中文
E8 Intelligence / AI Workflow

A cognitive operating system for the post-LLM economy.

Not a hedge fund. Not a prompt framework. A platform where human's uniquely irrational creativity meets AI's extreme rational validation — tested first in investment research, proven on a second unrelated domain, built for the profound disruption coming to knowledge work.

Document / Founder Brief · 2026-04-20
24 Agents 2 Domains Live 6,400+ Backtests

Most knowledge workers will become unable to contribute cognitively before 2035.

As LLM capability compounds, human relative contribution in most workflows approaches zero — not because AI replaces jobs or knowledge creation, but because most people stop thinking alongside AI. Only a small cohort will actively compound their cognitive capability through LLM collaboration, retaining meaningful leverage in the AI era. Everyone else will see their contribution decline until they are assigned to roles that no longer matter.

01

The Math

Human capability growth rate must stay positive relative to LLM capability growth rate. Most workers have zero or negative growth — they treat the LLM merely as a tool, not as a partner that elevates their thinking. In any workflow, their relative weight is declining toward zero.

02

The Exception

A small group treats LLM conversation as rational-depth accelerant. Deeper rational mastery reveals the boundary of reason — and beyond that boundary lies the territory of irrationality. LLM cannot enter this territory. Only human can, and the advantage gained there is both structural and durable.

03

The Market

As AI disruption displaces knowledge work, demand for teachable methodologies of human-AI cognitive partnership will multiply. Not prompt engineering courses — those commoditize fast. Actual frameworks for sustained cognitive sovereignty.

04

The Build

Agent OS is not an investment tool. It is a working prototype of a replicable human-AI compounding cognitive model — built around cognitive deconstruction and reconstruction, institutional memory, governance discipline, and platform-domain separation. Investment is the first proving ground.

A framework, not a methodology.

Every Agent OS action — across all domains, all agents, all timescales — follows one recursive discipline. It is the reason this platform produces different outcomes from generic AI tooling.

Think out of the box.
Implement in the box.
scenario analysis
= quantified uncertainty
+ strong logic argumentation
+ evidence backup
Layer 0 · Generation · Human Only
Irrationality at the Boundary of Reason
Off-consensus hypothesis generation from cross-domain synthesis. LLM cannot operate here — by mathematical structure, not capability limitation.
Layer 1 · Quantification · Human + LLM
Rational Scaffolding
Turn irrational conviction into testable claims. Magnitude ranges, probability distributions, falsification conditions, evidence backup.
Layer 2 · Execution · Agent OS Autonomous
Systematic Implementation
Rule-based entry, rebalance, exit. Zero discretion once thesis committed. Checkpoint-based long-running pipelines.
Layer 3 · Monitoring · Multi-LLM Adversarial
Continuous Validation & Assumption Deconstruction
Multiple LLMs challenge the thesis weekly. Falsification conditions tested against live data. Type-1 vs Type-2 classification continuous.
Layer 4 · Memory · Platform Infrastructure
Institutional Compounding
Forbidden Conclusions register. Research Conclusions tracker. Multi-tier knowledge compression. Every outcome feeds future thesis generation.

Platform and domain, cleanly separated.

The reason Agent OS extends beyond investment is architectural. Platform layer owns governance, communication, memory, and cumulative knowledge management discipline. Domain modules own their subject matter. The interface between them is contract-defined and platform-enforced.

24Specialized Agents
2Domains Live
6,400+Backtests Executed
80+Forbidden Conclusions
Platform Layer · Non-Negotiable Foundation
Governance · Communication · Memory · Cumulative Knowledge Management
Role boundary enforcement. Two-phase status tracking. Cross-agent messaging protocol. Institutional memory infrastructure. Cumulative knowledge management with pipeline checkpoint discipline. Dual-track fallback. Audit requirements.
XiaoHui · COO XiaoWai · Deputy XiaoJian · Audit XiaoTie · IT Ops XiaoLu · Compliance
Module 01 · Live · Investment Research
Quantitative Equity + Regime-Matched ETF
S&P 500 factor research with multi-factor ensemble. Macro-regime framework with multi-signal integration. 8-regime ETF rotation matched to current regime state. Multi-scenario overlay system. First live proving ground for platform.
XiaoTong · Lead XiaoYin · Factor XiaoYan · ETF XiaoJing · Macro XiaoZhan · Geopolitics XiaoZhi · AI/Tech XiaoDun · Risk XiaoDao · Execution (planned)
Module 02 · Production · ZiWei Diagnostic System
Individual Chart / Family Chart / Family Dynamics / Disease Prediction
Four production-ready capabilities on traditional Chinese astrological framework. Full JSON schema validation (600+ lines), 413-line orchestrator, 179 KB curated knowledge base. Chart engine now integrated into investment module as second signal for backtesting — direct cross-domain validation.
XiaoHeng · Lead XiaoWei · Chart XiaoXing · Reading XiaoYuan · Customer
Module 03 · Building · Content Pipeline
External Communication Layer
Three-layer content generation: Daily Memo, Weekly Digest, CEO Letter. Three-party external review (Opus / ChatGPT / human). SSOT data binding policy for performance reporting accuracy.
XiaoMo · Content XiaoGuan · Data XiaoTie · Automation
Module 04-N · Planned · Future Domains
Cross-Domain Expansion
Traditional Chinese arts and body-mind-spirit modern applications. Each new domain tests platform generalizability and adds institutional revenue. Same governance, different subject matter.
To be defined

Platform claims are cheap. Two domains in production are not.

