Products · Reference implementation

What RJL runs of its papers, and what is live.

Feature by feature and paper by paper: what is live, what is built and what is not there yet.

What runs

Three products on one governed path.

RJL Context Engine

Connects sources, serves approved meaning and governed metrics, and gives AI governed access through an MCP gateway that decides each call before any read. Every call is sealed in an audit trail. Sign-in through Amazon Cognito, by invitation.

engine.rjlsol.com

Workflow Studio

Runs a workflow on approved meaning with a person deciding: steps pinned to a release, cited drafts checked against their sources, and a named reviewer. The guided demo needs no account; the builder does.

rjlsol.com/workflow

Ontology Studio

Drafts and reviews business terms, imports FIBO, gist and MDRM report items as proposals, and publishes versioned releases the engine loads. Guest mode is free.

rjlsol.com/ontology/studio

RJL runs these three products on AWS, on fictional companies and data, as its own implementation of the patterns in its whitepapers. Two use cases run end to end: insurance claims for a fictional insurer, and bank regulatory reporting for a fictional bank. Engine links on this page need an invited account; ask for a walkthrough if you have none.

How to read the statuses

Five labels, used the same way everywhere.

  • Live Tested live at least once. Deployed, with a recorded live check.
  • Beta Built and tested, live check still to come. Some run in warn mode or are partial; the row says so. Features that ship switched off say so in the row.
  • Planned Recorded, not built.
  • Not supported The product does not do this.

Today: 73 features live, 126 in beta (20 of them off until switched on), 2 planned, and 7 patterns the papers describe that the product does not support.

Architecture

How the three products fit together.

RJL architecture: Ontology Studio defines meaning, Workflow Studio runs the work with a person deciding, and the Context Engine serves governed context. A control plane hosted in the engine serves operator roles. Catalogs, DSPM tools, warehouses and AI platforms are integrated, not replaced. Everything runs on AWS.RJL architecture: Ontology Studio defines meaning, Workflow Studio runs the work with a person deciding, and the Context Engine serves governed context. A control plane hosted in the engine serves operator roles. Catalogs, DSPM tools, warehouses and AI platforms are integrated, not replaced. Everything runs on AWS.
The architecture as designed. Not every part is live: one sign-in across all three products is built and off by default, and the catalog and DSPM integrations are built and tested with fakes. The status lists below say which. Select the drawing to open it full size.
From source to governed answer in six steps: sources, connect, context, check, guarantee and use, with one sealed audit trail beneath every step.From source to governed answer in six steps: sources, connect, context, check, guarantee and use, with one sealed audit trail beneath every step.
From a source to a governed answer. Each call is decided at the gateway before any read and sealed in the audit trail. Some checks on this path, such as row-level rules and delegated identity, run in warn mode today. Select the drawing to open it full size.

Feature status

Every feature, by area.

Open an area to see each feature, the product it lives in, any gap the records state, and a See it working link where the feature is live.

