everything else is decoration.

Discuss a pilot