← Back to interactive architecture

Risks, Assumptions, Constraints & NFRs

The registers that keep this architecture honest: risks with owners and mitigations, labeled assumptions, hard constraints, open questions, dependencies, and the NFR register with measurement methods.

Why this document exists: the brief demands unknowns be visible, not hidden. Everything below is load-bearing: nodes reference these IDs, the app renders them in inspectors, and phase gates check them. Likelihood scores are deliberately unassessed until P0 scoring with the organization present.

1 · Risk register

IDRiskImpact / likelihoodMitigation (controls / initiatives)Owner
R-01Strangler stall
Extraction waves lose funding/momentum; dual-run (cloud + legacy + ESB) becomes permanent, doubling run cost.
high-impact
unassessed (placeholder - score at P0)
Wave exit criteria include decommission evidence; steering reviews drawdown curve; init-legacy-retire is a first-class initiative.Executive steering
R-02Legacy knowledge loss
The people who understand monolith semantics leave before extraction completes.
high-impact
unassessed (placeholder - score at P0)
P1 documentation sprint; ACL contract tests encode behavior; pairing rotations.Legacy platform team
R-03Messaging semantics misuse
Teams put work on streams or facts on queues; DLQs unowned; replay untested.
medium-impact
unassessed (placeholder - score at P0)
ADR-07 decision tree; golden-path templates; DLQ drills as consumer onboarding gate (C-INT-02).Integration platform
R-04Data-quality debt propagates
Ungoverned sources feed products and AI; trust collapses on first bad executive number.
high-impact
unassessed (placeholder - score at P0)
Quality gates on zone promotion; certification before consumption; catalog-first onboarding (RB-09).Data governance council
R-05AI retrieval leakage
RAG surfaces content the caller isn't entitled to see - a data breach through the assistant.
high-impact
unassessed (placeholder - score at P0)
Authorization-aware retrieval (C-AI-03); entitlement-bypass tests as release gate; only cataloged sources embeddable.AI platform + security
R-06Agent overreach
An agent with broad permissions performs unintended production actions.
high-impact
unassessed (placeholder - score at P0)
Tool allowlists + transaction limits + HITL + kill switch (C-AI-04/05, RB-12); autonomy levels graduated by gov-ai.AI governance council
R-07Cloud cost overrun
Untagged sprawl, idle capacity, unwatched AI token burn.
medium-impact
unassessed (placeholder - score at P0)
Tag-or-no-deploy policy; budgets + anomaly alerts; monthly FinOps review; gateway token metering (ADR-16).FinOps practice
R-08Platform team bottleneck
IDP becomes a ticket queue; teams route around it; shadow platforms return.
medium-impact
unassessed (placeholder - score at P0)
Platform-as-product operating model with SLOs + adoption metrics; self-service golden paths; gov-platform arbitration.Platform engineering
R-09Unexamined lock-in
Provider coupling grows silently beyond the consciously accepted level.
medium-impact
unassessed (placeholder - score at P0)
Annual lock-in/exit review (ADR-02); open table formats at the data layer; gateway abstraction at the AI layer.Enterprise architecture
R-10Detection gaps
New platforms ship without SIEM onboarding; incidents are invisible.
high-impact
unassessed (placeholder - score at P0)
SIEM onboarding in every go-live checklist (C-OPS-03); coverage map reviewed by gov-secrisk.Security operations
R-11Ransomware reaches backups
Attacker with admin credentials encrypts/deletes primary backups.
high-impact
unassessed (placeholder - score at P0)
Immutability windows + isolated cyber vault with separate credentials (C-RES-01/02); vault recovery drills.SRE + security
R-12B2B modernization under-adoption
Partners stay on legacy exchange paths; the B2B gateway serves a fraction of traffic.
medium-impact
unassessed (placeholder - score at P0)
Onboarding runbook with lead-time target; per-partner migration plan in init-b2b; partner portal incentives.Partner operations
R-13Skills gap
K8s + data + AI platform skills are scarce; hiring lags the roadmap.
medium-impact
unassessed (placeholder - score at P0)
A-04 validation at P0; training budget; managed services preferred; PaaS escape valve (ADR-04).CTO organization
R-14Governance theater
Boards meet without decision rights or SLAs; teams bypass; exceptions never expire.
medium-impact
unassessed (placeholder - score at P0)
Charters with decision rights + review SLAs; policy-as-code enforcement; exception expiry automation (PR-12).Executive steering
R-15Compliance scope surprise
Payment/privacy obligations turn out broader than assumed (A-09/OQ-03).
high-impact
unassessed (placeholder - score at P0)
Early org-input checkpoints at P0/P1; configurable overlays ready (C-DP-05, C-PRV-01); no scope-narrowing design bets before review.Risk & compliance