Meaning 16 features: 7 live, 9 beta
  • Live Mapping from sources to an approved business model; build refuses until every mapping is approvedContext Engine See it working (invited sign-in)
  • Live Pack meaning served only from an approved Ontology Studio release, cited on answers, with a served release ledger and package fingerprintContext Engine · Fingerprint and ledger came in release 2026-11-18d; the release card is part of the Live CTX-006 row See it working (invited sign-in)
  • Live Industry models: FIBO (MIT) and gist (CC BY 4.0) search and alignment proposals; imported terms keep version and licenseContext Engine See it working (invited sign-in)
  • Live Semantic layer compiled to fixed SQL, versioned apart from the ontology, exported to dbt MetricFlow and Snowflake semantic viewsContext Engine See it working (invited sign-in)
  • Live Purpose-specific measures with owners: ’open claim’ resolves by purpose (service, reserving)Context Engine See it working (invited sign-in)
  • Live Proposed, disputed or expired terms refused at resolution; only approved mappings buildContext Engine See it working (invited sign-in)
  • Live Knowledge graph generated as openCypher with lineage, quality and policy on nodes and edgesContext Engine See it working (invited sign-in)
  • Beta Effective-dated definitions with as-of lookup; known-at historyContext Engine · as_of and known_at tool arguments wait for the GW-023 reviewer sign-off
  • Beta Namespaces, domain roles and a central steward for shared core entitiesContext Engine · Not wired into Ontology Studio approval
  • Beta Every data product attribute linked to an approved ontology attribute, checked at packagingContext Engine · Warn mode in both packs
  • Beta Regulatory rules and edit checks bound to ontology elements and carried with resultsContext Engine · Warn mode
  • Beta AI metadata proposals (descriptions, synonyms, terms) served only after a logged steward approvalContext Engine · Model rewrite provider tested with a fake only
  • Beta Catalog search scored on a 200-query set (0.94 top-5); misses become context gapsContext Engine · Test set written by RJL
  • Beta Graph formats: Gremlin, RDF, SPARQL and SHACL from the ontology release; Neptune Analytics optionContext Engine · Neptune and Neptune Analytics off for cost
  • Beta Conversational BI over governed metrics only: no SQL from the model, cites definition, owner, release and trust signalsContext Engine · On for banking since release 2026-11-21 (owner decision); no feature-level live check recorded
  • Beta Visual design studio: map columns to approved attributes, generate YAML diff, sealed change package approved by someone elseContext Engine · On for banking; no feature-level live check recorded
Data products 22 features: 13 live, 9 beta
  • Live Data product manifest (rjl.data-product.v1) and ODCS-shaped contract checked in every buildContext Engine See it working (invited sign-in)
  • Live Breaking interface change without a major version fails the buildContext Engine See it working (invited sign-in)
  • Live Lifecycle states (sandbox, managed sandbox, certified) shown with every result; promotion gatedContext Engine · Both demo products are managed_sandbox See it working (invited sign-in)
  • Live Certification only through a certification record bound to the manifest digest; build and gateway fail without itContext Engine · No demo product is certified (run kit rehearsal missing, by design) See it working (invited sign-in)
  • Live Run kit release gate opens only after a non-builder rehearsalContext Engine · No rehearsal recorded, so certification stays blocked See it working (invited sign-in)
  • Live Pipeline tests by class (contract, transformation, grain and join, reconciliation) fail publishContext Engine See it working (invited sign-in)
  • Live In-line quarantine with reasons; unknown reference codes held and shown unresolved, never mapped by guessContext Engine See it working (invited sign-in)
  • Live Quality mode per product (warn or allow); definitions stay available when data is blockedContext Engine See it working (invited sign-in)
  • Live Periodic product reviews with a retirement flagContext Engine See it working (invited sign-in)
  • Live Usage, calls, latency and cost per data product on the control towerContext Engine See it working (invited sign-in)
  • Live Values labelled with status and period; preliminary and final never combinedContext Engine · Banking only; claims data_status off See it working (invited sign-in)
  • Live Contracts inherit classification, purposes and prohibited uses; license terms on external dataContext Engine See it working (invited sign-in)
  • Live Master data matching with confidence, review queue and golden recordsContext Engine See it working (invited sign-in)
  • Beta Data product registration and certification workflow: six release checks, named approver per check, signed change packageContext Engine · Needs demo approver accounts
  • Beta Certified product past review or in breach is degraded, tools denied with reason, consumers notifiedContext Engine · Exercised in tests only; notifications off
  • Beta Restatements as new versions with notice; valid and recorded time; bounded backfillContext Engine · Example restatement record on fictional data
  • Beta Warehouse delivery of gold products into Redshift, Snowflake, Databricks and Iceberg with DDL plan and independent approvalContext Engine · Tested with fakes; live check pending
  • Beta Continuous quality monitor (completeness, validity, freshness, reconciliation, anomalies) with owner alertsContext Engine · Schedule created disabled · Off until switched on.
  • Beta Delivery channels beyond MCP: SQL (restricted grammar), REST, CSV, change events, read-only GraphQL, with a channel parity suiteContext Engine · MCP over HTTPS is live; every other channel off in both packs · Off until switched on.
  • Beta Copy or query as a recorded choice: virtual views served under D-16 conditions, materialized fallback with reasonContext Engine · Execution mode recorded with reasons (Beta); virtual serving off · Off until switched on.
  • Beta Golden record sync and governed Kafka change eventsContext Engine · Off until switched on.
  • Beta Data product marketplace with owner-approved, purpose-bound access requestsContext Engine · Off until switched on.
Sources 10 features: 10 beta
  • Beta Connector catalog (48 entries): enterprise connectors (Oracle, SQL Server, Db2, Snowflake, Redshift, Athena, Salesforce, Kafka, SharePoint, ServiceNow, SAP, Teradata, BigQuery, ADLS, GCS)Context Engine · Partial: tested with fakes; live checks need client accounts; core banking platforms declared only
  • Beta Connections configured in the product: request, independent approval, test, profile, activate; signed configuration history; credentials only in Secrets ManagerContext Engine · Deployed in release 2026-11-20; no feature-level live check recorded
  • Beta Profiling and incremental crawl; scheduled crawl of active connectionsContext Engine · Schedule off as shipped
  • Beta Legacy sources: COBOL copybooks, VSAM, Db2 unload, IMS, XML shredding, replayable MQ and IoT filesContext Engine · Fictional files only
  • Beta Change capture from DynamoDB Streams into a canonical store with version history and as-of lookupsContext Engine · Live as part of write-back (DP-006); latency metric Beta
  • Beta Operational demo estate: fictional bank and insurer in their own AWS systems, reached only through approved connectionsContext Engine · Separate stack; owner steps pending
  • Beta SharePoint as a governed document collection: passage acl applied before rankingContext Engine · Tested with a fake
  • Beta Two-way catalog sync (Glue, DataZone, Unity Catalog, Horizon, Collibra, Alation, Atlan, data.world)Context Engine · Fakes only; live check pending
  • Beta Sensitive data classifier measured honestly: held-out precision 0.80, recall 0.70 (below target)Context Engine · Partial: below the 0.95/0.90 targets
  • Beta DSPM integration hooks: findings from eight vendors become steward proposals only; comparison with engine classifyContext Engine · We integrate DSPM; we are not one · Off until switched on.
Gateway and policy 28 features: 14 live, 14 beta
  • Live Governed MCP gateway: policy decision before any read on every callContext Engine See it working (invited sign-in)
  • Live Role, tool, purpose, metric, entity and classification policy with masking; restricted fields never returnedContext Engine See it working (invited sign-in)
  • Live Every tool typed and bounded; free-text SQL refusedContext Engine See it working (invited sign-in)
  • Live Tool registry with pinned digests and a security test pack; unregistered or changed tools refusedContext Engine See it working (invited sign-in)
  • Live No agent holds a harmful tool or outbound channel; agents cannot writeContext Engine · Lambda network egress not restricted See it working (invited sign-in)
  • Live Every tool result screened for planted instructions (backstop)Context Engine See it working (invited sign-in)
  • Live Labels follow content to results, answers and logs; restricted data never reaches an unapproved model routeContext Engine See it working (invited sign-in)
  • Live Policy propagation drift flagged; uncertain propagation denies data calls (fail closed)Context Engine See it working (invited sign-in)
  • Live Sensitivity reconciliation of the built warehouse at packaging; a gap fails packagingContext Engine See it working (invited sign-in)
  • Live Run limits and spend caps per agent with defined fallbacksContext Engine See it working (invited sign-in)
  • Live compile_context: smallest permitted context with provenance and a token budgetContext Engine See it working (invited sign-in)
  • Live Write-back by the signed-in person only: field allowlist, intent audited first, operation IDs, conditional writeContext Engine · Approval references warn; memory reconciliation off See it working (invited sign-in)
  • Live Governed shared agent memory: declared readers, writers, retention, consent, eraseContext Engine · Seeing it needs a memory role account (none invited) See it working (invited sign-in)
  • Live Gateway refuses to start outside listed regionsContext Engine See it working (invited sign-in)
  • Beta Row-level security rulesContext Engine · Warn in both packs; enforce tested
  • Beta Aggregate-only purposes and purpose binding sealed on every agent callContext Engine · Warn by owner choice
  • Beta Delegated token per answer naming person, agent, audience and purposes (RFC 8693 shape); narrowing only agent to agentContext Engine · Warn mode
  • Beta Approval bound to the exact change with expiry and single useContext Engine · Warn mode
  • Beta Rate limits per person (enforced), per agent and product (warn) with HTTP 429Context Engine · Per agent and product limits warn
  • Beta Enforcement points listed with negative tests on seven pathsContext Engine · Overview role filter in warn; control tower not role-separated
  • Beta Policy exported as Cedar and OPA Rego with equivalence testsContext Engine · Export only; the engine decides
  • Beta Policy change workflow: structured edit, diff, replay of recent audit records, evals, independent approval, signed change packageContext Engine · Needs demo usernames
  • Beta MCP at revision 2026-07-28: Streamable HTTP, authorization on every call and prompt fetchContext Engine · Not checked against an external conformance suite
  • Beta Revocation windows stated and tested; revoked person refused within 60 secondsContext Engine · Not measured on the deployed engine
  • Beta Signed context reference as a runtime gate: data calls without a valid context are deniedContext Engine · Off on both live packs until CONTEXT_REF_KEY is set · Off until switched on.
  • Beta Vector retrieval on OpenSearch Serverless, access filtered before rankingContext Engine · Paid option, demo windows only · Off until switched on.
  • Beta Tokenization of restricted identifiers, two-person detokenize, training export gate, consumer noticesContext Engine · Off until switched on.
  • Beta Attestation-gated key and data release by data labelContext Engine · Demo verifier only; no confidential computing runtime · Off until switched on.
Answers 16 features: 8 live, 8 beta
  • Live Governed agent answering questions through the gateway with model routing and per-answer costContext Engine · The primary model provider is unavailable on RJL’s account; answers run on a second provider as a fallback, and route evaluations run in warn mode See it working (invited sign-in)
  • Live Defined incomplete state for missing, stale, conflicting or unresolved evidence; definition fallbackContext Engine See it working (invited sign-in)
  • Live Numbered sources with versions and as_retrieved digests; each sentence marked fact, interpretation or uncitedContext Engine See it working (invited sign-in)
  • Live Metric results carry metric versions; an unreturned reportable number fails evaluationContext Engine See it working (invited sign-in)
  • Live Separate trust signals per answer with a rely flagContext Engine · No live agent eval run recorded, so answers stay limited See it working (invited sign-in)
  • Live Procedure document search using the version in force; passages carry it; failing documents kept outContext Engine See it working (invited sign-in)
  • Live Two end-to-end use cases on fictional data: claims (insurer) and banking regulatory reportingContext Engine · Four demo questions checked after each deploy See it working (invited sign-in)
  • Live Regulatory reporting: MDRM-coded FFIEC 041 and FR Y-9C lines with edit checks; NAIC claim counts with listed gapsContext Engine · Edit-check publications not yet verified (POL-015 Partial) See it working (invited sign-in)
  • Beta Model routing (fast and deep routes); router version and per-route eval check on every recordContext Engine · Route evals warn; Bedrock suspended; fallback serves without a recorded eval run
  • Beta Approved-model register with data terms per route; unapproved route gets no toolsContext Engine · Partial: routes provisional, data terms not verified
  • Beta Governed policy decision step (evaluate_policy): no model, every step cites its clause, recommendation only, regulated decisions at L0Context Engine · Not checked live
  • Beta Six playbooks as use case templates: Calder Ridge covenant and credit review, KYB, AML alert triage, loan file completeness, covenant monitoring, credit committee memoContext Engine · Page presence checked 21 Nov; no functional live check
  • Beta Published-style edit checks for FFIEC 041, FR Y-9C and NAIC, each citing its source with a verification statusContext Engine · Thresholds illustrative
  • Beta Several main agents per pack, each with contract, lifecycle and limitsContext Engine
  • Beta Bedrock AgentCore agents use the governed MCP gateway with on-behalf-of identityContext Engine · Off until switched on.
  • Beta A2A Agent Cards and task admission through the same checksContext Engine · Specification not verified · Off until switched on.
Evaluation 9 features: 3 live, 6 beta
