No capital action is authorized. The investment-signal assessment is incomplete; this is not a finding that nothing changed.
AGI ALPHA ENGINE / PUBLIC BUILD26 positions8 layersUpdated 04 Sep 2026
AI-NATIVE ALPHA LABORATORY
We are building a global system for finding AGI alpha.
AGI Hedge Fund is building a research and investment system for long-term ownership of companies creating the foundations of future AGI. Every day we publish one verified answer: what changed—or did not change—in AGI, why it matters economically and what it means for capital. We help long-term investors understand how AGI is reshaping the economy and where ownership may be warranted—when evidence and price support it.
Built in public · forecasts timestamped · capital human-authorized
TODAY / RESEARCH RESULT
Nvidia, Google, and Emerald AI organize around flexible AI power demand
Infrastructure operators are increasingly framing AI data centers as flexible grid participants: Nvidia, Google, and Emerald AI reportedly launched a consortium focused on reducing or shifting electricity demand during grid stress. Separate conference takeaways described continued progress on projects already being built, but offered little evidence that new-approval queues are clearing. Policy debate moved in two directions: CNAS argued for tighter US AI export controls, while the US-China Business Council urged cooperation on AI risk management. Markets remain sensitive to both AI enthusiasm and interest rates, with commentary describing alternating buying in AI names and selling in chips. These developments matter because compute availability, power access, policy, and financing conditions jointly shape the speed and cost of AI deployment. However, the packet does not verify a new capability, company-level financial impact, or measurable expectation gap. The appropriate conclusion is heightened monitoring, not a capital action.
No capital action is authorized. The investment-signal assessment is incomplete; this is not a finding that nothing changed.
No capital action is authorized. The investment-signal assessment is incomplete; this is not a finding that nothing changed.
HOW IT WORKS / 5 SECONDS
One answer. Three auditable steps.
The entrance is simple. The evidence machine behind it stays deep, dated and open to inspection.
Collect primary sources
Official releases, filings, market receipts and research enter with a source, time and identity.
Test the claim—not the marketing
AI and fixed rules separate observation from interpretation, seek corroboration and preserve a counterfactual.
Seal a dated answer
We state what changed, why it matters economically, what it means for capital and what would falsify it.
ONE REAL CASE / IBM
A partnership announcement is not yet an investment signal.
This is how one public claim moves through the machine without being promoted into a portfolio decision.
IBM and OpenAI announced a partnership to expand secure enterprise AI deployment.
The announcement is confirmed in IBM’s official newsroom.
The reproducible bridge from AI bookings to recurring software margin and free cash flow per diluted share is not yet admitted.
As of 11 September 2026, IBM stays in deep monitoring at 0% portfolio weight. Evidence is insufficient for a capital position.
PLAIN LANGUAGE / CONTROLLED GLOSSARY
The difficult words have short definitions.
Hover, focus or tap a term. The same controlled meaning is used across the site.
AGI?Artificial General Intelligence: systems able to perform a broad range of economically useful cognitive work across domains, not just one narrow task.alpha?Return above a relevant benchmark that survives costs, risk and forward testing—not simply a rising market.counterfactual?The best available answer to: what would probably have happened without the observed change?evidence receipt?A dated, hashed record proving which source object the system actually observed.forward-only observer?A test that records predictions before outcomes and never rewrites the past with future knowledge.baseline receipt?The frozen starting state used later to measure a real change without hindsight.capital firewall?A governance boundary that prevents research software from changing live weights or placing trades.
START HERE / WHAT IS AGI?
AGI = Artificial General IntelligenceAGI is not a model name or a launch date.
For this fund, AGI means a broadly capable machine system that can autonomously complete economically valuable cognitive work with durable real-world reliability and viable cost. We track the transition through measurable evidence—not a press-release verdict.
Twenty-five is chapter zero. The mandate is global systematic AGI alpha discovery.
Read Manifesto 2.0 ↗25STARTER EQUITIES
21MONITORED SOURCES
06FALSIFIABLE MECHANISMS
00ALPHA CLAIMS
Research status, not a performance claim.
THE COMPOUNDING ASSET
The portfolio is a snapshot. The research system must keep learning.
