Non-Trivial Intelligence Systems

NTIS is a provenance-preserving longitudinal reasoning analysis platform. Its purpose is to keep the path from source material to judgment inspectable as it changes:

source → evidence → premise → inference → conclusion → change → provenance

A short description is observability for human reasoning. NTIS asks three distinct questions: What evidence was available? How was it connected to judgment? How did the evidence, reasoning, and judgment change?

Why ordinary model output is not enough

A plausible answer does not show where it came from, whether a statement was attributed to the right person, or what changed between two analyses. NTIS preserves source identity and exact spans, then makes the interpretive steps explicit.

Its governing boundary is probabilistic perception; deterministic analysis. AI/ML can help classify, extract, retrieve, route, propose, critique, or align candidate artifacts. A model’s confidence is producer metadata—not analytical weight, evidence strength, or a determination of correctness.

Canonical meaning requires adjudication before deterministic analysis. Learned output does not directly decide drift, reasoning quality, evidence strength, validity, cognitive condition, confidence degradation, or longitudinal scores.

What is working today

The stabilized foundation includes provenance-preserving source records and exact spans; attribution that distinguishes sender, subject, contributor, and proposition holder; explicit proposition observations; immutable correction history; curated reasoning states; and a deterministic Recursive Comparison Engine (RCE) for accepted reasoning states.

The RCE compares eligible, explicitly aligned states structurally. It can represent transitions such as retained, introduced, changed, reversed, missing, not-applicable, and not-comparable. It does not infer analytical quality or validate a drift method.

What remains experimental

Automatic perception and extraction are human-supervised research work. Current focus includes proposition-holder resolution, temporal and status interpretation, reasoning-role classification, quotation and report attribution, and selective abstention or review. Generic linguistic structure and local learned models may improve candidate perception, but unattended reasoning reconstruction is not solved.

NTIS is pre-commercial research and development, not a production deployment or validated measurement product.