Human judgment, compounded by governed intelligence
Risk Dimensions joins durable evidence, institutional memory, deterministic data services and bounded AI into one analytical system. The goal is not maximum automation. It is better preparation, clearer judgment and reliable execution with authority kept in human hands.
Operating principle: human judgment leads, agents extend
The design draws on Iain McGilchrist's The Master and His Emissary and W. Edwards Deming's System of Profound Knowledge.
Four-layer operating model
Evidence
Market observations, company information, documents, communications and public sources enter with provenance, timestamps and stable identity.
Memory
Accepted findings, contradictions, unresolved questions and superseded conclusions form a durable institutional record.
Judgment
Research frameworks and bounded models compare new evidence with prior understanding and return attributable judgments.
Delivery
Analytical products, research surfaces and operating workflows turn governed intelligence into practical decisions.
Platform foundation
Canonical data
Structured ledgers preserve securities, observations, relationships, revisions and analytical results without erasing source identity.
Deterministic services
Collectors, validators and analytical services normalize data, enforce schemas and expose consistent product interfaces.
Scheduled intelligence
Repeatable workflows refresh evidence, monitor change and produce reviewable outputs on defined schedules.
Governed inference
Models are selected by task and information boundary, with structured outputs, explicit limitations and human review.
Data and knowledge planes
Market and investment data
Security Master, listed price history, rates, macro events, news, outliers, sovereign data and private/bespoke portfolio return streams feed the analytical products.
Business and relationship data
CRM, contacts, Exchange mail, iMessage, WhatsApp, Substack, calendar evidence and meeting material are normalized while preserving provider identity and provenance.
Institutional knowledge
A durable knowledge corpus preserves doctrine, architecture, decisions and reusable context. Bounded context retrieval supplies the smallest task-relevant packet, while governed promotion controls what becomes institutional memory.
Intelligence compounding loop
Architecture library
The estate map is supported by focused schematics. The animated explainers show how intelligence compounds, how frameworks are tested and how attention is protected. The inline diagrams below show memory, data provenance and authority.
The Compounding Engine →
How temporary research becomes verified, durable and selectively retrievable intelligence while contradictory and superseded history remains visible.
The Regime Sentinel →
How events and RSS observations test, harden, challenge, falsify or warn of regime change in enduring frameworks.
The Space Between →
How governed filtering keeps advertising, engagement hooks, irrelevant stories and noise from consuming Mark's attention.
Memory Compounding ↓
Proposal, verification, acceptance, durable state and narrow retrieval across future work.
Data and Provenance ↓
Provider observations, immutable identity, normalized ledgers, APIs and human-facing surfaces.
Human Authority and Execution ↓
Human objectives, governed orchestration, bounded specialists, deterministic controls and exact readback.
Schematic: memory compounding
Schematic: data and provenance
Schematic: human authority and execution
Operating capabilities
Research workspace
Documents, communications and prior work can be reviewed together with source identity, citations and visible limitations.
Relationship intelligence
People, companies, conversations and follow-ups are connected without collapsing distinct source contexts.
Market intelligence
Cross-asset prices, rates, fundamentals, sovereign conditions, events and outliers share canonical identifiers and provenance.
Portfolio analytics
Listed assets and bespoke return streams support comparable simulations, diagnostics and complete-period analysis.
Institutional memory
Prior conclusions, contrary evidence, decisions and reusable context remain available for selective future retrieval.
Governed AI judgment
Research doctrine, model judgment and deterministic conversion remain separate so conclusions are attributable and reviewable.
Products and public surfaces
Global Scorecard
Cross-asset risk dashboard, securities universe, news monitor and events.
Portfolio Analyzer
Listed and bespoke return-series analysis, simulations and portfolio diagnostics.
Sovereign Fiscal Monitor
Debt, deficit, maturity and fiscal-dominance monitoring.
Company Fundamentals
Single-company research, comparisons, valuation and governed Coach context.
Data Resources
Protected aggregate Equity research, including the country-level Equity audit, published without ticker or issuer disclosure.
Kestrel
Official product landing page with supporting one-pagers for the Kestrel stack.
Risk Dimensions
Company website and public identity.
Research and media
The Macro Case for Bitcoin, Substack research and The New Barbarians Podcast.
System Architecture
This public overview at system.riskdimensions.io.
Authority and security boundaries
Least privilege
Dedicated identities, narrow paths, read-only interfaces and exact provider scopes limit every integration.
Human authorization
Consequential actions remain distinct from analysis and require accountable human authorization.
Far-side verification
A successful process is not enough. The resulting state is observed before success is reported.
Untrusted inputs
Documents, webpages and model outputs are treated as evidence, not instructions with independent authority.
Separated authority
Identity, data access, analytical judgment and consequential action are kept as separate control boundaries.
Visible uncertainty
Incomplete evidence, model disagreement, abstention, drift and superseded conclusions are recorded rather than hidden.