Fraud Detection Pipeline — System Design Academy

Complete System Design Tutorial

The design separates the latency-sensitive online path, the correctness-critical durable transition, and asynchronous projection or repair work. Each chapter below explains one decision shown in the diagrams.

1. Requirements and scope

Functional requirements

Execute the primary fraud detection pipeline workflow and return an observable result.
Expose the concrete operations shown below with stable retry identities.
Allow operators to inspect, disable, repair, or replay ambiguous state.
Keep derived analytics and notifications off the correctness-critical path.

Non-functional requirements

Design for 100K events/s · decision < 100 ms · low false-positive budget.
Protect the named business invariant under concurrency and partial failure.
Scale by an explicit aggregate/partition key without one global mutex.
Degrade intentionally when cache, queue, store, or regional ownership is unavailable.

Primary invariant: each accepted fraud detection pipeline operation has one stable result and retries do not repeat its business effect.

↑ top

2. Capacity assumptions

State assumptions before choosing technology. The supplied workload is 100K events/s · decision < 100 ms · low false-positive budget. Convert it into peak throughput, retained records, bandwidth, and failure budget; then connect each number to a component.

Dimension	Working estimate	Design implication
Traffic	100K events/s · decision < 100 ms · low false-positive budget	Separate online compute, durable state, and asynchronous consumers.
Peak factor	5–10× average	Admission control, autoscaling headroom, and hot-key protection.
Stored state	write rate × retention × record bytes	Partition count, indexes, compaction, backup, and archival tier.
Network	peak operations × payload size	Batching, compression, edge placement, and egress budget.
Availability	99.9–99.99%	Multi-zone replicas, tested failover, and dependency error budgets.
Recovery	RPO/RTO by invariant	Ack boundary, replay retention, reconciliation, and operator runbooks.

Do not cargo-cult the numbers: show the arithmetic, name the uncertain variable, and say which production metric will replace the interview estimate.

↑ top

3. API and data model

POST /v1/decisionssynchronous risk decision

POST /v1/labelschargeback/review outcome

PUT /v1/rules/{id}versioned policy

POST /v1/models/{version}/shadowcompare only

Authoritative records and indexes

decision(decision_id, subject_id, outcome, score, reasons, policy_version)

feature_snapshot(decision_id, names, values, event_times)

case(case_id, decision_id, queue, state, reviewer)

label(decision_id, label, observed_at, source)

Command and event envelopes

# Command envelope
operation: POST /v1/decisions
request_id: req_01J...
idempotency_key: idem_7f2...
aggregate_key: fraud-detection_123
expected_version: 17
payload: {...}

# Durable event
event_id: evt_01J...
aggregate_key: fraud-detection_123
event_type: fraud_detection_committed
schema_version: 3
trace_id: tr_8ac...

The authoritative aggregate and idempotency record share a transactional or conditional-write boundary. Event streams, search indexes, caches, analytics, and dashboards are rebuildable projections.

↑ top

4. Why this architecture and what it costs

The boxes are not a shopping list. Each boundary exists because it owns a different invariant, scaling dimension, data lifecycle, security policy, or failure domain. The table makes those choices explicit and shows when the design should remain simpler.

