Validation architecture Governed enterprise runtime

bp Sphere Runtime Architecture

bp Sphere is the governed runtime between enterprise signals and enterprise action. It sits above systems of record, data platforms, workflow tools, policy repositories, and AI services to make decisions governed, explainable, auditable, replayable, and operationally useful.

Open Fabric Documentation Open Platform Architecture Open Agent Registry
Positioning
Decision + action runtimeNot another reporting layer or chatbot surface.
Runtime chain
Signal to replaySignals, context, agents, policy, evidence, decision, action, replay, learning.
Control model
Human accountableAutomation is bounded by policy, authority, evidence, and approval gates.
Enterprise fit
Systems stay in placeSAP, Ariba, Databricks, Palantir, ServiceNow, and workflow systems remain sources and execution rails.
Reference architecture

Condensed runtime architecture

Existing bp platforms feed the signal layer; bp Sphere then resolves context, orchestrates agents, applies policy, proves evidence, creates decisions, routes human action, and preserves replay.

01
bp Systems of Record
SAPAribaDatabricksPalantirServiceNowWorkdayTreasuryDocument Stores
then
02
Enterprise Signal Layer
Event captureChange detectionSignal classificationSource lineage
then
03
Context and Ontology Runtime
Business object resolutionEntity graphProcess contextHistorical context
then
04
Agent Orchestration Runtime
Specialized agentsTool callsMission plansRuntime contracts
then
05
Policy and Control Runtime
AccessAuthoritySOXCybersecurityHuman approvalWrite controls
then
06
Evidence Intelligence Fabric
DiscoveryResolutionExtractionLineageEvidence packViewer
then
07
Decision Runtime
Decision crateRecommendationAlternativesRiskConfidenceImpact
then
08
Human Supervisor Runtime
ApproveRejectModifyEscalateOverrideExplainReplay
then
09
Action Runtime
Draft actionWorkflow triggerAPI write-backTicket creationCommunication
then
10
Replay, Audit and Learning Runtime
Full traceEvidence snapshotPolicy resultHuman judgmentOutcome learning
How to read this architecture

The runtime loop bp should remember

Everything on mission pages should map back to this chain. KPIs and dashboards are presentation. The runtime is the governed path that turns an enterprise signal into an explainable decision and a controlled action.

SignalSomething changes in SAP, Ariba, Databricks, workflow, documents, markets, or operational queues.
Contextbp Sphere resolves the business object, owner, history, policy domain, and similar cases.
Evidence + PolicySource records, documents, controls, authority gates, and approval requirements are assembled.
Decision + HumanAgents create a decision crate, but humans remain accountable where judgment is required.
Action + ReplayApproved actions route back to systems and the full decision trace becomes replayable.

Core positioning

bp Sphere is not the system of record, model provider, or data lake. It is the governed decision and action runtime that composes those assets into accountable enterprise operations.

Technology credibility test

Every meaningful decision should produce runtime records for signal, context, evidence, policy checks, agent steps, human judgment, action, outcome, and replay.

Detailed runtime layers

These layers are the operating model behind mission pages such as P2P, R2R, FP&A, Treasury, Credit, Journal, and Transformation Intelligence.

01

Enterprise Signal Layer

Detects what changed across systems and turns raw events into business signals.

Runtime objects
SAPAribaDatabricksServiceNowWorkdayTreasuryDocuments
What it controls or produces
Duplicate invoice riskJournal anomalyApproval boundary exceededSupplier risk changed
02

Context and Ontology Runtime

Resolves the business meaning of each signal using entities, relationships, ownership, and history.

Runtime objects
SupplierContractPOInvoiceJournalCost centerLegal entity
What it controls or produces
What changed?Who owns it?What policies apply?What evidence is required?
03

Agent Orchestration Runtime

Coordinates specialized agents rather than relying on one generic chatbot.

Runtime objects
Signal agentEvidence agentPolicy agentRisk agentSimulation agentAction agent
What it controls or produces
Select missionRun agent planValidate evidenceGenerate decision crate
04

Policy and Control Runtime

Applies authority, SOX, cyber, data, approval, and write-back gates before any decision or action.

Runtime objects
Authority matrixSegregation of dutiesPII rulesData residencyWrite permissions
What it controls or produces
Can recommendCan draftMust escalateHuman approval required
05

Evidence Intelligence Fabric

Builds credibility by retrieving, extracting, linking, and displaying evidence before recommendation.

Runtime objects
InvoicePOGoods receiptContract clausePolicy documentSAP transaction
What it controls or produces
DiscoveryResolutionExtractionLineageEvidence packViewer
06

Decision Runtime

Creates the governed decision object the human can inspect, challenge, approve, or replay.

Runtime objects
Signal summaryImpactRoot causePolicy checksAlternativesConfidence
What it controls or produces
Decision crateHuman ownerApproval requirementReplay link
07

Action Runtime

Moves from recommendation to governed execution across three maturity horizons.

Runtime objects
Decision supportGoverned actionControlled autonomy
What it controls or produces
Draft holdCreate ticketRoute exceptionSend communicationWrite back by API
08

Human Supervisor Runtime

Makes accountability explicit across analysts, supervisors, tower leads, controllers, and CFO roles.

