Architecture · technical detail
How the platform is built
How a card issuer turns network files into governed data products, and runs models and AI on them without loosening control. This page is for the engineering, security and model-risk teams.
From network file to governed answer
Each layer answers a question an examiner will ask. The entitlements layer is live in this product: click it.
- Network and ledger filesauthorizations · settlement · disputes
- Bronze → Silver → Goldrefined and contract-checked
- Data productsowned · contracted · catalogued
- Entitlementsprogram scope · purpose · masking · logLive · try it
- Modelsfeature tables · model cards
- AI analystsemantic layer · masked views
One ACID store for raw auth streams and curated products - no warehouse/lake split to reconcile at examiner time.
A disciplined refinement path from raw network files to certified products - replayable, quarantined, contract-validated.
Card domains (spend, rewards, risk, partner economics) own and publish their own products; no central team bottleneck.
Policy is computed and enforced uniformly - row scope, column masking, purpose, lineage, audit - so a regulated bank can prove control.
Decentralized ownership of analytical data by the domain that knows it best.
Each domain serves its data as a discoverable, addressable, trustworthy product.
A platform that lets domains build and run products without bespoke plumbing.
Global policies (security, interop, quality) enforced as code across all products.
How Atlas enforces entitlements
The controls a CISO will ask about, each one enforced by Postgres and exercised by the live access-control screen.
Catalog, contracts, lineage, policies and model cards are public. No fact row, view, aggregate or log row answers to it.
Callable by the server only. Each runs as a role that cannot bypass Row Level Security, so the calling key never decides what is visible.
Purpose grant, program scope and the fraud-detail gate, combined with AND. An ungranted purpose returns nothing for any persona.
Masking and the rows-versus-aggregates grant live in the views, which are the only row-level read surface.
Partner metrics are computed where the data lives; cells under 25 distinct cardholders never leave it.
The model fills a typed spec; the server compiles it onto the views; the database re-checks it and runs it as a role that can read the views and nothing else.
Every read is logged with persona, purpose, outcome and a chain hash. Visitors see only their own events.
Every persona, purpose and data product asserted against the claims above, on the live data, before and after every change.
Every control above is exercised live on the access-control screen, with the policy SQL, a decision trace and the self-test result beside each answer.
Open access controlsMedallion refinement, mapped to card data
The Gold tables are the published data products in the catalog.
Network and ledger events land exactly as received. PAN tokenized at landing.
- Card auths + settlement/clearing (ISO 8583 / ISO 20022)
- Visa / Mastercard network files
- Disputes & chargebacks, rewards events
- PAN tokenized at landing; append-only; replayable
- Ingested via streaming / Auto Loader
Cleaned, deduped, joined to reference data, and validated; bad rows quarantined.
- Joined to merchant / MCC / account / cardholder reference
- Deduped & type-conformed
- Validated with declarative pipeline Expectations (DLT / Spark Declarative Pipelines)
- Non-conforming rows quarantined, not silently dropped
Certified aggregates and ML feature tables. These Gold tables ARE the data products in the catalog.
- Certified business aggregates (spend, approval, fraud, attrition)
- ML feature tables with online/offline parity
- Contract + SLA + entitlement attached per product
- Consumed by BI, ML, partner reporting, and GenAI - never the raw lake
Data products and contracts
A data product is data plus an owner, an SLA, a contract and known consumers. Without a contract it is just a table.
Inception → Design → Creation → Publish → Operate / Govern → Consume → Retirement
Bitol project, Linux Foundation AI & Data - originated at PayPal
Open Data Product Standard (ODPS) v1.0 - the product wrapper around the contract
See the contracts in the catalogEntitlements on the common enterprise stacks
The same controls, expressed natively on the two platforms most issuers run. Atlas's layer sits on top of either.
- ABAC policies
Row-filter and column-mask policies bound to governed tags - attribute-based, not per-table hand-coding.
- Governed tags + classification
Auto-tag PII / financial data, aligned to GDPR / PCI / HIPAA controls.