2 · Assumptions

IDAssumptionStatement
A-01Illustrative baselineThe source document defines no real organization. The 'current state' here is an illustrative baseline constructed ONLY from challenges the source describes (S4, S27-S28, S49, S51, S56-S58): legacy monolith, ESB point-to-point sprawl, overnight-batch reporting, SaaS integrated ad hoc, departmental app sprawl. Phase 1 discovery replaces it with evidence.
A-02Target content is proposalTarget-state components and controls are design proposals / industry-practice extensions consistent with source themes - the source mandates almost none of them specifically. Each node's source_traceability field distinguishes source-derived themes from proposals.
A-03Single primary cloud pending ratificationOne primary cloud provider is assumed (ADR-02); the provider identity itself is an organization selection, not made here.
A-04Platform team capacityA platform team capable of operating managed Kubernetes is assumed (drives ADR-04). If untrue, the PaaS-first variant in ADR-04 applies.
A-05All recovery/availability values are placeholdersNo BIA exists. Every availability_target/rto/rpo value is a tier-band placeholder, never a commitment.
A-06Residency unknownData residency/sovereignty obligations are unknown (OQ-02); residency is modeled as a configurable control overlay (C-DP-05).
A-07Provisional service tiersCriticality/tier assignments are provisional judgments pending BIA.
A-08Vendors are illustrationsRepresentative products in the integration catalog are illustrative options with selection criteria - none are selections or endorsements.
A-09Payment scope unreviewedPayment flows assume provider-side tokenization; actual compliance scope requires organization-specific assessment. No compliance claim is made anywhere in this package.
A-10Volumes unknownLoad, growth and storage volumes are unknown (OQ-04); capacity records carry placeholders plus the test method that will replace them.
A-11No certification claimsRegulatory/compliance frameworks are modeled as configurable overlays; nothing here asserts certification or compliance.
A-12Ownership placeholdersAll owner fields are role placeholders to be bound to named people/teams at Phase 0.

3 · Constraints

IDConstraintStatement
CT-01Static deliveryThis package is standards-based HTML/CSS/JS with no build step and no CDN dependencies (offline-capable by design; supply-chain surface minimized).
CT-02Single canonical modelAll views render from data/architecture.json + views.json; no view may contradict the canonical model.
CT-03WCAG 2.2 AA targetAccessibility target for all HTML deliverables; table alternatives exist for every graph.
CT-04Legacy continuityThe legacy core must keep operating until strangler waves complete - business continuity outranks migration speed.
CT-05SoR discipline retainedERP/CRM/HRIS remain SaaS systems of record; modernization re-platforms the integration around them, not the systems themselves.
CT-06Governed integration onlyCross-domain access happens via APIs/events/governed pipelines - never direct database access across ownership boundaries.

4 · Open questions (require organization input)

IDQuestion
OQ-01Region strategy & DR region pair (feeds OD-01).
OQ-02Data residency / sovereignty obligations by jurisdiction.
OQ-03Applicable regulatory overlays (privacy, sector-specific).
OQ-04Volumes: traffic, orders, partners, data growth, peaks (capacity baselines).
OQ-05Platform team size/skills reality (validates A-04).
OQ-06SaaS data backup scope per vendor shared-responsibility review.
OQ-07Payment scope: card-present? stored credentials? provider model? (validates A-09).
OQ-08Workflow/case realization preference (feeds OD-03).
OQ-09AI use-case inventory with risk classes (feeds gov-ai intake).
OQ-10Existing contracts/licenses constraining product options.
OQ-11Current identity estate (greenfield CIAM or consolidation?).
OQ-12Any concrete edge-computing use case with NFRs (feeds OD-05).

5 · Initiative dependencies

Sequencing constraints between roadmap initiatives (full detail incl. exit criteria and rollback stances in the roadmap view).

