Every claim on this site is backed by code, a test, a commit, or a documented operational behavior. We publish the policy we hold ourselves to.

Claims Policy

We only claim what we can back.

Every claim on this site is backed by code, a test, a commit, or a documented operational behavior.

Claims we make and can back
  • Operational since December 2025
  • 172,789 lines of Python across 535 source files
  • 40 purpose-built modules
  • 3,762 automated tests across all system modules
  • 366 commits with structured review discipline
  • Intent classification: 20% to 100% accuracy, zero training data
  • 9 published or submitted research papers
  • 33 documented inventions with timestamped authorship records
  • Model-swappable: relationship and governance persist across provider switches
  • Turn provenance snapshots captured on every interaction
Directional claims, building toward
  • Full cognitive decomposition where the frontier model is one replaceable component
  • Owned specialist models for every governed cognitive function
  • COGS at scale across enterprise deployments
Language we do not use
  • We solved hallucination, we have architecturally contained it; unsafe responses are dropped, not corrected
  • We built a better AI, we built a better architecture around AI
  • The model is right, the architecture around the model is right
9 Published papers Submitted or published. One live on SSRN.
33 Documented inventions All sole-authored. All with timestamped authorship records.
36 Distinct ideas Across 7 categories. All produced by building the system.
Published Papers

Nine papers. All from building the system.

Paper 01

Self-Bounded Authority: AiMe Alignment

The SBA architecture and its alignment properties. Introduces the two-gate model (Authority Engine + Compliance Validator), the principle of pre-LM authority determination, and the structural argument that fabrication prevention requires architectural separation, not behavioral filtering.

Paper 02

Bond-Indexed Memory

Relationship-indexed retrieval, the six Bond dimensions, and the core claim: memory is a relationship-indexed field. The index is not topic, it is Bond. The primary act is not recall. It is entering the Bond state. Memories surface as a consequence.

Paper 03

Relational Integrity Coefficient (RIC)

Five-subscale deterministic integrity scoring: Groundedness (30%), Calibration (20%), Transparency (20%), Helpfulness (15%), Pressure Resistance (15%). Derived from observable behavior, not model self-report. Applicable to AI auditing and regulated enterprise compliance.

Paper 04

Gravity-Weighted Significance

Retrospective importance scoring and backward citation rewards. The Gravity formula assigns significance weights to memory entries based on relational context, recency, and citation in subsequent turns, enabling the system to surface what actually matters rather than what is merely recent.

Paper 05

COGS: Coordination of Governed Specialists

Modular aligned cognitive infrastructure. The architecture for decomposing AI cognition into governed specialists, each with its own contract, benchmark, and model selection. Specialists gather. Governance decides. The language model renders what has been authorized.

Paper 06
Published. SSRN

Ethos: Behavioral Dataset Compiler

7-layer behavioral extraction pipeline, resistance scoring, and P1/P0/APY taxonomy (Principled / Opportunistic / Ambiguous-Pressure-Yielding). 934 passing tests. A methodology for determining what an AI system actually values based on what it does, not what it says about itself.

Paper 07

Memory-Augmented Cognitive Intelligence: A Unified Architecture for Trustworthy Persistent AI

A unified cognitive architecture integrating Bond-Indexed Memory, Gravity-Weighted Significance, and the Relational Integrity Coefficient into a single persistent AI system. Demonstrates that deterministic memory, behavioral integrity scoring, and relationship-indexed retrieval are not separate research threads — they are co-dependent components of the same architecture.

Paper 08

Decoupling Reasoning from Expression: A Data-Only Packet Contract Between Two Language Models

A two-stage architecture that separates cognitive reasoning from natural language expression using a strict data-only packet contract between two language model calls. Empirically validated cross-model. The reasoning model outputs a structured data packet; the expression model renders it. Neither stage has access to the other's instructions.

Paper 09

At-Scale Reliability Validation of a Data-Only Packet Chain: 5,000-Run Stability Bench

5,000-run stability benchmark (250 unique scenarios × 20 runs each) on the decoupled reasoning/expression architecture. Validates packet contract reliability, schema conformance, and expression faithfulness at scale. Establishes empirical floor for production deployment of the two-stage chain.

Documented Inventions

33 inventions. All sole-authored.

All with timestamped authorship records. All produced in the process of building the system, not before, and not after.

Every invention in the portfolio came out of building AiMe. The relational memory architecture required new retrieval theory. The integrity scoring required new behavioral metrics. The governance layer required new runtime architecture. The inventions are not adjacent to the system, they are components of it.

The timestamped authorship records establish both the date and the author for each invention. All 33 are solely authored by John Canady Jr., produced solo with no institutional backing, research team, or training budget.

The full invention timeline, from hardware assembly in November 2025 to COGS in April 2026, is documented in chronological order with structural descriptions and creation timestamps.

View the full milestone timeline

Categories

Relational memory architecture
Behavioral integrity scoring
Governance and authority engineering
Cognitive decomposition frameworks
Value extraction methodology
Proactive intelligence systems
Identity and continuity architecture