Auditable credit guidance as a system design problem
A research note on making credit guidance explicit, logged, and reviewable.
mAI Economy (“monetary AI economy”) is a structured AI and control layer that generalizes historical window guidance into transparent, auditable, model-based credit allocation guidance. The intended audience is central banks, supervisors, and large financial institutions that need to influence credit composition without opaque discretion or black-box AI.
The core claim is not that another macro model is needed. It is that a policy operating system is needed: explicit rules, reviewable state variables, and feedback loops grounded in Kantian methods and Thermo-Credit.
Shortcuts to the underlying theory note, dashboards, and related preprints.
1. Problem: we manage volumes, but not structure
Traditional tools move the level of credit more easily than its structure: sector exposure, maturity, collateral, securitization, and risk concentration. Historical window guidance showed that composition guidance can be powerful, but opaque when rules and records are not explicit.
Today, large models and dashboards exist, but they mostly:
- Predict outcomes instead of defining accountable rules.
- Focus on bank-by-bank supervision, not system-level allocation incentives.
- Rely on complex, unstable AI without clear calibration or audit trails.
mAI Economy addresses this gap: it is a soft, rules-based, explainable “guidance layer” that recommends credit allocation corridors while staying observable, contestable, and regulatorily defensible.
2. Design: “Window Guidance as Code”
The system is designed as a control layer, not a command center. It does three things:
- Maps policy objectives (financial stability, climate, productivity, housing risk) into measurable targets.
- Translates them into allocation guidelines across sectors, tenors, instruments, and securitization channels.
- Monitors outcomes with stability & fairness metrics and provides explanations and logs.
Data & structure layer
Unified schema across:
- Regulatory texts and supervisory expectations (Basel, local rules, guidance).
- Bank disclosures: sectoral lending, RWA breakdown, securitization exposures, liquidity profiles.
- Market data: spreads, term structure, MBS/RMBS volumes, CDS where relevant.
This layer is encoded as a domain graph aligned with our Thermo-credit (QTC) indicators (e.g. capital “temperature”, securitization “channels”, off-balance φ metrics).
Model & control layer
- Uses QTC-style potentials to represent balance-sheet stress and misallocation.
- Applies feedback control logic to test how guidance rules affect stability and risk.
- Integrates AI components only where they are calibrated, monitored, and auditable.
Policy Studio (for authorities)
Define objectives, explore credit and capital corridors, and see projected impacts with documented assumptions.
Bank Console (for institutions)
Each bank can map its own book to the guidance corridors, understand deviations, and receive explainable suggestions instead of opaque scores.
Stability & Fairness Monitor
Tracks whether guidance remains proportionate, non-discriminatory, and within legal mandates. Every scenario and recommendation is logged for ex-post review.
3. Why this is hard to copy (and why we build it)
The value is not a prompt or a dashboard. It is the integration of:
- Theme 1: Kantian / feedback-based reliability research (AuditLoop) for measuring model stability, calibration, and hallucination risk.
- Theme 2: Thermo-credit (QTC) economic theory linking balance-sheet mechanics, securitization, and macro outcomes.
- Custom data & schema: Long-horizon datasets and a domain graph aligned to regulatory, accounting, and securitization regimes.
- Deployment model: Designed to run inside central bank / supervisor / G-SIB environments, with strict logging and review.
The value sits in tested pipelines, metrics, and institutional fit.
4. Governance: what mAI Economy is not
To avoid misuse and legal risk, the concept is framed with explicit constraints:
- It does not issue binding lending orders to any institution.
- It does not replace human or statutory decision-making.
- It does not provide retail investment advice or product recommendations.
- It is not a backdoor for discriminatory or non-transparent quotas.
Instead, it provides:
- Scenario tools with documented assumptions.
- Explainable metrics and corridors that can be debated and revised.
- Audit-ready logs for internal, external, and parliamentary oversight.
5. Status & next steps
mAI Economy is an R&D direction. Current work focuses on:
- Formalizing the QTC framework and publishing transparent theory notes.
- Building reproducible datasets and open demonstration code where feasible.
- Defining a deployable, regulator-friendly architecture for closed pilots.
Parties interested in joint research or tightly scoped PoC discussions (central banks, supervisors, large banks) may contact: info@toppymicros.com