Component	Owned responsibility	Why separate it	Rejected alternative	Cost and evolution trigger
Transaction API	Owns authentication, validation, quotas, deadlines, and routing. In this design it participates in: Resolve entity/account/device keys and fetch streaming/profile features in parallel.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents the external caller from also owning the contract and failure behavior of Decision Gateway.	Put protocol, authentication, quotas, and business transitions in one process. Acceptable at low scale, but it couples external traffic risk to domain releases. Here that simpler choice would couple “Online feature missing at decision time” to the external caller.	Costs another hop, policy distribution, and a high-blast-radius fleet. Introduce it when many clients or services need one protocol and protection boundary. Measure the latency, saturation, lag, or mismatch at the Transaction API boundary before extracting or scaling it.
Decision Gateway	Owns authentication, validation, quotas, deadlines, and routing. In this design it participates in: Evaluate deterministic rules and model under a fixed feature/model version budget.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Transaction API from also owning the contract and failure behavior of Streaming Feature Engine.	Put protocol, authentication, quotas, and business transitions in one process. Acceptable at low scale, but it couples external traffic risk to domain releases. Here that simpler choice would couple “Adversary probes threshold” to Transaction API.	Costs another hop, policy distribution, and a high-blast-radius fleet. Introduce it when many clients or services need one protocol and protection boundary. Measure the latency, saturation, lag, or mismatch at the Decision Gateway boundary before extracting or scaling it.
Streaming Feature Engine	Preserves durable ordering, replay, and consumer checkpoints. In this design it participates in: Combine outputs through policy thresholds and return allow/challenge/review/deny.	Decouples bursts and failure from user acknowledgment. It prevents Decision Gateway from also owning the contract and failure behavior of Online Feature Store.	Call every downstream consumer synchronously. That couples user latency and availability to analytics, notification, indexing, and audit dependencies. Here that simpler choice would couple “Manual-review backlog grows” to Decision Gateway.	Costs eventual consistency, schema evolution, replay tooling, idempotent consumers, and lag operations. Earn it when bursts or downstream failure must not block acknowledgment. Measure the latency, saturation, lag, or mismatch at the Streaming Feature Engine boundary before extracting or scaling it.
Online Feature Store	Owns authoritative versioned state or immutable evidence. In this design it participates in: Persist feature snapshot, reasons, and versions before asynchronous case routing.	Defines the source of truth and commit boundary. It prevents Streaming Feature Engine from also owning the contract and failure behavior of Profile Store.	Let each service persist its own interpretation. That removes one clear commit boundary and makes retries, support, and reconciliation ambiguous. Here that simpler choice would couple “Chargeback label arrives months later” to Streaming Feature Engine.	Costs schema migration, replication, backup, and shard ownership. Split storage only when access patterns, retention, or compliance boundaries materially differ. Measure the latency, saturation, lag, or mismatch at the Online Feature Store boundary before extracting or scaling it.
Profile Store	Owns authoritative versioned state or immutable evidence. In this design it participates in: Join delayed labels, monitor drift/calibration, and shadow new models before promotion.	Defines the source of truth and commit boundary. It prevents Online Feature Store from also owning the contract and failure behavior of Rules Engine.	Let each service persist its own interpretation. That removes one clear commit boundary and makes retries, support, and reconciliation ambiguous. Here that simpler choice would couple “New model shifts false-positive rate” to Online Feature Store.	Costs schema migration, replication, backup, and shard ownership. Split storage only when access patterns, retention, or compliance boundaries materially differ. Measure the latency, saturation, lag, or mismatch at the Profile Store boundary before extracting or scaling it.
Rules Engine	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Resolve entity/account/device keys and fetch streaming/profile features in parallel.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Profile Store from also owning the contract and failure behavior of Model Serving.	Keep this behavior inside an adjacent service. Prefer that simpler option until independent scale, ownership, policy, or failure isolation is measurable. Here that simpler choice would couple “Online feature missing at decision time” to Profile Store.	Costs an interface, deployment, telemetry, ownership, and failure mode. Keep it in-process first; extract when independent scale or reliability pays for the boundary. Measure the latency, saturation, lag, or mismatch at the Rules Engine boundary before extracting or scaling it.
Model Serving	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Evaluate deterministic rules and model under a fixed feature/model version budget.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Rules Engine from also owning the contract and failure behavior of Ensemble / Policy.	Keep this behavior inside an adjacent service. Prefer that simpler option until independent scale, ownership, policy, or failure isolation is measurable. Here that simpler choice would couple “Adversary probes threshold” to Rules Engine.	Costs an interface, deployment, telemetry, ownership, and failure mode. Keep it in-process first; extract when independent scale or reliability pays for the boundary. Measure the latency, saturation, lag, or mismatch at the Model Serving boundary before extracting or scaling it.
