ALPHA LAB / PUBLIC OPERATING MANDATE · v2.0.0 · 2026-09-06

An AI-native laboratory for systematic AGI alpha discovery

Twenty-five US equities are experiment zero—not our universe. We are building a global, evidence-bound system that maps advances in machine intelligence through bottlenecks, business models, geographies and second-order consequences into falsifiable investment hypotheses.

We seek an evidence process that can earn alpha; we do not claim that it already has.

α / 00

The story is not “AI will rise.” The investable question is where reality will diverge from the expectations already priced—and whether that divergence survives data quality, access, liquidity, governance, costs and a hostile counter-case.

26model specifications

0IC approvals

0trade authorizations

The system we are building

01
Map change before tickers: capability → adoption → constraint → cash flow → price expectation.
02
Build a global security master with point-in-time data, source lineage, access and investability gates.
03
Run Alpha vs Alpha experiments under matched factor, currency, liquidity, governance, tax and execution assumptions.
04
Publish the forecast before the outcome; preserve failures; let the forward record—not charisma—decide.

A global frontier map

Research queues are not portfolios. Every market must pass access, governance, evidence, valuation and execution gates.

US-0live baseline

United States / experiment zero

25 equities plus a reserve map today’s public AGI stack. It is a research baseline, not a completed global portfolio.

CN-1discovery planned

China / first Alpha vs Alpha challenger

AI acceleration plus depressed valuations may create a wider expectation gap than in the US. This is a hypothesis—not a recommendation or an explanation based on one geopolitical story.

ASIA-2research queue

Asian industrial chain

Taiwan, Japan, Korea, Malaysia and Singapore may own less visible manufacturing, memory, optics, packaging and data-center control points.

INDIA-3discovery queue

India / deployment at population scale

Low-cost intelligence may alter digital distribution, services, labor productivity and demand at a different adoption curve.

UA-4access research

Ukraine / autonomy and defense technology

A live innovation environment may reveal autonomy, sensing and resilient-infrastructure learning before it reaches public markets.

EARTH-5causal mapping

Land, power, water and environmental externalities

AGI infrastructure and productivity can reprice scarce physical inputs—but the transmission may benefit owners, utilities, regulators or communities differently.

Alpha vs Alpha experiments

Comparable experiments turn a seductive story into a contest that can be lost.

EXP-00 · live research baseline

US AGI stack v1

Can a bottleneck-weighted stack outperform broad and thematic comparators after costs without hidden factor concentration?

EXP-01 · designing universe

China AGI prototype

Does valuation compensate for governance, policy, FX, access and accounting risk while preserving AI-linked cash-flow upside?

EXP-02 · causal map

Scarcity migration basket

Can leading evidence identify where rent moves before consensus financial estimates?

EXP-03 · stress design

Displacement chain

Which apparent AI losers are truly exposed, which can adapt, and where is the asymmetric hedge?

Six falsifiable mechanisms

A1 · 1–8 quartersresearch-priority

Capability → economic deployment lag

Mechanism: Benchmarks can jump before reliability, workflow integration, liability and customer budgets allow broad substitution. The lag, not a guessed AGI date, determines revenue and displacement.
Leading evidence: Task-level autonomy, intervention rate, cost per completed workflow, paid retention, enterprise seat contraction and daily work-use intensity.
Fails when: No durable improvement in end-to-end task economics or adoption after two major capability generations.
Portfolio use: Own beneficiaries only when adoption economics improve faster than price-implied expectations; treat benchmark headlines as discovery, not trades.

A2 · 2–12 quartersresearch-priority

Scarcity migrates through the stack

Mechanism: The limiting input can move from accelerators to HBM, packaging, optics, switchgear, turbines, grid interconnection or permitted land. Equity rents migrate with the binding constraint.
Leading evidence: Lead times, book-to-bill, utilization, cancellation clauses, price/cost spread, megawatts energized—not announced—and customer concentration.
Fails when: Capacity additions remove lead times without margin or utilization resilience.
Portfolio use: Re-rank layers from measured scarcity rents; never extrapolate one bottleneck forever.

A3 · 4–16 quartersresearch-priority

AI capex must clear a cash-return hurdle

Mechanism: Revenue and backlog can rise while per-share owner cash deteriorates because of leases, replacement cycles, dilution and financing. The market often prices the build before cohort returns are observable.
Leading evidence: Incremental gross profit versus cash capex and leases, utilization, depreciation life, RPO cash schedule, dilution and net debt.
Fails when: Cohort cash returns remain below the required return through a full capacity ramp.
Portfolio use: Prefer companies converting scarcity into per-share cash; reduce conviction when accounting growth outruns funded economics.