  • Live Context evals (terms, metric golden values, policy, masking) with a critical veto that blocks packaging and deployContext Engine See it working (invited sign-in)
  • Live Negative release suite on every packaging run (routes, revocation, cross-tenant, passages)Context Engine See it working (invited sign-in)
  • Live Every live answer scored deterministically; hourly drift alertsContext Engine · Observes only; never refuses See it working (invited sign-in)
  • Beta Agent eval suite and release gate per agent with thresholds; scorecards per agent and modelContext Engine · Warn; no live agent eval run recorded (Bedrock suspended)
  • Beta Governed eval sets: expert criteria approved before prove, two experts per critical case, blind baseline, held-out checkContext Engine · Fictional experts
  • Beta Red-team library generated from policy (over 1,000 cases per pack); gateway cases run every test runContext Engine · Model-level cases wait for a model
  • Beta Fairness tests per jurisdiction and line of business; a named duty blocks promotionContext Engine · Illustrative; not legal advice
  • Beta Review queue sampling by autonomy level, gold items, reviewer accuracy, override rate, capacity forecastContext Engine · No live reviewers
  • Beta Answer-level metrics: claim-level groundedness by a model grader, token F1Context Engine · Partial; never a release gate · Off until switched on.
Registry and governance 20 features: 6 live, 14 beta
  • Live Versioned agent contract with an L0 to L4 level per action, compared with the deployed composition at packagingContext Engine · Not validated against rjl.agent-product.v1 See it working (invited sign-in)
  • Live Agent lifecycle with gated promotion, stop, degrade and retire; Workflow Studio evidence on promotion recordsContext Engine · Retirement cleanup in dry run See it working (invited sign-in)
  • Live Kill switches per agent, tool, product and role, effective in about 30 secondsContext Engine · Set in DynamoDB; no UI See it working (invited sign-in)
  • Live AI inventory with discovery check and sealed exportContext Engine See it working (invited sign-in)
  • Live Release ID and policy hash on every audit record; a changed composition without a passed evaluation serves no agentContext Engine · Route evals warn See it working (invited sign-in)
  • Live Decision-rights register with independence checks applied by every workflowContext Engine · Fictional people See it working (invited sign-in)
  • Beta Agent registration and change workflow (owner, roles, tools, level per action, limits, routes) with impact reportContext Engine · Partial
  • Beta Tier per use case on one named scale (OCC or CSBS), equal to the audit tierContext Engine · Tiers provisional
  • Beta Prompt registry: approved prompt versions only, promotion needs an eval run and an independent approverContext Engine · 1.0.0 versions carried over with no eval run recorded
  • Beta Three lines: every role carries its line, internal audit read-only, second-line approval for critical items, findings boardContext Engine · Warn; no auditor accounts
  • Beta Exceptions with owner, compensating control, expiry and retest; workbench with independent acceptanceContext Engine · Workbench needs its Cognito groups
  • Beta Release decision records signed by three different people; promotion gateContext Engine · Gate in warn
  • Beta Applicability records for consequential decisions, confirmed by the second lineContext Engine · Counsel’s law list not started
  • Beta Controls mapped to policies, tests, evidence and regulatory register rows; capability records block promotion when emptyContext Engine · Template records
  • Beta Supplier records with exit plans; portability drillContext Engine
  • Beta Examiner and model risk portal: time-limited read-only examiner access, tiering questionnaire, evidence pack by OCC pillarContext Engine · Needs examiner and demo accounts
  • Beta Control plane across Ontology Studio, Workflow Studio and engine deployments: one inventory, approvals inbox, evidence index, impact analysisContext Engine · Studio data are fixtures; studio APIs off
  • Beta Use cases as versioned configuration: draft, review, independent approval, activation by verified build, templatesContext Engine · Deployed; no feature-level live check
  • Beta Role workspaces (CDAO, CFO) as use case configuration; digital workforce tileContext Engine
  • Beta Roles and permissions screen: role matrix with simulate, decision rights, sign-in bindingContext Engine
Audit and operations 35 features: 12 live, 22 beta, 1 planned
  • Live Keyed two-layer hash-chained audit of every call, Object Lock copy, nightly verification; fails closedContext Engine See it working (invited sign-in)
  • Live Audit records name principal, agent, purpose, versions, decision, citations, model route and prompt versionContext Engine See it working (invited sign-in)
  • Live Correlation ID per question; permitted call, denial, failure and human intervention reconstructed from the trailContext Engine See it working (invited sign-in)
  • Live Replay of any audited call; evidence pack replay linksContext Engine · Model-level replay waits for a live model run See it working (invited sign-in)
  • Live Element-level lineage from source column to product, cited by answers; lineage canvasContext Engine See it working (invited sign-in)
  • Live Retention classes per record; legal hold; content withheld above the reader’s clearanceContext Engine · Demo classes not counsel approved See it working (invited sign-in)
  • Live Audit search by person, agent, product, tool, decision and timeContext Engine See it working (invited sign-in)
  • Live Incident records, reporting clocks per regime, postmortemsContext Engine · Demo clocks; not legal advice See it working (invited sign-in)
  • Live Security event feed with eleven detections, alarms and a response playbookContext Engine · No client SIEM connected See it working (invited sign-in)
  • Live Control tower: audit, registry, cost against budget, feedback, change capture, risk view, committee packContext Engine See it working (invited sign-in)
  • Live Reference architecture and capability map: every layer marked working, partial or not yetContext Engine See it working (invited sign-in)
  • Live Per-answer cost and model spend against budget; gateway cost per 1,000 callsContext Engine · List prices, not the bill See it working (invited sign-in)
  • Beta OpenLineage and OpenTelemetry exportsContext Engine · Export files; no collector
  • Beta Evidence pack (JSON and PDF) tagged by OCC pillarsContext Engine
  • Beta Service objectives in consumer terms with error budgets; postmortem when one incident uses over 20 percentContext Engine · Example targets
  • Beta Authority ladder for AI operations agents, enforced and denial-testedContext Engine · No operations agent deployed
  • Beta Restore and verify drills measuring RPO and RTOContext Engine · Not yet run against staging
  • Beta Staging stack with capped load test and smoke check; dev, test and prod configs with promotion checkContext Engine · Load test not yet run
  • Beta Release integrity: post-deploy smoke of four demo questions, deploy preflight, package fingerprintContext Engine · Smoke step skips until the owner updates the deploy role
  • Beta Cost per accepted task; routing by quality barContext Engine · Partial: no accepted tasks recorded
  • Beta Showback by workflow and persona; Cost Explorer and SaaS metering optionsContext Engine · Partial: no chargeback; metering off
  • Beta Portfolio intake with five readiness questions and value-by-readiness quadrant; stop points agreed before resultsContext Engine · Example entries
  • Beta Benefits contracts with fully allocated, incremental and avoidable cost views; freed capacity kept apart from cashContext Engine
  • Beta Adoption records: sponsor, owner, change leader, signed one-page alignment, enablement by role, AI literacy, handover without buildersContext Engine · Templates
  • Beta One feedback event schema with routes, owners and SLAs; week-one measuresContext Engine · FBK-004 Live
  • Beta Signed change packages a person commits (policy changes, agent registration, design studio, certification)Context Engine
  • Beta Impact analysis: a proposed change lists affected agents, products and reportsContext Engine · Partial: studio data fixtures
  • Beta Admin console with sealed quarterly access reviews; SAML and SCIM optionsContext Engine · On by owner decision; SAML and SCIM off
  • Beta Simulation environments from governed products, masked per roleContext Engine · Phase 2 preview
  • Beta Skills registry and internal expert networkContext Engine · Phase 2 preview, off
  • Beta Process mining of the audit trail; proposals scored by context evalsContext Engine · Off until switched on.
  • Beta Pooled SaaS tenancy with tenant negative suiteContext Engine · Off until switched on.
  • Beta Notifications hub (email, Slack, Teams), critical types not mutableContext Engine · Off until switched on.
  • Beta One RJL sign-in: the engine Cognito pool for all three productsContext Engine · Off until switched on.
  • Planned Annotation, labeling and fine-tuning data engineContext Engine · Phase 2 by design; partner by default (P-05)
Workflow Studio 27 features: 9 live, 18 beta
  • Live Versioned document workflows with a pinned, reviewed Ontology Studio releaseWorkflow Studio · Live acceptance 10 Nov 2026 (before the 19 Nov AWS move; AWS deploy smoke green 20 Nov) See it working
  • Live Human review: reviewer inbox, due times, email alert, accept, return or rejectWorkflow Studio · Live acceptance D (10 Nov) and email check (9 Nov) See it working
  • Live Cited drafts with exact-quotation checks against frozen sources; evidence beside findings; reviewer correctionsWorkflow Studio · Live acceptance C: 4 of 4 checks See it working
  • Live Saved test cases and a release gate that refuses until tests passWorkflow Studio · Live acceptance E gate See it working
  • Live Same test cases on two providers, side by sideWorkflow Studio · Live acceptance F See it working
  • Live Personal and patient data check before any AI callWorkflow Studio · Protection passed live; the message fix (11 Nov) not re-verified See it working
  • Live Background steps that continue with the page closedWorkflow Studio · Live acceptance C See it working
  • Live Healthcare payer prior authorization and provider denial appeal templatesWorkflow Studio · Live acceptance E See it working
  • Live Workspace provider keys sealed server side; 13 providersWorkflow Studio · Live acceptance B See it working
  • Beta Scorecard: cases reviewed, acceptance, correction and rework rates, tokens per accepted case; CSV exportWorkflow Studio · Candidate link /workflow/builder
  • Beta Cost per accepted task against an owner’s acceptance ruleWorkflow Studio
  • Beta Experts as evaluators, criteria approved before drafting, calibrated AI judge with chance-adjusted agreementWorkflow Studio · Candidate link /workflow/builder/guide/experts
  • Beta Critical-failure veto on the release gate: averages never offset a critical failureWorkflow Studio · coverage.md EVAL-016 gap row is out of date on this
  • Beta Reviewer oversight: decisions, corrections and override rate per reviewerWorkflow Studio
  • Beta Signed evidence export (Ed25519), signed review packet and a browser verification pageWorkflow Studio · Candidate link /workflow/evidence/verify
  • Beta Security log export for workspace administratorsWorkflow Studio
  • Beta BPMN modeler and import with ontology term links, release impact on the process, process handbookWorkflow Studio · BPMN is descriptive (no process engine)
  • Beta Three lines: line roster, second-line acceptance of critical exceptions, read-only audit viewWorkflow Studio
  • Beta Send to engine: signed workflow evidence package for engine promotion recordsWorkflow Studio · Download built; network send off
  • Beta Governed document extraction to a schema bound to approved termsWorkflow Studio · Not checked with a live key
  • Beta Model documentation pack for an approved versionWorkflow Studio
  • Beta Platforms page with a production readiness pack per platformWorkflow Studio · Sign-in required; candidate link /workflow/builder/platforms
  • Beta Lending and financial crime playbooks sharing ids with the engineWorkflow Studio
  • Beta Operations view: review queue ageing, runs needing attention, token totalsWorkflow Studio
  • Beta Engine tool steps and policy decision steps calling governed engine tools for the signed-in personWorkflow Studio · Off until switched on.
  • Beta Control plane token and read-only API for the engineWorkflow Studio · Off until switched on.
  • Beta Pro interface: drag and drop steps, board inbox, command paletteWorkflow Studio · Off until switched on.
Ontology Studio 18 features: 1 live, 16 beta, 1 planned
  • Live Topic-first drafting with the user’s own key; drafts arrive as labelled proposals that never self-approveOntology Studio · Live acceptance I (10 Nov) See it working
  • Beta Domain contract: purpose, scope, questions, owner, jurisdiction, effective date stamped on releasesOntology Studio · Candidate link #scope
  • Beta Private local document collections with cited proposals and source rechecksOntology Studio · No enterprise connector or web search provider
  • Beta Terms with stable IDs, typed relationships and process term kinds (role, activity, event, decision)Ontology Studio · Candidate link #define
  • Beta Local review queue with user-entered reviewer names; approval record CSV per releaseOntology Studio · No authenticated identity or role enforcement
  • Beta Scoped meanings: several definitions per term, each tied to a context and effective dateOntology Studio
  • Beta Structural checks and knowledge health; SHACL shape exportOntology Studio · No SHACL engine or OWL reasoner
  • Beta Use with AI: portable context files, approved-only and private-excluded by defaultOntology Studio · Candidate link #use
  • Beta Standard exports: Turtle, JSON-LD, SKOS, SHACL, Apache Ossie; SKOS import as proposalsOntology Studio · No arbitrary OWL import
  • Beta Immutable releases with semantic versions by kind of change, change list and fingerprintOntology Studio · Releases not signed
  • Beta Engine release package (approved terms only) that the engine loads as a reviewed change; lineage import from the engineOntology Studio · Release API off by default
  • Beta Industry models browser: FIBO, gist and MDRM report items, imported only as proposalsOntology Studio · Candidate link #models
  • Beta Semantic layer references checked against dbt, Snowflake or Databricks definition filesOntology Studio
  • Beta Manual compare-answers screen: with and without the context packOntology Studio · No measured LLM improvement
  • Beta Invitation-only cloud save, conflict checks and expiring read-only share linksOntology Studio · Pending hosted acceptance
  • Beta Claims engine example ontology (rjl.claims) that the engine serves as an approved releaseOntology Studio · Candidate link #examples
  • Beta Send a release to Workflow StudioOntology Studio
  • Planned Question routing to evidence and interface; authorization outside the prompt; GitHub and catalog publishingOntology Studio

