Backend-for-Frontend Architecture Architect

Wrote docs/architecture/bff-architect/00-overview/bff-architect-overview.md and docs/architecture/bff-architect/verification/fitness-self-check.md. Complete Backend-for-Frontend (BFF) architecture specification establishing dedicated mobile aggregation boundaries, parallel downstream fan-out, 45 KB payload pruning, and frontend team ownership.

---

skill: bff-architect

Backend-for-Frontend Architecture: Mobile Banking App [BFF-MOB-001]

Summary

This specification establishes the Backend-for-Frontend (BFF) architecture, client-tailored aggregation boundaries, and downstream service fan-out contracts for mobile-banking-bff v2.4 under run ID banking-bff-arch-001. It powers the Retail Consumer Mobile Banking App serving 4.2 million active iOS and Android users across 18,000 peak requests/second. It decisively eliminates the mobile network bloat and severe latency demonstrated in incident MOB-4914 (where forcing mobile clients to query a generic desktop API required 12 sequential HTTP roundtrips over cellular networks, delivering 850 KB of unneeded desktop data and stalling app startup to 9.5 seconds). The architecture enforces a dedicated

Mobile BFF owned by the mobile engineering squad, parallel asynchronous downstream fan-out via gRPC to 8 core banking microservices, aggressive field pruning capping dashboard payloads at

<= 45 KB, resilient fallback degradation for non-critical widgets, and strict isolation prohibiting cross-BFF calls.

Detailed Description

Forcing diverse client form factors (such as native mobile apps, desktop web browsers, and smartwatch widgets) to query a single universal API introduces severe trade-offs. Mobile devices operate over lossy, high-latency cellular connections with constrained battery and CPU power; they require compact, pre-aggregated view models that render in a single network roundtrip. Desktop web browsers operate over high-bandwidth fiber connections and can easily absorb sprawling relational graphs. A dedicated Backend-for-Frontend isolates the unique user experience needs of the mobile app, decoupling mobile release cycles from backend microservice deployments.

iOS / Android Native Client (18,000 req/sec over 4G/5G)
                         │
                         ▼ (Single HTTP/2 Request: `GET /v1/dashboard`)
[ Dedicated Mobile BFF: Node.js / Fastify Cluster on AWS EKS ]
  ├── 1. Client Identity & Device Context Gate (TouchID / FaceID Assertion)
  ├── 2. Parallel Downstream Fan-Out (Scatter-Gather via Internal gRPC in < 25ms):
  │      ├── Accounts Service ────► Checking & Savings Balances
  │      ├── Credit Card Service ─► Available Credit & Minimum Payment
  │      ├── Rewards Service ─────► Loyalty Points Balance
  │      └── Notifications Svc ──► Unread Security Alert Badges
  ├── 3. View-Model Synthesis & Field Pruning (Strips 94% of unneeded desktop fields)
  └── 4. Graceful Degradation (If Rewards times out > 40ms, returns `points: null`)
                         │
                         ▼
HTTP 200 OK: Compact Mobile View Model (Payload Size: 38 KB, p99 <= 58 ms)

Criteria and weights

Alternatives rejected

Contracts and Invariants

Forty-Five Kilobyte Payload Ceiling [INV-BFF-01]
  Mobile BFF view-model payloads must not exceed 45 KB uncompressed.
  Exposing raw, unpruned backend relational entity dumps to mobile clients is prohibited.

Strict Prohibition of Cross-BFF Communication [INV-BFF-02]
  A Backend-for-Frontend must never invoke or query another BFF service.
  All data must be sourced directly from backend domain services or shared event topics.

Single Network Roundtrip Guarantee [INV-BFF-03]
  The Mobile BFF must assemble complete mobile screen state in a single client network roundtrip.
  Requiring mobile applications to chain multiple sequential HTTP requests for one screen is prohibited.

Ownership and Handoffs

Traceability

Verification

No validator was supplied, so no command was run.

Reviewer self-check against BFF architecture standards:

• Client-Centricity: PASS. Dedicated to mobile form factor; delivers tailored 38 KB screen view model.
• Aggregation Efficiency: PASS. Parallel gRPC scatter-gather executes in < 25 ms, meeting the 65 ms SLA.
• Resilience: PASS. Non-critical rewards timeouts degrade gracefully without blocking balance display.
• Markdown Hygiene: PASS. Native Markdown syntax strictly adheres to rule_markdown.md.

Open Decisions

• DEC-BFF-01: David O'Reilly to determine whether HTTP/3 (QUIC) should be enabled on edge ingress proxies to further improve mobile reconnection performance over unstable cellular connections (Owner: David O'Reilly).

Next steps

1. Mobile Engineering squad initializes the mobile-banking-bff Fastify service repository.
2. Platform team sets up gRPC client stubs and mutual TLS certificates for downstream core banking services.
3. Conduct staging performance drill validating 18,000 TPS throughput and asserting 38 KB payload constraints.

---

skill: bff-architect

Mobile Banking BFF — Fitness Self-Check [BFF-MOB-FIT-001]

Summary

This fitness self-check evaluates the Backend-for-Frontend (BFF) architecture against three critical red-capable domain failure probes: shared mutable ownership, leaky abstraction, and implicit coupling. All targeted probes pass by design construction. A self-check is supporting evidence, never the authoritative gate. Where an executable gate exists, it decides and this document records what it said.

Detailed Description

Residual Risk

• Cellular network IP hopping during train travel may force occasional TLS renegotiations on client connections. Accepted by David O'Reilly with client-side retry idempotency headers.

Traceability

Verification

No validator was supplied, so no command was run.

Open Decisions

None.

Next steps

1. Platform Security team incorporates synthetic probes FIT-1, FIT-2, and FIT-3 into the master CI pull-request validation workflow.
2. Mobile team verifies payload size linter enforcing <= 45 KB in pre-commit git hooks.
3. Conduct staging resilience drill simulating backend service timeout to confirm graceful degradation.