SYSTEMS / PUBLIC REGISTRY

Systems built around evidence, state and authority.

Real systems at their real maturity. Each page says what exists, what is private, and what is still unfinished.

01Working system

Implemented and exercised in a bounded environment.

02Private runtime

Operational architecture with private data and infrastructure.

03Research kernel

Production-shaped mechanisms tested against synthetic scenarios.

04Reference implementation

Reusable contracts and runnable patterns, not a hosted service.

SYSTEM FOUNDATIONS

Four systems underneath the lab.

Four shared systems support delivery, operations, research and data quality.

Private system

Multi-Agent Engineering Control Plane

A governed delivery system in which product, architecture, implementation, review and release remain separate accountable roles.

SNAPSHOT / AUG 2026 — 293 commits in first 3 weeks · 145 test files · 18 migrations · independent verifier on every release

Decision problem

Agentic delivery becomes unreliable when the same agent defines the task, writes the solution and approves its own work.

Architecture

Machine-readable contracts coordinate specialist agents. A fresh verifier must reproduce the evidence before a release receipt can be assembled.

Verified mechanism

Persistent task context · independent review · executable quality gates · controlled release authority

Public boundary

Internal operating system. Project code, delivery data and runtime identities remain private.

Private runtime

Persistent AI Operations Runtime

An event-driven runtime for AI operators working across messaging, structured memory and controlled action queues.

SNAPSHOT / AUG 2026 — 545 test files · 56 versioned migrations · gated releases with install receipts and evidence bundles

Decision problem

A useful agent must survive restarts, duplicate events and long-running work without gaining unrestricted real-world authority.

Architecture

Verified ingress, durable queues, version-pinned sessions, risk gates and least-privilege workers create explicit control boundaries.

Verified mechanism

Verified ingress · durable jobs · persistent context · approval-aware outbound actions

Public boundary

Private runtime. No personal messages, identities or operational records are exposed publicly.

Working research system

Quant Research Knowledge Engine

A research environment for navigating a large technical corpus through lexical, semantic and evidence-aware retrieval.

Decision problem

Finding a plausible paper is easy. Reconstructing an evidence trail across thousands of documents is not.

Architecture

A portable full-text database and server-backed hybrid retrieval combine dense and sparse recall, reranking and source-level traceability.

Project snapshot / July 2026

25,095 abstracts · 18,376 full-text papers · 365,179 chunks · 558 MB portable FTS5 index

+ 61-volume technical book corpus alongside the paper index

Public boundary

The research layer and corpus scale are verified. Source documents, proprietary notes and retrieval-quality claims are not published here.

Research kernel

Evidence & Data Correctness Fabric

A source-to-answer architecture for systems that must know where data came from, whether it may be used and when it became invalid.

Decision problem

Retrieval quality alone cannot protect a decision when sources are stale, revoked, duplicated or incorrectly projected.

Architecture

Source contracts, immutable evidence, lineage, rights and freshness rules are tested against replay, duplicate delivery and recovery scenarios.

Verified mechanism

Provenance · re-authorization · abstention · quarantine · deterministic replay and recovery

Public boundary

Production-shaped R&D validated with synthetic fault scenarios, not a public production data plane.

APPLIED R&D PORTFOLIO

Six programs turning the architecture into products.

Six active programs apply the same architecture to real research and operating problems. Their labels show what is ready today.

A1Working local system / Private beta

Investor intelligence

Investor Decision Radar

Public-evidence investor monitoring

A research operating system for monitoring publicly disclosed decisions by venture, growth and crossover investors.

MECHANISM: Separates managers, companies, investment events and follow-up work; preserves provenance and history in PostgreSQL; assigns persistent research threads to each entity.
PUBLIC BOUNDARY: Private R&D system. It monitors public evidence; it does not claim privileged investor information or produce investment advice.

A2Working self-hosted product / design-partner pilots

Counterparty intelligence

RHENIUM 75

From counterparty facts to a controlled decision

Establishes the material facts about a counterparty, shows the source behind every conclusion, and turns them into a decision a company can defend: proceed, proceed on conditions, or decline.

Research coverage across 10 evidence domains · up to 3 targeted research passes per dossier · durable job queue with resume and cancel

MECHANISM: One identified legal entity, research in several passes, every material fact kept with its source and date, an explicit completeness score, then draft deal conditions — advance limits, security, payment terms — plus an affiliation graph and monitoring that checks later changes against the live contract.
PUBLIC BOUNDARY: A registry counts as connected only where a direct connector exists; agent web research is labelled separately. Draft recommendations stay drafts until a person reviews them. Not legal advice. Counterparties, source documents and private dossiers are never published.

A3Private research platform / MVP handoff candidate

Long-horizon research

Tenbagger

AI-native research workstation

An evidence-first environment for long-horizon company research, source graphs, dossiers, valuation gates and alternative signals.

SNAPSHOT / AUG 2026 — 339 dated research documents since April 2026 · forecast-vs-fact journal

MECHANISM: Source ingestion, immutable raw records, evidence extraction, research-agent runs, scoring and review gates are exposed through a dashboard and bounded MCP tools.
PUBLIC BOUNDARY: Research tooling, not an audited investment strategy or performance record. No recommendation is implied by inclusion here.

A4Reference implementation

Agent orchestration

Loop

Verified agent-loop harness

Reusable contracts for orchestrator–worker–verifier systems with durable state, explicit budgets, stop rules and anti-spin protection.

MECHANISM: Deterministic software owns control flow and evidence gates while language models perform bounded reasoning inside the loop.
PUBLIC BOUNDARY: A reusable implementation pattern and runnable examples—not a claim that every loop is autonomous or production deployed.

A5Applied R&D bundle / Working prototype

Computational design

AI Architect

Computational design control plane

A workflow connecting AI agents, parameter specifications, Rhino and Grasshopper to deterministic geometry, metrics and validation artifacts.

MECHANISM: Agents create bounded geometry specifications; deterministic tools build and score the model; view captures, logs and project packages preserve the result.
PUBLIC BOUNDARY: Project-specific R&D. AI assists automation and communication; geometry and metrics remain subject to deterministic validation.

A6Supervised MVP / Implementation package

Executive intelligence

TRadar

Controlled signal intake

A fail-closed intake layer that turns approved communication streams into commitments, risks, deadlines and concise executive digests.

MECHANISM: Allowlisted read-only collection is separated from AI interpretation, task candidates and drafts that still require explicit approval.
PUBLIC BOUNDARY: Private communication data is never shown. The public claim is limited to the architecture and implementation package.

CLAIM STANDARD

Mechanism before marketing.

We publish architecture, maturity and verified snapshots. Private repositories, client material, source documents, messages and operating data stay private.

Start a conversation