Our ambition is not a permanent list of AI stocks. It is an auditable operating system that can search every investable market for expectation gaps created by AGI.
01 / causal map◎
Start with change, not a ticker
Capability becomes investable only through adoption, constraints, cash flows and expectations already in price.
02 / Alpha vs Alpha◎
Make markets compete
The US baseline will face China and later global challengers under matched risk, currency, liquidity and cost assumptions.
03 / public memory◎
Publish before knowing
Plans, forecasts, revisions and failures form a forward record. The audience can challenge it; popularity cannot approve capital.
DAILY ANSWER / P06-NET.1
One verified answer about AGI and capital. Every day.
P06-NET freezes the promise before the audience can reward it: material change, honest null result or blocked evidence—always with a source, causal path and decision boundary.
INTERNAL KITCHEN / OPTIONAL AUDITOpen the full P00–P12 research machine and public logRoadmap, receipts, hashes, corrections and every research log remain public. They are folded here so a first-time reader sees the answer before the machinery.
PUBLIC RESEARCH PROGRAM / LIVE · v3.97.0
Watch the laboratory think, build, test and change its mind.
Every phase exposes its objective, current evidence, blocker and next decision. The log records the work in sequence so results cannot quietly replace the plan that preceded them.
13phases
7active now
2evidence/capital gates
00alpha claims
CANONICAL JSON
Program published: · Canonical file auto-checks every 60 seconds
Waiting for first browser checkCharter the laboratory
- objective
- Define where AGI alpha may exist and what would falsify the mandate.
- evidence
- Manifesto 2.0 publishes the causal map, evidence boundaries and capital protocol.
- blocker
- No blocker.
- next
- Keep it versioned when evidence changes the mandate.
Open the operating record
- objective
- Make plans, builds, results, lessons and decisions externally inspectable.
- evidence
- Canonical public program and timestamped research log are rendered on the homepage.
- blocker
- Static releases are near-real-time, not a continuous event stream.
- next
- Add signed entry history and a challenge intake ledger.
Build the source and data foundation
- objective
- Acquire primary filings, policy, physical-system and market data with provenance.
- evidence
- Twenty monitored baseline channels are live. GuruFocus remains a rejected canonical source. The Financial Datasets one-time audit now covers all 92 tracked routes and 87 unique symbols: 43 exact catalog matches, 35 issuer-verified identities, 18 complete private research inputs and 2 standardized-only inputs. The audit used 189 local API attempts with auto-reload and recurring spend disabled.
- blocker
- The measured gaps are material: 44 tracked symbols are absent from the exact-symbol catalog, five collide with the wrong issuer, fourteen have incomplete financial or provenance coverage, six captured snapshots are stale and one provider listing conflicts with the canonical security master. The current plan does not permit public redistribution of underlying vendor values.
- next
- Use Financial Datasets only as a private secondary discovery and normalization layer for identity-verified symbols. Route native non-US listings to primary filings or separately licensed sources, verify every decision-grade field against primary evidence and keep scheduled vendor spend disabled until coverage economics justify it.
Normalize the US baseline
- objective
- Turn 25 US equities into comparable owner-cash, expectations and risk workpapers.
- evidence
- 25 specifications exist; none has passed the normalized valuation approval gate.
- blocker
- Point-in-time shares, segment cash flows and dilution need verification.
- next
- Resolve the first five core-company workpapers with reproducible inputs.
Map China Discovery Universe
- objective
- Separate investable securities, private labs and inaccessible strategic assets.
- evidence
- Universe v0.1 maps 12 public candidates, 3 private labs and 8 pre-investment gates.
- blocker
- Identity mapping is not valuation, ownership verification or approval.
- next
- Verify filings, security access, VIE exposure and economic linkage company by company.
Build China company dossiers
- objective
- Create point-in-time fundamental, governance and valuation workpapers.
- evidence
- P05.21 froze 17 byte-exact artifacts, 7 portable controls, the five-issuer maximum-contrast selection rationale and 6 monitored routes. P05 is complete; unresolved evidence continues under the Evidence Engine.