V11 Agent OS was built around investment research. It was then deployed — without architectural modification — to a completely unrelated domain: family dynamics analysis using traditional Chinese astrological frameworks. The chart engine has since been integrated back into the investment module as a second signal for backtesting. Both directions of integration work on identical platform infrastructure.

Module 01 · Production

Investment Research

Quantitative equity + regime-matched ETF
  • InputS&P 500 price series, macro indicators, fundamentals 2003–2026
  • Core LogicMulti-factor ensemble, multi-axis regime framework, 8-regime ETF rotation
  • ValidationWalk-forward, bootstrap 1000x, PIT-clean backtest, 100% trade audit
  • OutputWeekly portfolio, daily memo, investor letter
  • GovernanceCompliance gate, three-party external review, SSOT binding
  • Assets6,400+ backtests, 80+ forbidden conclusions, 11-step pipeline
Module 02 · Production

ZiWei Diagnostic System

Four production-ready capabilities
  • Capability 1Individual chart construction and reading
  • Capability 2Family chart integration across members
  • Capability 3Family dynamics analysis with pair scoring
  • Capability 4Disease etiology prediction via chart analysis
  • IntegrationChart engine feeds investment module as second signal for backtesting
  • AssetsSchema validated, 179 KB knowledge base, validated on real cases
The Critical Observation

The architectural primitives are identical.

Both modules use the same Agent OS platform layer: role-bounded agents, two-phase status tracking, pipeline checkpoint discipline, knowledge isolation via lazy-loaded context, sequential orchestration to prevent redundant reasoning, audit hooks for output verification, and living-document SSOT governance. The subject matter differs by orders of magnitude. The infrastructure is the same stack. That is platform value.

Shared platform primitives

These are what actually transfer across domains.

01

Role-Bounded Agents

Each specialist has defined scope and non-overlapping authority. Platform enforces, not subject matter.

02

Sequential Orchestration

Earlier outputs constrain later reasoning. Prevents redundant analysis and context contamination.

03

Knowledge Isolation

Each module loads only its own knowledge base. Investment and ZiWei never cross-contaminate.

04

Schema-Validated I/O

Every inter-agent handoff is schema-checked. Factor CSV or diagnostic JSON — same discipline.

05

Checkpoint Pipeline

Long-running multi-step workflows resume from last success. Backtest or diagnostic — same infrastructure.

06

Audit Hooks

Every computation verifies its own output. Trade count matching or score sum validation — same principle.

07

Institutional Memory

Forbidden Conclusions. Research Conclusions. Pending registry. Every decision leaves an audit trail.

08

Two-Phase Status

Claim before action, done after delivery. Enforced across all 24 agents in both modules.

09

External Review Gate

Three-party validation (two external LLMs plus human). Applied identically across all output types.

One platform. Many proving grounds.

Each domain validates the platform from a different angle. Investment tests quantitative discipline and governance under regulatory scrutiny. ZiWei tests absorption of non-quantitative signal and qualitative reasoning. Content tests external communication with SSOT binding. Future domains extend proof surface further.

01
Investment Research
V11 CORE
Quantitative equity alpha + regime-matched ETF sleeve. Multi-axis macro framework, PIT-clean backtest discipline, multi-factor ensemble, automated weekly pipeline. Live · 2025-09
02
ZiWei Diagnostic
紫微斗數
Four production capabilities: individual chart, family chart, family dynamics analysis, disease prediction. Chart engine integrated into V11 as second signal for backtesting. Production
03
Content Pipeline
Investor Comms
Three-layer content generation (daily memo to weekly digest to CEO letter). Three-party external review. SSOT data binding policy prevents performance misrepresentation. Building
04
Scenario Satellite
Thematic Basket
Multi-scenario basket construction. Pre-registered thesis, falsification conditions, time-boxed duration. Maximum 10% AUM allocation. Design · Q2
05
Chinese Medicine Research
中醫問診
Traditional Chinese diagnostic reasoning extended through multi-agent framework. Platform generalization test into adjacent traditional knowledge domain. Natural extension from ZiWei module. Future · 2027
06
AI Collaboration IQ System
Meta-Domain
Multi-specialist consultation protocol with differential reasoning via multi-agent framework. Type-1 vs Type-2 irrationality distinction applied to epistemology. Teachable framework for human cognitive amplification. Future · 2028
07
Scientific Research
Hypothesis Lab
Hypothesis generation plus adversarial validation for research teams. Via Negativa discipline. Forbidden conclusions registry applied to research hypotheses. Future · 2029

The most valuable asset is not the fund. It is the operating system that lets one human compound cognitive capability faster than the machine disappears them.

V11 Founder Thesis · 2026