Cauce installs inside the financial institution's infrastructure and verifies compliance on every credit decision before it is executed. It does not route applications, does not score the applicant and does not take the decision. Your customers' data never leaves your infrastructure.
Cauce is not the pipeline: it is one stage of it. The others belong to the institution.
The institution obtains the applicant's data with the applicant's consent. Cauce takes no part and obtains no consent.
The institution's own scoring engine. Cauce does not score the applicant and does not replace that engine.
This is where Cauce operates. It checks the decision already reached against rules configured by the institution: due diligence, sanctions and PEP lists, limits and policy, consent. It requires human confirmation where the law demands it, generates the explanation of the decision, and keeps the audit trail.
The institution executes its decision. Cauce takes no part.
Cauce does not compete with the institution's decision engine: it verifies its decisions and keeps the evidence.
Loop: execute → evaluate → modify → verify → retain, always in a sandbox and validated by CASE before production. It never changes live on its own — rollback and human oversight.
Every decision comes with a readable explanation traceable to the source data. Designed with XAI audits in mind.
Map of rules studied per jurisdiction (research): Ecuador, Mexico, Brazil, Colombia, Chile, Peru, Argentina.
Versioned REST/JSON and signed webhooks. The prototype runs on synthetic data; there are no production integrations.
Banks, credit unions and cooperatives, finance companies and supervised credit fintechs. Cauce is not licensed or sold today. Should the project move to a commercial stage, the only contractual customer would be the supervised institution, never the public.
CACE-Bench is an open synthetic benchmark (MIT) for evaluating credit AI. Its reference run regenerates from a single command and the same seed (0): compliance false positives from 22.51% to 4.96% (−77.9%) over 23,000 synthetic cases, with 95% Wilson intervals. In that same run, 15.57% of cases come out undecidable: the provider chain is not enough to conclude. Figures from a synthetic benchmark, illustrative of the method and not of production performance. Validation on production data has not been done.
The same benchmark, second domain: DeFi track →
The CASE compliance engine audits every AI decision and maps its metrics to the region's regulatory frameworks. Its methodology and a reproducible synthetic benchmark are open — CACE-Bench: measure us against our own standard. Encrypted data, consent and data-subject rights.
Reference authorities in the regulatory map; they are named descriptively.
On-premise or in the institution's own private cloud. No transfer of information outside the institution's infrastructure or outside the customer's jurisdiction. The regulatory obligation rests with the institution; Cauce produces the evidence with which it is demonstrated. Cauce is today a research prototype, with no incorporated entity and no commercial offering: we do not claim organizational certifications we do not hold.
Data minimization: the demo persists no personal data and the benchmark runs on synthetic data. We do not resell or monetize data.
Encryption in transit and at rest. Per-tenant isolation. Per-country residency, as an option.
Cauce acts as a processor by design: the controller is the obligated party. Non-delegation clause — scoring, the decision and the report to the FIU stay with the client. We deliver signals and attributions, not a solvency assessment.
Only if the project moves to a commercial stage and an entity is incorporated: SOC 2 Type II, ISO/IEC 27001, 27017 and 27018, ISO/IEC 42001; subprocessor registry, incident SLA and DPA templates. Current status: not audited — we will publish each milestone with its date.
Cauce is a project of Digital Economy Lab focused on regulatory verification prior to the credit decision in Latin America. It is a research prototype: it is not incorporated as a company and is not sold today. We say so plainly, because in RegTech traceability starts with transparency about who is behind it.
The open CACE-Bench benchmark and the reference agent are authored by Ravil Akhtyamov — Head of research project, Digital Economy Lab.
Cauce and Digital Economy Lab are affiliated. To avoid conflict of interest, the lab does not audit Cauce deployments, and this affiliation is disclosed in the benchmark methodology. CACE-Bench evaluates third-party models and agents by the same standard.
[email protected] · Open CACE-Bench repository: github.com/rav11l/cace-bench
Cauce is a research prototype: we do not currently offer pilots or commercial services. Write to us to discuss the methodology or collaborate on research.
Regulatory map for research purposes, updated 22.10.2026. It does not constitute an offer of services in any of these jurisdictions and is not directed at persons located in them. Cauce is neither registered with nor supervised by any of the authorities mentioned; they are named descriptively.
| Country | Regulator · FIU | Verifies before the decision |
|---|---|---|
| Peru | SBS / UIF-Perú | Due diligence and enhanced regime, sanctions and PEP lists (Res. SBS 2660-2015); human oversight and documentation of the logic and of the data sources in high-risk systems (Ley 31814 and DS 115-2025-PCM, in force since 22.01.2026; credit-scoring is high-risk; the adaptation window for the economy and finance sector closed in September 2026). |
| Chile | CMF / UAF | Due diligence, sanctions and PEP lists (Ley 19.913); right to object to decisions based solely on automated processing and, where the exceptions apply, guarantees of human intervention, expression of the data subject's point of view and review (Ley 21.719, art. 8° bis; full effect expected 01.12.2026; a pending bill proposes postponing it to 01.12.2027). |
| Brazil | BACEN / COAF | Periodic model assessment, including backtesting, by a unit other than the one that developed the model, and stress-test documentation (Res. CMN 4.557/2017, arts. 9 and 12; segments S1–S4); right to request review of decisions taken solely on automated processing, not necessarily by a natural person (LGPD art. 20). |
| Mexico | CNBV / UIF | Banks: Disposiciones under art. 115 of the LIC. Non-financial lenders (actividades vulnerables): KYC and a 10-year file; risk-based customer classification with semi-annual reassessment and automated monitoring and alerting mechanisms (Reglas de la LFPIORPI, as amended by Acuerdo 115/2026, DOF 07.08.2026; staged entry into force 30.11.2026 · 01.03.2027 · 01.06.2027). |
| Colombia | SFC · SES / UIAF | Risk-based due diligence and reporting through the UIAF system (SARLAFT 4.0 — Circular Externa 027/2020); open-finance system mandatory for SFC-supervised entities, being implemented (Decreto 0368 of 07.04.2026). |
| Ecuador | SEPS · SB / UAFE | Due diligence, sanctions and PEP lists, consent and limits; right not to be subject to decisions based wholly or partly on automated assessment (LOPDP art. 20); a specific rule on AI and personal data requiring a prior impact assessment and an audit of the system (Resolution SPDP-SPD-2026-0009-R, Registro Oficial 240, 10.03.2026; amended by Resolution SPDP-SPD-2026-0037-R). |
| Argentina | BCRA / UIF | Risk-based due diligence (Res. UIF 14/2023 as amended), sanctions and PEP lists; model inventory, explainability, independent validation and assessment of external AI providers (BCRA supervisory guide on the responsible and safe use of AI, June 2026 — supervisory expectations, not binding rules). |
This is not legal advice and not a certification of compliance: the regulatory obligation rests with the institution, and compliance must be assessed with local counsel.
python examples/benchmark_your_pipeline.py --demo --n 3000
python examples/benchmark_your_pipeline.py --endpoint https://tu.api/verify
See the adapter (Case → Narrative) →