Autonomous Economic Intelligence
A private research program on reliable decision-making when governing mechanisms are incompletely known and errors are costly.
- Status
- Private architecture · public research boundary
- Evidence
- Environments
- Synthetic or stylized; not prevalence estimates
Capability is not reliable action.
High predictive or reasoning capability does not by itself guarantee reliable action when the decision depends on identifying the governing rule, establishing whether available evidence supports that rule in the region the plan will visit, stress-testing rare regimes, or preserving future options while acquiring information.
- evidence
- governing structure
- support
- stress
- planning
- verified execution
The public research boundary can be summarized as: evidence → governing structure → support → stress → planning → verified execution
Selected internally labeled evidence.
The private research archive includes hidden-rule economies, document-to-decision tests, rare-catastrophe evaluations, covenant-constrained planning, adaptive adversaries, and interactive decision worlds. These experiments are synthetic or stylized and are not presented as real-world prevalence estimates.
handled in one hidden-rule family versus 0/12 for the comparison model
coverage in a code-enabled 24-economy evaluation; comparison systems covered at most 11/24
structural method in controlled crash-rarity tests at 20%, 5%, and 1%
claims inspected in a dedicated claim-discovery audit
Internal evidence. External replication, externally specified worlds, and real-world validation remain decisive next tests.
Standalone research probes
REC, microstructure, and private credit are not simply marketing examples for AEI. They are standalone research probes of different pieces of the broader problem: timing information acquisition, valuing an inquiry as an intervention, and identifying when model error moves a reached feasibility boundary.
Not publicly disclosed
- The production architecture
- Exact identification machinery
- Acquisition / probing policies
- Architecture-selection work
- Implementation details sufficient for reconstruction
- Successor-system design
The decisive next step.
The existing evidence does not establish broad general intelligence, universal superiority over frontier models, or production reliability in open-world finance. The next decisive step is externally specified and independently auditable validation across new world families, model providers, and increasingly realistic economic structure.