- Row filters & column masks
Enforced uniformly across SQL, ML, and BI consumers.
- Automatic lineage
Table- and column-level lineage captured automatically for impact analysis and audit.
- System tables
system.access.audit, lineage, and billing tables - examiner evidence by query.
- Masking + external tokenization
Dynamic data masking and external tokenization integrations at column level.
- Row access policies
Row-level filtering bound to roles / context.
- Tag-based masking
Object tags drive masking policies so classification propagates automatically.
- Trust Center
Posture monitoring across the account.
- Classification / dependencies / access history
Data Classification, Object Dependencies, Access History, Data Metric Functions.
Vaultless tokenization + format-preserving encryption (FPE) applied to PAN at Bronze landing. Detokenization is itself an entitlement - granted only to roles and purposes that need clear values. Representative vendor: Protegrity (representative vaultless-tokenization vendor).
What JPMorgan Chase has published about its data mesh
JPMorgan Chase has described publicly a data mesh on AWS in which each data product lives in its own lake and is shared in place rather than copied. Atlas follows the same principle; nothing here claims knowledge of the firm's internal card platform.
Each product and producer has its own storage, so isolation is physical, not a naming convention.
Per-product catalogs roll up to a central catalog, so the firm can see who can read what.
Consumers are granted access where the data lives, down to column level, instead of receiving copies.
Consumers query what they are entitled to; extracts stop multiplying.
Models and AI on curated data
Generative AI consumes only certified products, through the same entitlements as people. It gets no path of its own.
Atlas's AI analyst works this way: questions resolve to a validated spec over a semantic layer and run on masked views.
In Atlas: a site-wide daily model budget, verified questions with no model call, and a database executor that refuses anything outside the masked views.
Model risk under SR 26-2
SR 26-2, issued by the Federal Reserve on 17 April 2026, superseded SR 11-7 with a risk-based framework: controls scale with a model's materiality instead of one validation standard for every model. Atlas tiers its models accordingly and applies the same principles to its AI analyst.
Controls scale with model materiality and risk - not one-size-fits-all.
Risk managed across the full model lifecycle, not at a single validation gate.
Independent, competent challenge remains the backbone of validation.
Ongoing performance, drift, and outcome monitoring in production.
The AI analyst is inventoried, monitored and constrained like a model: a validated output format, a budget and a full access log.
| Requirement | Where it is met |
|---|---|
| Model inventory | Unity Catalog tables / model registry |
| Model cards | MLflow + UC Model Registry, bound to the production version |
| Validation / effective challenge | MLflow Evaluate: AUC / KS / PSI; LLM: groundedness, toxicity, PII-leak, LLM-as-judge |
| Continuous monitoring | Lakehouse Monitoring + Gateway telemetry |
| Promotion controls | UC Registry aliases + ABAC; Tier-1 dual control |
| Examiner RFI | Answered via a governed Genie space over the inventory + validation log |
Reference components
The building blocks by layer, for the engineering conversation.
Sources and method
- 01Data mesh - Zhamak Dehghani's four principles: domain ownership, data as a product, self-serve platform, federated computational governance.
- 02Open Data Contract Standard - ODCS v3 and ODPS v1, Bitol project, Linux Foundation AI & Data.
- 03Databricks Unity Catalog: row filters and column masksdocs.databricks.com/aws/en/data-governance/unity-catalog/filters-and-masks
- 04Snowflake Cortex Analyst: verified query repositorydocs.snowflake.com/en/user-guide/snowflake-cortex/cortex-analyst/verified-query-repository
- 05JPMorgan Chase: the evolution of data mesh architecturewww.jpmorganchase.com/about/technology/blog/evolution-of-data-mesh-architecture
- 06Federal Reserve SR 26-2 - Supersedes SR 11-7 with a risk-based model-risk framework.www.federalreserve.gov/supervisionreg/srletters/SR2602.htm · as of 17 April 2026