- blocker
- Any inconclusive cell remains an explicit evidence-acquisition target. A signed FAIL rule identifies an absent quantified criterion in its bound packet; it is not a negative issuer verdict. Expected dilution, executable cross-venue pricing and licensed point-in-time market data also remain unresolved.
- next
- Keep the six carry-over evidence routes monitored without reopening P05; use the frozen handoff as the immutable P06 input.
Run Alpha vs Alpha / Core × Adaptive
- objective
- Test whether governed evidence-to-position adaptation adds net, factor-aware value over the frozen Core under identical capital, timestamps, FX, liquidity and costs.
- evidence
- P06.2 sealed the forward-only observer; P06.3 sealed governed decisions and automatic no-action. P06.4 now freezes the first-future-close pricing receipt, modeled-cost debit and later-observation carry into Adaptive.
- blocker
- Observation zero still awaits the first synchronized eligible close; therefore the forward ledger correctly contains 0 return observations. Licensed execution prices and short-borrow inputs also remain absent.
- next
- Let the nightly workflow reach observation zero, a completed decision window and one approved target set. The pricing engine will then wait for—not backfill—the first synchronized close recorded after approval.
Prove evidence-to-position across Asian supply chains
- objective
- Turn an eligible forward AGI event into a reproducible probabilistic capital consequence before outcome, valuation, weight or price is known.
- evidence
- P07.0–P07.6 freeze the complete forward chain from event and evidence clock through bounded Bayes, cost-aware sizing, independent governance and first-future-close pricing. P07.7 now seals the append-only audit ledger, four entry thresholds and the future calibration, net-return, benchmark, factor and integrity calculations.
- blocker
- The audit is deliberately not eligible: 0/12 governed dockets, 0/4 priced shadow positions, 0/4 resolved outcomes and 0/90 elapsed days. The machine records zero rather than backfilling history.
- next
- Let real P07 dockets and time accumulate while P08 is designed in parallel. Audit only after all four thresholds pass; even then publish a finite process audit, not a durable-alpha or live-capital claim.
Map India and deployment markets
- objective
- Test where abundant machine cognition becomes owner cash flow or displacement risk before market expectations adjust.
- evidence
- P08.2A turns all 6 canonical intake packets into downloadable, route-bound JSON handshakes. Every blank template is demonstrably rejected by the same 10 frozen gates it asks a real provider to pass.
- blocker
- No authentic schema receipt, rights receipt, transfer authorization or matched dataset exists. Therefore workflow adoption, displacement and counterfactual economics remain unmeasured, and P08.3 is locked.
- next
- After a separate founder authorization, send the priority software-delivery template through a secure channel; admit only an authentic schema-only receipt that passes all ten gates, then begin rights review without moving rows or P08.3 early.
Build the Global AGI Equity Observatory
- objective
- Give each admitted listed security a point-in-time AGI causal-impact page spanning direct exposure, displacement and second-order power, land, water, logistics, defense and labor effects.
- evidence
- P09.0–P09.4 engineering is complete/build-ahead: all 50 Global 50 identities sit behind five causal routes, ten executable gates, an append-only Daily AGI Answer screening ledger and a deterministic resolution queue. Model output is proposal-only; code admits; the outcome clock schedules; future evidence resolves.
- blocker
- 0 event-company edges and 0/100 resolved records exist. All four pre-declared calibration rates are null—not zero—and prices, scores, valuation and capital remain closed.
- next
- Let the scheduled Daily AGI Answer screen every post-freeze packet in order. Admit only complete ten-gate records, append future resolutions, and publish latency, precision, revision and no-action rates only after 100 outcomes.
Accumulate a prospective record
- objective
- Resolve timestamped forecasts without hindsight or survivorship edits.
- evidence
- P10.1 keeps the three-source intake live; P10.2 freezes four clocks and an append-only outcome chain; P10.3 precommits three forward audits; P10.4 now freezes the complete governed handoff contract to P11. A CI acceptance test carries 100 explicitly synthetic, memory-only records across the complete P10.1→P10.4 chain and proves that one missing outcome receipt keeps publication and handoff closed. No synthetic fixture enters production state. 0 real records are admitted, 0/3 audit families are published and 0/5 handoff artifacts are sealed.
