Phase B — Business Architecture

The capabilities Schema-to-Screen introduces or changes, the value stream from a request to a governed screen, the end-to-end process with its decision points, and who is accountable for what.

B1 · Business capability mapLevel-1 capabilities with level-2 detail, heat-mapped by change

new introduced by the platform · enhanced existing capability materially changed · existing integrated as-is. Reflows to one column on narrow screens.

1 · Demand & intake new

Turning business prose into a typed, checkable use-case spec.

Request captureSpec typingArchetype selectionClarification round-tripsActor & scope validation

2 · Schema knowledge new

Knowing the schema well enough to retrieve the right seven tables out of three thousand.

Crawling & profilingCatalog cardsDescription reviewJoin-path miningGlossary & synonymsVersioningSubgraph retrieval

3 · Data governance enhanced

Classification and entitlements become executable inputs rather than documents.

Column classificationEntitlement policyOwner reviewCoverage gatingDrift signalling

4 · Contract generation new

Composing a GraphQL slice and its persisted operations from a subgraph and the policy catalog.

Slice compositionPolicy applicationOperation setProposal designCost weighting

5 · Experience generation new

Producing the on-demand screen as a spec, and custom components as reviewed exceptions.

UI spec compositionArchetype layoutsBinding to operationsCustom-component requestsRegistry curation

6 · Verification new

Deterministic proof that an artifact is safe to publish.

CompositionPolicy coverageEntitlement compileCost & complexityBinding validityGolden testsPII scanAccessibilityGroundedness

7 · Review & publish new

Human judgment applied to contracts, with evidence, in minutes.

Review workspaceApproval routingPublish & registrationAuto-publish rules (read-only)

8 · Runtime delivery enhanced

Serving every use case through one graph and one runtime, with authorization compiled in.

Supergraph registryPersisted operationsQuery compilationSpec renderingProposal handlingTelemetry

9 · Change & operations new

Making schema, policy and model change computable rather than discovered.

Drift detectionRe-verificationDegraded handlingRetirementRelease-unit pinning

10 · Platform governance existing

The enterprise controls the platform plugs into rather than replaces.

Identity & claimsAuditCost managementModel lifecycleSecrets

B2 · Value streamFrom a business question to a governed screen

Each stage names the value it adds, who participates, and the KPI that tells you the stage is healthy. Machine stages are minutes; the two human stages are the designed bottleneck.

01
Request

Value: a question stated in the business's words.

Who: requester.

KPI: requests/week.

02
Specify

Value: a typed spec with actors, questions, allowed actions.

Who: intake agent, requester.

KPI: clarification rounds; refusal rate.

03
Ground

Value: the right subgraph, with join paths and citations.

Who: Scout over the catalog.

KPI: subgraph precision/recall; ambiguity rate.

04
Compose

Value: a GraphQL slice and a UI spec, policies applied.

Who: composers.

KPI: first-pass gate success; escape-hatch rate.

05
Verify

Value: deterministic proof of safety.

Who: verifier, sandbox.

KPI: gate failure taxonomy; time in gates.

06
Approve

Value: accountable human judgment on evidence.

Who: domain owner; Finance for writes.

KPI: reviews/hour; rejection reasons.

07
Publish

Value: one more slice in one graph; one more spec in one runtime.

Who: registry, runtime.

KPI: time request → live.

08
Operate

Value: screens that stay correct as the schema moves.

Who: crawler, gates, owners.

KPI: drift → re-verify latency; degraded count.

B3 · Process flow (swimlanes)The generation run with its decision points, retries and parking

Solid arrows are the happy path; dashed arrows are loops (clarify, retry) and the drift trigger. Every loop is bounded by the supervisor; every dead end parks with a person.

REQUESTER INTAKE &SUPERVISOR SCOUT &COMPOSERS VERIFIER REVIEWER /DATA OWNER RUNTIME &OPERATIONS Submit requestprose + users Clarifyanswer the question Type into specarchetype · actors ambiguous? Retrieve subgraphcite-or-refuse tables clear? Composeslice + UI spec Run nine gatessample sandbox pass? Parkednotes → next run Review evidenceevidence bundle approved? Publish & pinslice · spec · unit Catalog driftcrawler bumps version yes · ask no no · two look-alikes → ask yes fail → responsible composer, ≤2 retries still failing pass reject + notes yes re-verify every affected use case (gates only; no model)

B4 · Actor / role → capability matrixWho is responsible, who approves, who is consulted, who is informed

CapabilityRequesterPlatform agentsData ownerDomain reviewerSecurityPlatform engBusiness user
Demand & intakeRRIC
Schema knowledgeRA (descriptions)CR (crawler)
Data governanceIRCAC
Contract generationRIIC (policy catalog)C
Experience generationCRIA (custom components)C
VerificationRIA (gate set)R (sandbox)
Review & publishICACR (publish)I
Runtime deliveryCRuses
Change & operationsIR (re-verify)A (drift)IIRI
Platform governanceAR
R responsibleA approves / accountableC consultedI informed

B5 · Business services catalogThe services the platform offers, and to whom

Business serviceDescriptionConsumerService level (target)
Use-case requestSubmit a business question; receive a typed spec or a clarifying questionRequestersSpec or question within minutes; ≤2 clarification rounds before parking
Governed screen deliveryA published, versioned screen and its GraphQL slice for an approved use caseDomain teamsMachine time in minutes; review SLA owned by the domain
Catalog stewardshipOwner review of table descriptions, classification coverage reports, drift noticesData ownersCrawl per schedule and on DDL events; drift notice within one crawl cycle
Policy enforcementEntitlements compiled into every query; classified fields policed on every sliceSecurity, compliance100% policy coverage before publish; zero unscoped operations reachable
Proposal handlingWrite actions delivered as proposals into the owning system's approval queueFinance, operations ownersNo direct writes; proposal status visible in the originating screen
Evidence & auditEvery run traced; every publish recorded against its release unit; any screen's SQL reproducibleAudit, securityTrace retained per policy; reproduction on demand
Component registryReviewed custom components available to every future specDevelopers, composersEscape-hatch rate tracked; registry growth reviewed monthly