Papers

Each paper, and the features it relies on.

304 claims in the whitepapers and the book were checked against the product records. Each panel shows up to six claims the product backs, and the most important claim it does not back yet.

Why Enterprise AI Stalls After the Demo 12 claims · Live 6 · Beta 3 · n/a 3

Why Enterprise AI Stalls After the Demo

12 claims checked · Live 6 · Beta 3 · n/a 3

  • Live Information needs agreed meaning and provenance.Section 02 Information · Mapping from sources to an approved business model See it working (invited sign-in)
  • Live None of them decides what a result means or when a source applies. That takes a business authoritySection 02 Information · Purpose-specific measures with owners: ’open claim’ resolves by purpose (service, reserving) See it working (invited sign-in)
  • Live Authority must survive the complete request path.Section 03 Authority · Governed MCP gateway: policy decision before any read on every call See it working (invited sign-in)
  • Live The capability needs an operating owner.Section 04 Ownership · Decision-rights register with independence checks applied by every workflow See it working (invited sign-in)
  • Live Start with one complete governed workflow.Section 06 What to do · Two end-to-end use cases on fictional data: claims (insurer) and banking regulatory reporting See it working (invited sign-in)
  • Beta The controls have to hold everywhere information is read or an action happens, including downstream services and cachesSection 03 Authority · Enforcement points listed with negative tests on seven paths

Honest gap. Delegated identity is recorded on every answer; enforcing it is still in warn mode.

Read the paper

100 AI Use Cases and Who Runs Them 10 claims · Live 3 · Beta 3 · n/a 4

100 AI Use Cases and Who Runs Them

10 claims checked · Live 3 · Beta 3 · n/a 4

  • Live Insurers have moved AI furthest into claims and underwriting intakeSection 04 Insurance · Two end-to-end use cases on fictional data: claims (insurer) and banking regulatory reporting See it working (invited sign-in)
  • Live Health plans use AI to speed approvals, find improper payments and support members.Section 06 Healthcare payers · Healthcare payer prior authorization and provider denial appeal templates See it working
  • Beta Seven patterns account for almost all of the hundred.Section 02 Patterns · Six playbooks as use case templates: Calder Ridge covenant and credit review, KYB, AML alert triage, loan file...
  • Beta Start where the pattern is proven and your readiness is high.Section 03 Choosing · Portfolio intake with five readiness questions and value-by-readiness quadrant

Honest gap. An industry catalogue: the closest RJL examples are Workflow Studio’s healthcare templates and the engine’s lending and anti-money-laundering playbooks, which are built but not live checked.