InitiativeWaveDepends onRetires risk
Governance & decision fabricwave-0R-14
Discovery, APM inventory & BIAwave-0init-governanceR-02
Cloud landing zones & network foundationwave-0init-governanceR-07
Identity foundation (workforce + CIAM + PAM + JML)wave-0init-governanceR-10
Observability & incident baselinewave-0init-landingR-10
FinOps baseline (tagging, allocation, budgets)wave-0init-landingR-07
Platform engineering MVP (golden paths)wave-0init-landing, init-identityR-08
API management & edge gatewaywave-1init-platform-mvpR-03
Event backbone & queueswave-1init-platform-mvpR-03
Legacy facade & anti-corruption layerwave-1init-api-platformR-01
B2B & partner exchange modernizationwave-1init-event-backboneR-12
Lakehouse, catalog & first data productswave-1init-landing, init-event-backboneR-04
Strangler wave 1: customer & order extractionwave-1init-legacy-facade, init-platform-mvpR-01
Strangler wave 2: fulfillment, billing & ESB drainwave-2init-strangler-1, init-b2bR-01
Resilience & cyber-recovery proofwave-2init-data-platform, init-strangler-1R-11
AI foundation: gateway, RAG, evals, costwave-2init-data-platform, init-identityR-05
Governed agentic automation (first use case)wave-3init-ai-foundationR-06
Legacy decommission with evidencewave-3init-strangler-2, init-data-platformR-01
Continuous optimization & radar cadencewave-3init-resilience, init-finopsR-07

6 · Non-functional requirements register

Requirement stances with the measurement method that will replace each placeholder. No numeric target below is a commitment until bound at P1–P2 with organization data.

CategoryRequirement stanceMeasurement / verification method
AvailabilityService tiers T1–T4 define expectations; numeric targets are assigned per service ONLY after the BIA prices downtime (P1). Multi-AZ is the T1/T2 baseline (C-RES-04); multi-region is decision OD-01.BIA report → tier assignment per service in the model; availability measured from synthetic + SLO data, reported monthly.
PerformanceLatency/throughput/concurrency/payload/batch-window targets are placeholders per component (capacity records) until OQ-04 volumes exist. Interactive paths carry an indicative sub-second p95 placeholder explicitly marked as unvalidated.Load tests against agreed profiles in a performance environment; RUM/synthetic percentiles in production; targets versioned in NFR register.
ScalabilityHorizontal-first (stateless services, HPA), queue-based load leveling for bursty work, partitioning for streams/data, read replicas where read-heavy, PaaS/serverless for spiky low-duty workloads. Scaling triggers are defined per component at onboarding (RB-05/RB-18).Scale tests to 2× expected peak placeholder; autoscale drill; capacity review cadence.
SecurityIdentity-first zero trust (C-ID-*), deny-by-default networks (C-NET-*), OAuth2/OIDC discipline (C-APP-01), encryption everywhere (C-DP-02), supply-chain gates (C-SC-*), AI-specific controls (C-AI-*). Full catalog: security model doc.Control tests with evidence per controls.json; red-team/pen-test scope per change class; posture scan trends.
ResilienceDependency failure = designed behavior: timeouts, circuit breakers, bounded retries with jitter, bulkheads, DLQs, graceful degradation matrices per consumer; recovery via tested backup/restore, tiered DR, cyber vault (C-RES-*).Chaos/failure drills per tier cadence; DR rehearsal timings vs objectives; degradation matrix accuracy reviewed at PIR.
OperabilityA service is done when observable (OTel + SLO + owner), on-call-staffed, runbook'd, capacity-planned, restore-tested. Platform SLOs cover the paths teams depend on.ORR checklist (P8); orphan-alert audits; runbook drill records.
MaintainabilityModular boundaries (ADR-05), golden-path consistency, contract-first interfaces, automated tests as release gates, tech-debt register reviewed quarterly with capacity actually allocated.DORA metrics; debt burn-down; standards-conformance scans.
PortabilityValued where cheap and real: open table formats (data), OTel (telemetry), OCI/K8s (compute), gateway abstraction (AI providers). NOT pursued via lowest-common-denominator abstraction layers (ADR-03). Lock-in is accepted consciously and reviewed annually.Annual lock-in/exit review per ADR-02; exit-cost estimate refresh.
ComplianceNo certification or regulatory compliance is claimed anywhere (A-11). Privacy, records, residency and sector overlays are modeled as configurable control sets (C-DP-05, C-PRV-01, C-DP-04) that activate once OQ-02/03 answers arrive.Overlay activation records; control evidence retention (C-GOV-03); external audit only after org scoping.
AccessibilityThis package itself targets WCAG 2.2 AA: keyboard-complete, visible focus, contrast-checked tokens, reduced-motion support, table alternatives for every graph, no color-only meaning.Keyboard walkthrough; automated a11y scan; screen-reader spot checks (validation section of README).
CostEvery product allocated (tags enforced at deploy), budgeted, anomaly-monitored; unit economics per product; AI token budgets per consumer; optimization cadence (rightsizing, commitments, storage lifecycle, egress review) monthly via FinOps review.Allocation coverage %; budget variance; anomaly MTTR; unit-cost trend per product.