Evidence checked · HNI-2026-001
Tool Scope, Rules Context, and Token Economics
A 12-run local factorial found that scoped tools dominated the measured savings while rules removal was approximately neutral at the median.
Read Experiment Note 001 →Research program
A paper trail that can disagree with us.
Hypernovelty Institute runs small, reproducible studies on agent systems, verification, forecasting, research tools, and institutional adaptation. Credibility comes from protocols, evidence, disclosed limits, and a record that keeps null and contrary results visible.
Current record
Research starts with what can be checked.
The library separates exploratory notes, registered studies, replications, benchmarks, and technical reports.
Operating commitment
Contrary results remain part of the Institute.
Protocols require publication of null, negative, mixed, and inconclusive findings alongside results that support the initial thesis.
Inspect the rules →Machine-readable
A registry, not a highlights reel.
Every formal study receives a stable identifier, protocol hash, evidence receipt, disclosed AI roles, limitations, and verification status.
Open registry JSON →Research lifecycle
From signal to durable record.
Find a question that can lose.
The observation becomes a falsifiable comparison with a rival or null explanation.
Fix the rules before the result.
Primary outcomes, sample plan, stopping rule, exclusions, analysis, ethics, and disclosure are hashed.
Separate measurement from self-report.
Locked automated checks and evidence hashes decide whether a result can enter the registry. Their level is labeled: internal checks are not external review.
Research posture
Independent does not mean unaccountable.
HNI does not claim university accreditation, institutional peer review, nonprofit status, or external ethics oversight that it does not possess. Support, computing credits, consulting relationships, and conflicts are disclosed per study.
How the Institute is organized →