Read the paper

Agent Patterns 12 claims · Live 4 · Beta 5 · n/a 3

Agent Patterns

12 claims checked · Live 4 · Beta 5 · n/a 3

  • Live Choose the highest row that does the job.Section 03 Control patterns · Versioned document workflows with a pinned, reviewed Ontology Studio release See it working
  • Live Prefer an API whenever one exists.Section 05 Environment · Every tool typed and bounded See it working (invited sign-in)
  • Live Autonomy is a setting, not a type of agent... read freely, draft at L0, write only at L2.Section 06 Autonomy · Versioned agent contract with an L0 to L4 level per action, compared with the deployed composition at packaging See it working (invited sign-in)
  • Beta Describe any agent with one choice from each dimension.Section 02 Five dimensions · Agent registration and change workflow (owner, roles, tools, level per action, limits, routes) with impact report
  • Beta Add agents only when one agent cannot hold the context or the work splits cleanly.Section 04 Topology · Several main agents per pack, each with contract, lifecycle and limits
  • Beta On 4 November 2026 the OCC issued Bulletin 2026-54 and a supervisory frameworkSection 07 Regulators and buyers · Examiner and model risk portal: time-limited read-only examiner access, tiering questionnaire, evidence pack by OCC...

Honest gap. Three lines of defense and agent-to-agent delegation are built, but run in warn mode or are off.

Read the paper

Choosing an Enterprise Platform for AI Agents 13 claims · Live 3 · Beta 6 · Not supported 1 · n/a 3

Choosing an Enterprise Platform for AI Agents

13 claims checked · Live 3 · Beta 6 · Not supported 1 · n/a 3

  • Live Run your own evaluation; treat leaderboards as a shortlist.Section 07 How to pick a model · Same test cases on two providers, side by side See it working
  • Live Which components are GA, preview, beta, early access, or restricted by region and edition?Section 23 Diligence · Reference architecture and capability map: every layer marked working, partial or not yet See it working (invited sign-in)
  • Beta Choose models by task class.Section 05 Models · Model routing (fast and deep routes)
  • Beta The same model can arrive through four different contracts.Section 09 Model access · Approved-model register with data terms per route
  • Beta Can the platform represent the user, the agent and the service separately, and enforce delegated authority with short-lived credentialsSection 18 Gates · Delegated token per answer naming person, agent, audience and purposes (RFC 8693 shape)
  • Beta Price a completed outcome.Section 22 Economics · Cost per accepted task against an owner’s acceptance rule

Honest gap. Model routing and the model register are built. The primary model provider is unavailable on RJL’s account, so answers run on a second provider and route evaluations run in warn mode.

Read the paper

The Connected Enterprise Data & AI Blueprint 13 claims · Live 10 · n/a 3

The Connected Enterprise Data & AI Blueprint

13 claims checked · Live 10 · n/a 3

  • Live Choose a workflow with a measurable outcome.Section 02 The first workflow · Two end-to-end use cases on fictional data: claims (insurer) and banking regulatory reporting See it working (invited sign-in)
  • Live the architecture’s layers group into five capability areas with distinct responsibilitiesSection 03 The foundation · Reference architecture and capability map: every layer marked working, partial or not yet See it working (invited sign-in)
  • Live A table becomes a product when someone keeps those commitments.Section 04 Capability area one · Data product manifest (rjl.data-product.v1) and ODCS-shaped contract checked in every build See it working (invited sign-in)
  • Live Separate calculations from relationshipsSection 05 Capability area two · Semantic layer compiled to fixed SQL, versioned apart from the ontology, exported to dbt MetricFlow and Snowflake... See it working (invited sign-in)
  • Live keep the document’s authority, scope, freshness and permissions intact as it gets parsed and indexedSection 06 Capability area three · Procedure document search using the version in force See it working (invited sign-in)
  • Live Identify both the caller and the agentSection 07 Capability area four · Audit records name principal, agent, purpose, versions, decision, citations, model route and prompt version See it working (invited sign-in)

Honest gap. DSPM appears only as integration hooks, which are off by default.

Read the paper

Adoption by Design 10 claims · Beta 5 · n/a 5

Adoption by Design

10 claims checked · Beta 5 · n/a 5

  • Beta Name the person who changes the work as well as the person who funds it.Section 02 Ownership · Adoption records: sponsor, owner, change leader, signed one-page alignment, enablement by role, AI literacy, handover...
  • Beta Experts will judge the system against their own work.Section 04 Pattern two · Experts as evaluators, criteria approved before drafting, calibrated AI judge with chance-adjusted agreement
  • Beta Change the process, the roles and the incentives along with the tool.Section 07 The work · BPMN modeler and import with ontology term links, release impact on the process, process handbook

Honest gap. Adoption records are templates; the product backs this paper through the review flow, not a live adoption program.

Read the paper

The 90-Day First Governed Capability 8 claims · Live 5 · Beta 2 · n/a 1

The 90-Day First Governed Capability

8 claims checked · Live 5 · Beta 2 · n/a 1

  • Live Prove a complete path through it with real information and real controls, then turn the approved slice into a service your own team can run.Section 01 The objective · Two end-to-end use cases on fictional data: claims (insurer) and banking regulatory reporting See it working (invited sign-in)
  • Live Freeze the candidate artifacts for the review.Section 05 The Prove decision · Release ID and policy hash on every audit record See it working (invited sign-in)
  • Live Build turns the accepted slice into a maintained service.Section 06 Build · Run kit release gate opens only after a non-builder rehearsal See it working (invited sign-in)
  • Live Assign decision, delivery and service owners during Frame.Section 07 Ownership · Decision-rights register with independence checks applied by every workflow See it working (invited sign-in)
  • Live Measure useful work, and enforce release conditions.Section 08 Measurement · Saved test cases and a release gate that refuses until tests pass See it working
  • Beta The weekly plan and its decision gates.Section 02 The plan · Portfolio intake with five readiness questions and value-by-readiness quadrant

Honest gap. The gates and evidence exist, mostly built rather than live checked.

Read the paper

The Data and AI Operating Model 11 claims · Live 6 · Beta 3 · n/a 2

The Data and AI Operating Model

11 claims checked · Live 6 · Beta 3 · n/a 2

  • Live An insurer brings four policy-in-force counts to a leadership meeting.Section 01 The executive case · Purpose-specific measures with owners: ’open claim’ resolves by purpose (service, reserving) See it working (invited sign-in)
  • Live Give meaning a named lead.Section 05 Domain delivery · Decision-rights register with independence checks applied by every workflow See it working (invited sign-in)
  • Live Enterprise oversight should be able to see the outcome, exposure and control evidenceSection 07 Governance and risk · Control tower: audit, registry, cost against budget, feedback, change capture, risk view, committee pack See it working (invited sign-in)
  • Live It retrieves permitted case information and current procedure guidance, cites the sources and prepares a summary. It doesn’t approve paymentsSection 11 Illustrative application · Two end-to-end use cases on fictional data: claims (insurer) and banking regulatory reporting See it working (invited sign-in)
  • Live Fund the supported service, not just its launch.Section 13 Product review · Periodic product reviews with a retirement flag See it working (invited sign-in)
  • Beta Catalog, definitions, owners and current productSection 04 Platform as a service · Catalog search scored on a 200-query set (0.94 top-5)

Honest gap. Three lines and internal audit are built in warn mode, with no auditor accounts yet.

Read the paper

Data Products in Practice 13 claims · Live 9 · Beta 2 · n/a 2

Data Products in Practice

13 claims checked · Live 9 · Beta 2 · n/a 2

  • Live A product is a promise to its consumers.Section 01 The executive case · Data product manifest (rjl.data-product.v1) and ODCS-shaped contract checked in every build See it working (invited sign-in)
  • Live Could a named consumer rely on itSection 02 Product, not project · Lifecycle states (sandbox, managed sandbox, certified) shown with every result See it working (invited sign-in)
  • Live Change the data, take on the contract.Section 04 Producers and consumers · Contracts inherit classification, purposes and prohibited uses See it working (invited sign-in)
  • Live State when the answer is fit to use.Section 07 Freshness and finality · Values labelled with status and period See it working (invited sign-in)
  • Live Choose the shared responsibility deliberately.Section 08 Master and reference data · Master data matching with confidence, review queue and golden records See it working (invited sign-in)
  • Live Test the data needed for the decision.Section 10 Quality and reconciliation · Pipeline tests by class (contract, transformation, grain and join, reconciliation) fail publish See it working (invited sign-in)

Honest gap. No demo product is certified yet, and delivery channels other than MCP are off.

