200
Independent four-arm runs
Fifty runs in each controlled prompt condition
For Investors and Strategic Partners
A calibration, optimization, and provenance company.
CONEXUS is building a common technical foundation beneath experimental AI methods and human-facing products. The investment case begins with evidence, intellectual property, working prototypes, and the work still required to validate and scale them.
d = 3.78
Neutral-to-CONEXUS effect
Run-level semantic-distance comparison in the tested configuration
30,800
Locked optimization trials
Controlled Forgetting Engine sweep
561%
Largest reported relative gap
One stated 3D protein-folding comparison, not a universal rate
One foundation, three layers
The company thesis is that calibrated search, subtractive optimization, and traceable provenance can reinforce one another.
Layer 1
Calibration
The Nine-Gear ECP architecture structures how models hold competing constraints and explore alternatives.
Layer 2
Optimization
The Forgetting Engine tests whether strategic candidate elimination can improve search under defined budgets.
Layer 3
Provenance
Traceable records connect inputs, transformations, outputs, and authorship history where implemented.
Research program
Controlled calibration experiments and computational optimization benchmarks are being organized into reproducible evidence packages with explicit limitations.
IP portfolio
Patent filings cover areas including calibration, symbolic compression, strategic forgetting, provenance, and collaborative authorship. Filing status is not a guarantee of issued claims or commercial exclusivity.
Product translation
NAiRTHEX and ECHOform test how the underlying architecture can support bounded reflection experiences while preserving human authority and clear product limits.
Diligence boundaries
What a serious investor should verify
CONEXUS does not present editorial interest, internal audits, patent applications, prototypes, or large effect estimates as substitutes for independent technical and commercial diligence.
The four-arm result currently covers one model family, one task, and one configuration.
Forgetting Engine percentages come from different benchmarks and cannot be combined into one performance score.
Internal reports and code availability do not substitute for independent replication or peer review.
Patent filings establish claimed priority positions, not guaranteed issuance, validity, scope, or freedom to operate.
Product concepts and prototypes are at different stages of deployment, testing, and commercial readiness.
Review the evidence before the narrative.
The public Evidence page separates measured findings, benchmark results, hypotheses, and limitations. Additional materials are available for qualified diligence.