Runtime objects
Inspect evidenceApproveRejectModifyEscalateOverrideReplay
What it controls or produces
Human remains accountable where policy, value, or judgment requires it
09

Replay and Audit Runtime

Stores the full trail needed to prove what happened, why it happened, and who approved it.

Runtime objects
SignalContextAgent stepsModel responsePolicy checksEvidenceOutcome
What it controls or produces
ReplayableAuditableTimestampedOutcome-linked
10

Observability and Reliability Runtime

Shows platform operators whether agents, tools, policies, evidence, and replay are healthy.

Runtime objects
LatencyToken usageFailed toolsPolicy denialsEvidence gapsApproval latency
What it controls or produces
Agent registryPolicy registryEvidence registryKill switchSafe degradation
Concrete execution path: P2P duplicate invoice

This is the concrete path bp technology reviewers should expect to inspect live: every decision must leave runtime records for signal, context, evidence, policy, agents, human action, and replay.

1

SAP/Ariba event arrives

Invoice enters the enterprise event stream with source lineage.

2

Duplicate signal detected

Signal layer identifies same supplier, amount, PO reference, and payment window.

3

Context resolved

Supplier, PO, invoice, cost center, payment history, and prior cases are linked.

4

Evidence retrieved

Invoice, PO, GRN, policy, and historical payment records become an evidence pack.

5

Policy checked

Duplicate payment control and authority boundary rules determine the allowed action.

6

Agents participate

Matching, evidence, policy, recommendation, and escalation agents create the decision.

7

Decision crate created

Supervisor sees risk, evidence, recommendation, alternatives, and confidence.

8

Human approves hold

Supervisor approves or modifies the payment hold with a reason code.

9

Action executed

Action runtime sends the governed hold instruction to SAP or Ariba workflow.

10

Replay and learning stored

The full trace is saved and the duplicate pattern improves future matching.

Likely validation questions

This page should make the platform boundary clear: bp Sphere complements bp systems of record and governed data platforms. It does not replace them.

Where does Databricks fit?

Databricks provides governed data products, analytical features, lineage, and quality metadata. bp Sphere consumes those products and turns them into explainable decisions and governed actions.

Where does Palantir / Foundry fit?

Foundry can provide ontology and business-object relationships. bp Sphere uses that ontology as context, then adds orchestration, evidence, policy, decision, action, replay, and human accountability.

Where does Yala fit?

Yala can be the deployment and enterprise hosting path. The runtime capabilities remain the same and can be deployed through bp-approved infrastructure.

Does bp Sphere replace SAP or Ariba?

No. SAP and Ariba remain systems of record. bp Sphere decides, explains, recommends, drafts, and routes governed actions back into those systems.

Proof required for credibility

Runtime records

Every agent run should produce signal, context, evidence, policy, recommendation, human action, and replay records.

Policy gates

Every recommendation and write-back path is bounded by authority, SOX, cyber, data, and human approval controls.

Evidence before opinion

Recommendations must show the invoices, POs, contracts, policy clauses, transactions, and source snapshots used.

Safe operations

Operators need health, latency, failures, evidence gaps, policy denials, approval latency, kill switch, and safe degradation controls.

Latest runtime wiring now implemented

The architecture now has code-backed seams for experience orchestration, governed chat, assistant portability, token economy, policy fail-closed behavior, replay determinism, signed evidence, runtime adoption auditing, learning promotion, autonomy demotion, case-state propagation, and observability readiness.

Experience Orchestration Layer

Surface available - Shared decision objects render across desktop, mobile, and chat surfaces.

/ui/mission/p2p, /ui/mobile-cockpit, /ui/enterprise-chat

Governed Chat Interface

Surface available - Chat resolves role, mission, case, evidence, policy, action contract, replay, and feedback.

/ui/enterprise-chat

Workforce Intelligence Integration Fabric

Surface available - Assistant registry, connector packages, certification, compliance mappings, identity resolution, failover drills, and workforce coordination APIs.

/ui/workforce-intelligence-fabric, /ui/assistant-operations-center

Token Economy Control Plane

Surface available - LLM calls are routed, cached, denied, downgraded, or escalated before spend occurs; token value is attributed to decisions.

/ui/ai-cost-usage, /api/iaf/decision-intelligence/token-economy/decision

Guardrail Policy Engine

Surface available - Policy evaluation fails closed in production posture; non-fail-closed failures route to escalation instead of silent execution.

/ui/policy-registry, /ui/control-plane

Decision Replay + LLM Determinism

Surface available - Execution seed, prompt/context hash, evidence hash, policy version, and replay records are persisted and validated.

/ui/decision-journey, /ui/single-decision-audit, /ui/decision-replay-studio

Still external, not hidden

  • SAP/SOR phase 2-3: RFC/OData reads, CDC streams, governed write-back, and master-data sync require customer credentials and agreed write boundaries.
  • Live Entra tenant/Graph sync and vendor assistant connector credentials are external deployment bindings.
  • Production OTEL/SLO alerting requires environment configuration and failure-mode drills in dev/prod.
  • Agent de-stubbing and route-level adoption breadth continue as capability migration work, not platform architecture gaps.

Strongest validation message

bp Sphere is the governed runtime between enterprise signals and enterprise action. It does not replace bp's platforms. It makes them work together through signal, context, policy, evidence, decision, action, replay, and human accountability.