Read the paper

The Federated Data Platform 10 claims · Live 5 · Beta 5

The Federated Data Platform

10 claims checked · Live 5 · Beta 5

  • Live Give shared measures an owner and a purpose.Section 05 Semantics · Purpose-specific measures with owners: ’open claim’ resolves by purpose (service, reserving) See it working (invited sign-in)
  • Live It doesn’t silently reinterpret the incoming value.Section 07 Control plane in action · In-line quarantine with reasons See it working (invited sign-in)
  • Live Connect tools to service responsibilities.Section 10 Capability map · Reference architecture and capability map: every layer marked working, partial or not yet See it working (invited sign-in)
  • Live Give AI a governed runtime.Section 11 AI services · Governed MCP gateway: policy decision before any read on every call See it working (invited sign-in)
  • Beta Build reuse into the next use case.Section 01 The executive case · Use cases as versioned configuration: draft, review, independent approval, activation by verified build, templates
  • Beta Domains publish supported products; shared infrastructure and a common control plane make their operation coherent.Section 02 Reference architecture · Control plane across Ontology Studio, Workflow Studio and engine deployments: one inventory, approvals inbox, evidence...

Honest gap. Warehouse delivery, virtual serving and the connectors are built and tested with fakes; none is live checked.

Read the paper

Context Before Query 11 claims · Live 7 · Beta 4

Context Before Query

11 claims checked · Live 7 · Beta 4

  • Live An authorized answer can still be wrong.Section 01 The executive case · Purpose-specific measures with owners: ’open claim’ resolves by purpose (service, reserving) See it working (invited sign-in)
  • Live obtain final June and September valuation versionsSection 04 Worked request · Values labelled with status and period See it working (invited sign-in)
  • Live Give every agent the same memory.Section 06 Enterprise memory · Governed shared agent memory: declared readers, writers, retention, consent, erase See it working (invited sign-in)
  • Live Keep document rights alive through retrieval.Section 07 Knowledge content · Procedure document search using the version in force See it working (invited sign-in)
  • Live Decide how the system stops or degrades.Section 08 Failure behavior · Defined incomplete state for missing, stale, conflicting or unresolved evidence See it working (invited sign-in)
  • Live Give every control an operator.Section 09 Ownership and evidence · Decision-rights register with independence checks applied by every workflow See it working (invited sign-in)

Honest gap. The signed context gate is built and is switched on per use case; the demo runs with it off.

Read the paper

Ontology or Semantic Layer? 13 claims · Live 6 · Beta 5 · n/a 2

Ontology or Semantic Layer?

13 claims checked · Live 6 · Beta 5 · n/a 2

  • Live A semantic layer says how to count. An ontology says what is being counted.Section 01 The answer first · Semantic layer compiled to fixed SQL, versioned apart from the ontology, exported to dbt MetricFlow and Snowflake... See it working (invited sign-in)
  • Live someone accountable approves each definition, drafts stay drafts, and every answer can be traced to the version it usedSection 08 Ownership · Proposed, disputed or expired terms refused at resolution See it working (invited sign-in)
  • Live Import terms from either as proposals, never as approved meaning.Section 08 Ownership · Industry models: FIBO (MIT) and gist (CC BY 4.0) search and alignment proposals See it working (invited sign-in)
  • Beta Ontology Studio lets you build the ontology for the same claim and export it.Section 03 One claim, three questions · Claims engine example ontology (rjl.claims) that the engine serves as an approved release
  • Beta Ontology Studio exportSection 04 Example products · Standard exports: Turtle, JSON-LD, SKOS, SHACL, Apache Ossie
  • Beta Definitions are portable | Export with owner, version and approval historySection 08 Ownership · Immutable releases with semantic versions by kind of change, change list and fingerprint

Honest gap. Ontology Studio records a reviewer name the user enters, not an authenticated approver.

Read the paper

The Four-Part System 11 claims · Live 9 · Beta 2

The Four-Part System

11 claims checked · Live 9 · Beta 2

  • Live Four responsibilities, delivered as one capability.Section 01 The premise · Reference architecture and capability map: every layer marked working, partial or not yet See it working (invited sign-in)
  • Live When a business decision is missing, it should say so.Section 03 The brain · Defined incomplete state for missing, stale, conflicting or unresolved evidence See it working (invited sign-in)
  • Live A knowledge corpus needs equivalent commitments for authority, applicability, permissions and upkeep.Section 04 Information · Data product manifest (rjl.data-product.v1) and ODCS-shaped contract checked in every build See it working (invited sign-in)
  • Live It carries identity and request context, invokes bounded operations and returns observable outcomes.Section 05 The nervous system · Governed MCP gateway: policy decision before any read on every call See it working (invited sign-in)
  • Live Rapid provisioning helps when the resulting environment also has ownership, bounded access and a path to support.Section 06 The body · Lifecycle states (sandbox, managed sandbox, certified) shown with every result See it working (invited sign-in)
  • Live Classifying data doesn’t, by itself, enforce any access decision.Section 07 The immune system · Role, tool, purpose, metric, entity and classification policy with masking See it working (invited sign-in)

Honest gap. The control plane across products is built with fixtures, and the three-lines crossing runs in warn mode.

Read the paper

The Reference Implementation 34 claims · Live 22 · Beta 9 · Not supported 2 · n/a 1

The Reference Implementation

34 claims checked · Live 22 · Beta 9 · Not supported 2 · n/a 1

  • Live A machine draft carries no authority on its own.Section 05 Meaning and knowledge · Topic-first drafting with the user’s own key See it working
  • Live filter each query by the entitlements in the verified token before anything reaches the modelSection 05 Meaning and knowledge · Procedure document search using the version in force See it working (invited sign-in)
  • Live Record lineage... so an answer can be tied to the exact source version it used.Section 05 Meaning and knowledge · Element-level lineage from source column to product, cited by answers See it working (invited sign-in)
  • Live Agents and applications reach data through narrow, versioned operations, with no open query access.Section 06 Interfaces · Every tool typed and bounded See it working (invited sign-in)
  • Live Keep four separate contracts.Section 06 Interfaces · Data product manifest (rjl.data-product.v1) and ODCS-shaped contract checked in every build See it working (invited sign-in)
  • Live Writes take a separate, certified path... an approval bound to the exact proposal with an expiry. Each intended write gets one operation IDSection 06 Interfaces · Write-back by the signed-in person only: field allowlist, intent audited first, operation IDs, conditional write See it working (invited sign-in)

Honest gap. Enforcement inside a client’s data platform is not supported yet: the gateway enforces over the store it fronts, and row-level rules run in warn mode.

Read the paper

Patterns and Product Fit 10 claims · Live 6 · Beta 3 · n/a 1

Patterns and Product Fit

10 claims checked · Live 6 · Beta 3 · n/a 1

  • Live One vocabulary, many enforcement points.Section 02 Classification · Labels follow content to results, answers and logs See it working (invited sign-in)
  • Live verify tag propagation, policy arguments and effective enforcementSection 03 Enforcement at the data · Policy propagation drift flagged See it working (invited sign-in)
  • Live Filter before the model sees anything.Section 04 Retrieval for AI · Procedure document search using the version in force See it working (invited sign-in)
  • Live One definition, every tool, every platform.Section 08 Semantic layers at scale · Semantic layer compiled to fixed SQL, versioned apart from the ontology, exported to dbt MetricFlow and Snowflake... See it working (invited sign-in)
  • Live Prove the negative before the positive.Section 12 Release tests · Negative release suite on every packaging run (routes, revocation, cross-tenant, passages) See it working (invited sign-in)
  • Beta Fix oversharing before you index it.Section 05 SharePoint as a source · SharePoint as a governed document collection: passage acl applied before ranking

Honest gap. Vector retrieval that filters before ranking is a paid option, off by default.