A4 · days–4 quartersexperimental

Narrative reflexivity creates price/evidence gaps

Mechanism: A coherent future-history can force investors to re-map correlated exposures before fundamentals change. Viral attention is a flow signal, not proof that the scenario is true.
Leading evidence: Source reach, cross-asset breadth, abnormal volume, factor-neutral residual returns, options skew and subsequent primary-data confirmation.
Fails when: The move is explained by macro, earnings or pre-existing momentum and reverses without new evidence.
Portfolio use: Run an event study before attributing causality. Never claim profit without positions, timestamps, borrow, fills and exits.

A5 · 2–12 quarterstail-risk

The intelligence-displacement transmission chain

Mechanism: If cognitive substitution outruns job creation and policy, seat-based software, friction tolls, high-income consumption, private credit and prime mortgages can become one correlated chain.
Leading evidence: White-collar postings/hiring, seat counts, renewal discounts, upper-income spending, HELOC/401(k) stress, software-credit marks and geographically concentrated delinquencies.
Fails when: AI remains complementary: hiring, real labor income and consumption broaden while diffusion rises, with timely fiscal/institutional adaptation.
Portfolio use: Treat this as a stress regime, not a central forecast. Test defensive assets and counterparties before the chain is visible in headline unemployment.

A6 · 2–20 quarterstail-risk

Security and policy can reprice the whole stack

Mechanism: Export controls, model-weight security, national compute projects, permitting and power policy can abruptly change accessible markets, cost of capital and strategic rents.
Leading evidence: Binding regulations, licences, appropriations, procurement, interconnection approvals, sanctions and disclosed geographic revenue exposure.
Fails when: Announcements do not become enforceable rules, funded contracts or capacity approvals.
Portfolio use: Use scenario caps and geographic diversification; a press release is not a policy cash flow.

Discovery loop

01
Write a causal graph from capability to cash flow before selecting tickers.
02
Write both a future-history and its strongest counter-scenario.
03
Translate each link into a dated leading indicator and explicit falsifier.
04
Estimate what the current price already requires; compare expectations, not stories.
05
Construct factor-aware baskets and predeclare costs, capacity and exit rules.
06
Log the forecast before the outcome; retain failed hypotheses and measure net alpha prospectively.

From evidence to capital

01
Observe: ingest allowlisted primary sources and market data with timestamps and lineage.
02
Verify: reconcile units, periods, corporate actions and contradictory disclosures.
03
Falsify: write the strongest counter-case and predeclare expiry and failure conditions.
04
Value: compare scenario cash flows with expectations embedded in price.
05
Construct: control factor, currency, geography, liquidity, tax, costs, capacity and exit risk.
06
Authorize: AI proposes; mandate checks and authorized humans approve, execute and audit.

Research in public

LIVE / OPEN NOTEBOOK

The work is the show: question → source → model → contradiction → decision → result → next chapter. We publish plans, revisions and failures across the site and media channels so an audience can inspect, challenge and improve the research. Participation is welcome; popularity never becomes evidence and capital never becomes content theatre.

AGI HEDGE FUND / PEOPLE ↗

25 normalization workpapers

The US baseline is chapter zero. Every equity has a model family, adjustment protocol and blocking condition; completion is not valuation approval.

Source map

publisher claimsCitrini Research — Aboutmethod · accessed 2026-09-06 ↗scenario not predictionThe 2028 Global Intelligence Crisisscenario · 2026-02-22 ↗institutional counterargumentThe 2026 Global Intelligence Crisiscountercase · 2026-02-24 ↗forecast not factSituational Awareness: The Decade Aheadforecast_map · 2024-06 ↗forecast not factFrom GPT-4 to AGI: Counting the OOMscapability_hypothesis · 2024-06 ↗rough scenario not budgetRacing to the Trillion-Dollar Clusterphysical_stack_hypothesis · 2024-06 ↗

SOURCE RADAR / DAILY

Allowlisted source change radar

A scheduled machine checks every allowlisted source through direct retrieval or an official-domain search fallback. A changed page does not become a fact or an order.

Last check2026-09-18T22:01:08.921Z

What runs automatically

GitHub Actions starts collection every day at 23:17 UTC. An AI analyst checks the sources and an independent AI editor decides whether an evidence-bound edition may publish under the standing mandate.

Machine: discover, verify, draft and independently edit
Human: standing mandate, corrections and real-capital boundary

21sources

16primary/system sources

13awaiting evidence

2automatic search fallback

RSS ITEM DIFF

Recent RSS item review queue

OpenAI News RSS · new itemAI check queued

How Cooley is accelerating IPO work with ChatGPT ↗Published: Sep 17, 2026, 12:00 PM UTC · Detected: Sep 18, 2026, 3:46 PM UTC

OpenAI News RSS · new itemAI check queued

Introducing Astra for Law ↗Published: Sep 17, 2026, 12:00 AM UTC · Detected: Sep 18, 2026, 3:46 PM UTC