- blocker
- Before the first real admission, the acceptance test exposed a zero-width evidence window in the P06/P07 adapters. P10.1 contract 1.0.1 now freezes seven calendar days after each due date for a machine receipt; P09 preserves its source-defined expiry. The revision receipt proves that no historical record was changed. The P10.4 handoff remains null until P10.3 publishes the complete eligible forward audit, and no valuation, target weight, rebalance rule, order or trade is created.
- next
- Keep both scheduled workflows advancing P10.1–P10.4 in order. Seal the handoff automatically only after all five P10.4 gates pass; P11 still requires a separate public contract.
Construct governed model portfolios
- objective
- Translate approved evidence into bounded weights, turnover and risk budgets.
- evidence
- Current US allocation remains a research baseline, not demonstrated alpha.
- blocker
- Forward record and approved valuations are insufficient.
- next
- Keep simulation and production mandates explicitly separated.
Authorize regulated execution
- objective
- Only authorized humans and regulated infrastructure may place real-money orders.
- evidence
- Real-money execution, custody and withdrawals are locked.
- blocker
- Legal entity, licensing, broker/custodian and mandate approvals are absent.
- next
- Do not solicit or execute capital until the legal and operational gates pass.
BUILD IN PUBLIC / APPEND-ONLY VIEW
Public research log
A timestamped narrative of what we planned, built, learned and decided.
LOG-151
A paid catalog is not global coverage. All 87 tracked tickers now have a measured data map.
A one-time Financial Datasets audit tested every currently tracked instrument: 92 listing routes and 87 unique symbols across the core AGI portfolio, Global 50, China routes, defense routes and QQQ/SPY benchmarks. The provider catalog contained 43 exact symbols; the identity gate verified 35. Eighteen symbols passed the full private research-input standard, two support standardized-only work and fourteen need field or provenance repair. Forty-four were absent from the exact-symbol catalog, five short symbols resolved to another issuer and four fund/private-market routes were structurally unsuitable for company fundamentals. Of 34 captured financial snapshots, 33 contain twelve quarters and 27 include the required as-reported packet; 27 are current, one lagging and six stale at the cutoff. The most important result is negative but useful: this provider can accelerate an identity-verified US lane, but it cannot serve as our global security master. A successful API response is never enough to join facts to a company. The audit also rejected an SPCX listing conflict and exposed sector-specific schema traps such as bank capex and null-versus-zero debt. 189 local API attempts were recorded; auto-reload, subscriptions and recurring calls remain off. Because the current plan does not grant underlying-data redistribution, the public receipt exposes classifications and ticker cohorts—not licensed values. No canonical evidence, valuation, portfolio weight or trade authority changed.
Inspect the 87-ticker coverage receipt ↗LOG-150
The last AI built by humans has not been built yet.
The 75-page v1 of “The Last AI Built by Humans” does not show that an autonomous takeoff has begun. It gives us something more useful: an evidence ladder for deciding what would count. B0 improves one output without inheritance. L1 executes a human-prescribed change; L2 chooses an improvement strategy; L3 chooses what experience to acquire; L4 persists adaptation from deployment; and L5 revises the improver, verifier or successor-generation mechanism itself. The decisive distinction is structural versus effective recursion: reusing a revised mechanism proves loop closure, while genuine benefit requires better later successors under matched total budgets, protected independent evaluation, held-out transfer and a complete regression history. The survey maps broad L1–L2 evidence, domain-dependent L3–L4 and only bounded L5 prototypes. Its HCI is not “percent to AGI,” and its post-2026 curve is illustrative rather than predictive. We therefore add the RSI Ladder as a standing track: every claim now receives a level, persistent-state receipt, successor re-entry test, verifier-authority record, matched-budget ledger, transfer result, generation count, rollback history, unit-economics measure, owner-economic KPI and falsifier. CDNS/SNPS may sit closest to executable RSI feedback; platform, compute and physical-infrastructure holdings gain only if validated research throughput closes into owner cash faster than efficiency or expectations absorb it. No capability prior, issuer verdict, valuation, weight or trade changed.