Read the paper

Agent Onboarding 12 claims · Live 9 · Beta 2 · n/a 1

Agent Onboarding

12 claims checked · Live 9 · Beta 2 · n/a 1

  • Live Onboard the capability, not the connection.Section 01 The premise · Versioned agent contract with an L0 to L4 level per action, compared with the deployed composition at packaging See it working (invited sign-in)
  • Live Bound memory as a separate dependency.Section 03 Memory · Governed shared agent memory: declared readers, writers, retention, consent, erase See it working (invited sign-in)
  • Live Authenticate the caller and identify the agent release.Section 04 The request path · Release ID and policy hash on every audit record See it working (invited sign-in)
  • Live An aggregate accuracy score can’t excuse an unresolved failure in disclosure, isolation, unauthorized action, integrity or evidence.Section 05 Evaluation · Context evals (terms, metric golden values, policy, masking) with a critical veto that blocks packaging and deploy See it working (invited sign-in)
  • Live Promote with inspectable release evidence.Section 06 Release · Agent lifecycle with gated promotion, stop, degrade and retire See it working (invited sign-in)
  • Live needs the authority to restrict itSection 07 Operation · Kill switches per agent, tool, product and role, effective in about 30 seconds See it working (invited sign-in)

Honest gap. The engine’s agent contract is not yet validated against the rjl.agent-product.v1 schema.

Read the paper

Confidential Computing for AI 7 claims · Live 1 · Beta 1 · Not supported 1 · n/a 4

Confidential Computing for AI

7 claims checked · Live 1 · Beta 1 · Not supported 1 · n/a 4

  • Live Labels decide where a workload runs. Attestation proves it ran there.Section 12 Where it fits · Labels follow content to results, answers and logs See it working (invited sign-in)

Honest gap. RJL has an attestation-gated release preview with a demo verifier only; there is no confidential computing runtime.

Read the paper

Data Product Certification Standard 7 claims · Live 4 · Beta 2 · n/a 1

Data Product Certification Standard

7 claims checked · Live 4 · Beta 2 · n/a 1

  • Live Certification records a decision about a version and a scope.Section 01 What certification means · Certification only through a certification record bound to the manifest digest See it working (invited sign-in)
  • Live schema_version: rjl.data-product.v1Section 02 The manifest · Data product manifest (rjl.data-product.v1) and ODCS-shaped contract checked in every build See it working (invited sign-in)
  • Live The service team owns alerts, incidents, recovery and access operations.Section 05 Continuing ownership · Incident records, reporting clocks per regime, postmortems See it working (invited sign-in)
  • Beta Run contract, policy, quality, lineage and service checks alongside a review of meaning and intended use.Section 04 Promotion · Data product registration and certification workflow: six release checks, named approver per check, signed change...

Honest gap. No demo product is certified, because the run kit rehearsal has not happened.

Read the paper

AI Governance Evidence 14 claims · Live 4 · Beta 10

AI Governance Evidence

14 claims checked · Live 4 · Beta 10

  • Live Carry a correlation identifier through the application, agent, retrieval, data product and tools.Section 06 Reconstruction · Correlation ID per question See it working (invited sign-in)
  • Live Keep explanatory evidence without building an uncontrolled store of sensitive information.Section 06 Reconstruction · Retention classes per record See it working (invited sign-in)
  • Live Link incidents, complaints, overrides, source changes and control exceptions to the same system identifier.Section 11 Operation · Incident records, reporting clocks per regime, postmortems See it working (invited sign-in)
  • Live Article 12 requires appropriate automatic logging capabilities for high-risk systems.Section 06 Reconstruction · Keyed two-layer hash-chained audit of every call, Object Lock copy, nightly verification See it working (invited sign-in)
  • Beta An enterprise should be able to show who approved an AI capability, what that approval covers and whether its controls workSection 01 The premise · Release decision records signed by three different people
  • Beta Supersedes SR 11-7 and SR 21-8.Section 02 US banks · Applicability records for consequential decisions, confirmed by the second line

Honest gap. Release decisions, three lines, fairness tests and applicability records are built in warn mode with template records.

Read the paper

The 10-Question AI Production-Readiness Diagnostic 11 claims · Live 9 · Beta 1 · n/a 1

The 10-Question AI Production-Readiness Diagnostic

11 claims checked · Live 9 · Beta 1 · n/a 1

  • Live Is there an owner and a measurable outcome?Section Q01 · Decision-rights register with independence checks applied by every workflow See it working (invited sign-in)
  • Live Are the information dependencies certified?Section Q02 · Certification only through a certification record bound to the manifest digest See it working (invited sign-in)
  • Live Is business meaning approved and tested?Section Q03 · Pack meaning served only from an approved Ontology Studio release, cited on answers, with a served release ledger and... See it working (invited sign-in)
  • Live Does the knowledge path preserve source controls?Section Q04 · Procedure document search using the version in force See it working (invited sign-in)
  • Live Is identity enforced through the complete path?Section Q05 · Negative release suite on every packaging run (routes, revocation, cross-tenant, passages) See it working (invited sign-in)
  • Live Is the agent limited to approved capabilities?Section Q06 · No agent holds a harmful tool or outbound channel See it working (invited sign-in)

Honest gap. On today’s records the engine would answer three of the ten questions Yes and seven Partial.

Read the paper

AI Token Economics 7 claims · Live 2 · Beta 4 · n/a 1

AI Token Economics

7 claims checked · Live 2 · Beta 4 · n/a 1

  • Live Forecast by workflow, from drivers, before the build.Section Executive summary · Per-answer cost and model spend against budget See it working (invited sign-in)
  • Live Limit by persona, measure by workflow... Caps are controls, with defined behavior when they bind.Section Executive summary · Run limits and spend caps per agent with defined fallbacks See it working (invited sign-in)
  • Beta Look at fully allocated, incremental and avoidable cost side by side.Section Executive summary · Benefits contracts with fully allocated, incremental and avoidable cost views
  • Beta Investigate rising tokens per task against accepted completion, review effort, latency and businessSection Rule · Scorecard: cases reviewed, acceptance, correction and rework rates, tokens per accepted case
  • Beta Build a forecast by workflow that reconciles to invoicesSection In brief · Showback by workflow and persona
  • Beta manage AI by the cost and value of accepted workSection Operating · Cost per accepted task against an owner’s acceptance rule

Honest gap. Costs use list prices, not the bill, and no accepted tasks are recorded yet.

Read the paper

Data Is Not Just Code 12 claims · Live 7 · Beta 4 · n/a 1

Data Is Not Just Code

12 claims checked · Live 7 · Beta 4 · n/a 1

  • Live A precise number can answer the wrong question.Section 01 The executive case · Purpose-specific measures with owners: ’open claim’ resolves by purpose (service, reserving) See it working (invited sign-in)
  • Live Context is an operational asset, not a longer prompt.Section 03 Context · compile_context: smallest permitted context with provenance and a token budget See it working (invited sign-in)
  • Live Use one delivery platform; match the gates to the consequences.Section 08 The release pipeline · Context evals (terms, metric golden values, policy, masking) with a critical veto that blocks packaging and deploy See it working (invited sign-in)
  • Live Reconciliation helps. It can’t be the only definition of correct.Section 09 Build and test · Pipeline tests by class (contract, transformation, grain and join, reconciliation) fail publish See it working (invited sign-in)
  • Live Breaking changes publish a new versionSection 10 Migrate and release · Breaking interface change without a major version fails the build See it working (invited sign-in)
  • Live Watch for plausible but wrongSection 11 Operate and change · Every live answer scored deterministically See it working (invited sign-in)

Honest gap. Answers as of a past filing date wait for a reviewer sign-off.