Inspect the B0–L5 ladder, proof gates and standing RSI contract ↗LOG-149
AGI is a field of inquiry—not a unit of measurement.
Roberto Pieraccini’s “Four Questions About AGI” correctly exposes the danger of letting one ambiguous banner stand in for intelligence, generality, autonomy, reliable deployment and economic value. Its strongest evidence is not that AGI is meaningless, but that these clocks can diverge: benchmark performance can leap while competence remains jagged; digital capability can scale while physical reliability moves slowly; disruption can precede institutional diffusion; and a laboratory can declare a milestone without independently establishing owner economics. The essay is a secondary synthesis, not a new capability receipt. Its “red herring” conclusion is an argument rather than consensus, and Rodney Brooks’s decade-per-nine reliability rule is a useful heuristic rather than a validated universal law. We therefore keep AGI as the name of our research field but prohibit the label itself from entering the evidence chain. Every future AGI claim must be decomposed into six measurable layers: capability, autonomy, reliability, diffusion, economics and price. For capital, the exploitable object is the gap between impressive capability and dependable economic deployment—and the bottlenecks that close or widen it. This is a methodological clarification only: no capability prior, issuer verdict, valuation, portfolio weight or trade changed.
Inspect the four-question audit and measurement stack ↗LOG-148
AGI is not the finish line. It is the start of a compounding race.
Google DeepMind’s 57-page “From AGI to ASI” gives us a more useful map than a single takeoff date. It defines competent AGI around median skilled-human performance on most cognitive tasks and ASI as intelligence exceeding large, coordinated groups of thousands or tens of thousands of experts across virtually all domains. Six native advantages—speed of I/O and processing, memory, substrate independence, lossless replication and high-bandwidth sharing—make stalling exactly at human level an unstable base case. Progress can run through four non-exclusive paths: scaling compute/models/data, a new algorithmic paradigm, recursive AI R&D and multi-agent organization. Each meets six possible frictions: data, resources/economics, paradigm limits, harder research, abstraction/physical latency and deliberate governance slowdown. The key fund insight is organizational: cheap AGI copies may form a superhuman collective before one model becomes solitary ASI. We therefore replace “when ASI?” with a forward matrix: path × friction × economic transmission × listed owner × price. Epoch AI’s current scaling indicators and METR’s field observations suggest scaling and R&D autonomy are advancing, but physical build times, judgment and reliability still bind. Every future event now needs a path, counter-friction, capability, physical, owner-economics and price receipt plus a horizon and falsifier. Five review triggers are proposed; none creates an issuer verdict, valuation, weight or trade.
Inspect the four paths, six frictions and forward contract ↗LOG-147
10,000 forecasts do not give us one AGI date.
AIMultiple’s “10,000 predictions” is a valuable discovery map, but not one comparable consensus distribution. It combines more than 6,000 participants across ten surveys, 18 named forecasts and roughly 3,900 community/market contributors while mixing AGI, weak AGI, HLMI, full automation of labor, a Turing-test outcome and a company announcement. The definition alone moves the 2023 expert-survey median from 2047 for HLMI to 2116 for full labor automation; live market pages also moved materially beyond the article’s August snapshots by our 15 September cutoff. We therefore reject one weighted calendar year as a portfolio input and split the question into six clocks: capability, reliable autonomy, declaration/contract, deployment/diffusion, cash-flow capture and market pricing. Calendar revisions now have proposed forward rules: a ≥10 percentage-point repricing with a liquidity receipt or a ≥12-month same-contract median shift can open review, but changing a research prior requires two independent capability/autonomy lanes; issuer escalation additionally requires owner economics and a point-in-time valuation gap. The March 2027 forecast remains our pre-registered hypothesis—not a consensus claim. No issuer verdict, valuation, weight or trade was created.
Audit the six AGI forecast clocks ↗PUBLIC RESEARCH / SEASON 01
Watch a fund learn in real time.
Question, source, model, contradiction, decision, result, next chapter. The site and media channels are one open notebook for a shared search—while evidence and capital governance remain non-negotiable.
YOUR NEXT STEP