Read the paper

Rewiring Data Management with AI 12 claims · Live 3 · Beta 7 · n/a 2

Rewiring Data Management with AI

12 claims checked · Live 3 · Beta 7 · n/a 2

  • Live An approved contract should drive the build.Section 05 Contract-driven delivery · Data product manifest (rjl.data-product.v1) and ODCS-shaped contract checked in every build See it working (invited sign-in)
  • Live Match the check to the use.Section 06 Quality · Quality mode per product (warn or allow) See it working (invited sign-in)
  • Live One claims request, end to end... It isn’t a client deployment or an achieved result.Section 10 A worked workflow · Two end-to-end use cases on fictional data: claims (insurer) and banking regulatory reporting See it working (invited sign-in)
  • Beta AI proposes. The business keeps its decisions.Section 03 The lifecycle · AI metadata proposals (descriptions, synonyms, terms) served only after a logged steward approval
  • Beta An owner should receive a proposed interpretation with its evidence, alternatives, affected consumersSection 04 Business participation · Impact analysis: a proposed change lists affected agents, products and reports
  • Beta Measure what the consumer experiences.Section 07 Operations · Service objectives in consumer terms with error budgets

Honest gap. Process mining and DSPM comparison are off by default.

Read the paper

Build It to Run 11 claims · Live 6 · Beta 2 · Not supported 1 · n/a 2

Build It to Run

11 claims checked · Live 6 · Beta 2 · Not supported 1 · n/a 2

  • Live No contract, no build.Section 06 Build from a contract · Data product manifest (rjl.data-product.v1) and ODCS-shaped contract checked in every build See it working (invited sign-in)
  • Live A pipeline is done when someone who didn’t build it can run it.Section 07 Ship the run kit · Run kit release gate opens only after a non-builder rehearsal See it working (invited sign-in)
  • Live Write the tests that production failures would have needed.Section 08 Test for what breaks · Pipeline tests by class (contract, transformation, grain and join, reconciliation) fail publish See it working (invited sign-in)
  • Live Every material incident buys one change to the build.Section 12 Close the loop · Incident records, reporting clocks per regime, postmortems See it working (invited sign-in)
  • Live status codes from an allowed listSection 14 Worked example · In-line quarantine with reasons See it working (invited sign-in)
  • Beta Set objectives in the consumer’s terms, and page only for what they would notice.Section 09 Detect before the consumer · Service objectives in consumer terms with error budgets

Honest gap. The authority ladder is built; the AI-assisted triage agent itself is on the backlog.

Read the paper

The Day After the Demo (book) 20 claims · Live 9 · Beta 9 · n/a 2

The Day After the Demo (book)

20 claims checked · Live 9 · Beta 9 · n/a 2

  • Live Separate four kinds of decision: business outcome, information meaning, the technical service and risk.Section Chapter 2 · Decision-rights register with independence checks applied by every workflow See it working (invited sign-in)
  • Live Redesign the work and the reviewSection Chapter 3 · Human review: reviewer inbox, due times, email alert, accept, return or reject See it working
  • Live I use the Four-Part System to assign the jobs.Section Chapter 7 · Reference architecture and capability map: every layer marked working, partial or not yet See it working (invited sign-in)
  • Live A data product is information, or a capability, that a defined consumer can find, understand and use through a supported contract.Section Chapter 10 · Data product manifest (rjl.data-product.v1) and ODCS-shaped contract checked in every build See it working (invited sign-in)
  • Live A context specification, owned and versionedSection Chapter 12 · compile_context: smallest permitted context with provenance and a token budget See it working (invited sign-in)
  • Live split the action into four steps: propose, approve, submit and confirmSection Chapter 13 · Write-back by the signed-in person only: field allowlist, intent audited first, operation IDs, conditional write See it working (invited sign-in)

Honest gap. The chapters map to the same features as the papers; see each paper’s panel.

Read the paper

Regulatory Reporting with Governed Context 6 claims · Live 4 · Beta 2

Regulatory Reporting with Governed Context

6 claims checked · Live 4 · Beta 2

  • Live Each report line is fed by an approved metric and carries its MDRM code.Section 03 · Regulatory reporting: MDRM-coded FFIEC 041 and FR Y-9C lines with edit checks See it working (invited sign-in)
  • Live The model never does the arithmetic: metrics compile to fixed SQL.Section 05 · Semantic layer compiled to fixed SQL See it working (invited sign-in)
  • Live Preliminary, final and restated values are never combined.Section 06 · Values labelled with status and period See it working (invited sign-in)
  • Live Lineage runs from the source column to the report line.Section 08 · Element-level lineage from source column to product, cited by answers See it working (invited sign-in)
  • Beta Edit checks cite their source and its verification status.Section 07 · Published-style edit checks for FFIEC 041, FR Y-9C and NAIC
  • Beta MDRM report items are imported into an ontology only as proposals.Section 09 · Industry models browser: FIBO, gist and MDRM report items

Honest gap. The edit-check publications are not yet verified against the publishers, and the reporting product is in a managed sandbox, not certified.

Read the paper

Governing AI from Registry to Runtime 9 claims · Live 5 · Beta 4

Governing AI from Registry to Runtime

9 claims checked · Live 5 · Beta 4

  • Live No record, no production: every audit record carries the release ID, and a changed composition without a passed evaluation serves no agent.Section 04 · Release ID and policy hash on every audit record; a changed composition without a passed evaluation serves no agent See it working (invited sign-in)
  • Live The registry is the AI inventory, with a discovery check.Section 04 · AI inventory with discovery check and sealed export See it working (invited sign-in)
  • Live Independence rules are checked by the system, not left to memory.Section 03 · Decision-rights register with independence checks applied by every workflow See it working (invited sign-in)
  • Live Guardrails run outside the model: a policy decision comes before any read.Section 08 · Governed MCP gateway: policy decision before any read on every call See it working (invited sign-in)
  • Live Kill switches stop an agent, tool, product or role.Section 13 · Kill switches per agent, tool, product and role, effective in about 30 seconds See it working (invited sign-in)
  • Beta Release decisions are signed by three different people.Section 06 · Release decision records signed by three different people; promotion gate
  • Beta Every role carries its line, and internal audit is read-only.Section 03 · Three lines: every role carries its line, internal audit read-only, second-line approval for critical items, findings board
  • Beta Exceptions carry an owner, a compensating control and an expiry.Section 06 · Exceptions with owner, compensating control, expiry and retest; workbench with independent acceptance
  • Beta Restricted identifiers are tokenized, and detokenizing needs two people.Section 09 · Tokenization of restricted identifiers, two-person detokenize, training export gate, consumer notices

Honest gap. Several second-line controls are built but run in warn mode rather than enforcing, and RJL’s own fallback model route was evaluated on 20 November 2026 and did not meet the threshold; fixes are being made and the route will be re-tested before it is relied on.

Read the paper

Built but not live

What is built, but not yet live, and why.

  • Warn mode. Several controls record what they would refuse before they refuse it: row-level rules, delegated identity per answer, approval binding, route and agent evaluation gates, and three lines of defense. Each moves to enforce once enough runs are recorded.
  • Beta, off until switched on. Paid options and integrations that need a client account ship switched off: vector retrieval, DSPM hooks, the marketplace, notifications, one sign-in across the three products, and the runtime context gate, which is switched on per use case.
  • Waiting for a live check. Connectors, warehouse delivery and catalog sync are tested with fakes; a live check needs a client system.
  • Model provider. The primary model provider is unavailable on RJL’s account, so answers run on a second provider as a fallback. The route is recorded on every answer, and route evaluations run in warn mode.
  • No certified data product. Both demo products sit in a managed sandbox, because certification needs a run kit rehearsal by people who did not build the product.

What it does not do

Patterns the papers describe that RJL does not run.

  • Not supported The paper’s example stack (Snowflake, Okta, Entra Agent ID, Hasura DDN, Cortex, Cyera) is not what RJL runs; the engine runs on AWS Lambda, Cognito, DynamoDB, S3
  • Not supported Enforcement inside a customer data platform (tag-based masking and row access at the warehouse)
  • Not supported AI-assisted operations triage agent
  • Not supported Confidential computing runtime (TEE, confidential GPU)
  • Not supported OWL reasoner or SHACL validation engine
  • Not supported Executable process engine (BPMN is descriptive only)
  • Not supported Workforce assistant selection, browser or computer-use agents

RJL is not a DSPM scanner, a process engine, an OWL reasoner or a confidential computing runtime, and no client system is connected. Nothing on this page is a compliance claim or legal advice.

Status as of 20 November 2026, from the product records (engine 7d286bd, Workflow Studio b5176b3, Ontology Studio 2f4d90a). A status never rises above a recorded check. It is a reference implementation, not a client deployment.