Ensemble / Policy	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Combine outputs through policy thresholds and return allow/challenge/review/deny.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Model Serving from also owning the contract and failure behavior of Decision Service.	Embed policy in every caller. Rules then drift by deployment version and emergency changes require fleet-wide redeployment. Here that simpler choice would couple “Manual-review backlog grows” to Model Serving.	Costs an interface, deployment, telemetry, ownership, and failure mode. Keep it in-process first; extract when independent scale or reliability pays for the boundary. Measure the latency, saturation, lag, or mismatch at the Ensemble / Policy boundary before extracting or scaling it.
Decision Service	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Persist feature snapshot, reasons, and versions before asynchronous case routing.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Ensemble / Policy from also owning the contract and failure behavior of Manual Review Queue.	Keep this behavior inside an adjacent service. Prefer that simpler option until independent scale, ownership, policy, or failure isolation is measurable. Here that simpler choice would couple “Chargeback label arrives months later” to Ensemble / Policy.	Costs an interface, deployment, telemetry, ownership, and failure mode. Keep it in-process first; extract when independent scale or reliability pays for the boundary. Measure the latency, saturation, lag, or mismatch at the Decision Service boundary before extracting or scaling it.
Manual Review Queue	Preserves durable ordering, replay, and consumer checkpoints. In this design it participates in: Join delayed labels, monitor drift/calibration, and shadow new models before promotion.	Decouples bursts and failure from user acknowledgment. It prevents Decision Service from also owning the contract and failure behavior of Case Management.	Call every downstream consumer synchronously. That couples user latency and availability to analytics, notification, indexing, and audit dependencies. Here that simpler choice would couple “New model shifts false-positive rate” to Decision Service.	Costs eventual consistency, schema evolution, replay tooling, idempotent consumers, and lag operations. Earn it when bursts or downstream failure must not block acknowledgment. Measure the latency, saturation, lag, or mismatch at the Manual Review Queue boundary before extracting or scaling it.
Case Management	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Resolve entity/account/device keys and fetch streaming/profile features in parallel.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Manual Review Queue from also owning the contract and failure behavior of Decision Audit Store.	Keep this behavior inside an adjacent service. Prefer that simpler option until independent scale, ownership, policy, or failure isolation is measurable. Here that simpler choice would couple “Online feature missing at decision time” to Manual Review Queue.	Costs an interface, deployment, telemetry, ownership, and failure mode. Keep it in-process first; extract when independent scale or reliability pays for the boundary. Measure the latency, saturation, lag, or mismatch at the Case Management boundary before extracting or scaling it.
Decision Audit Store	Owns authoritative versioned state or immutable evidence. In this design it participates in: Evaluate deterministic rules and model under a fixed feature/model version budget.	Proves correctness and supports controlled recovery. It prevents Case Management from also owning the contract and failure behavior of Label Pipeline.	Let each service persist its own interpretation. That removes one clear commit boundary and makes retries, support, and reconciliation ambiguous. Here that simpler choice would couple “Adversary probes threshold” to Case Management.	Costs schema migration, replication, backup, and shard ownership. Split storage only when access patterns, retention, or compliance boundaries materially differ. Measure the latency, saturation, lag, or mismatch at the Decision Audit Store boundary before extracting or scaling it.
Label Pipeline	Executes idempotent asynchronous work under lease and retry policy. In this design it participates in: Combine outputs through policy thresholds and return allow/challenge/review/deny.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Decision Audit Store from also owning the contract and failure behavior of Model Monitor.	Perform the work before acknowledging the caller. This gives simpler consistency but unbounded tail latency and fragile retries. Here that simpler choice would couple “Manual-review backlog grows” to Decision Audit Store.	Costs leases, checkpoints, retries, DLQs, and lag operations. Separate once execution time or burstiness no longer fits the request deadline. Measure the latency, saturation, lag, or mismatch at the Label Pipeline boundary before extracting or scaling it.
Model Monitor	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Persist feature snapshot, reasons, and versions before asynchronous case routing.	Proves correctness and supports controlled recovery. It prevents Label Pipeline from also owning the contract and failure behavior of Shadow Deployment.	Infer correctness from request logs. Logs do not prove durable business state, detect silent drift, or safely repair ambiguous outcomes. Here that simpler choice would couple “Chargeback label arrives months later” to Label Pipeline.	Costs an interface, deployment, telemetry, ownership, and failure mode. Keep it in-process first; extract when independent scale or reliability pays for the boundary. Measure the latency, saturation, lag, or mismatch at the Model Monitor boundary before extracting or scaling it.
Shadow Deployment	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Join delayed labels, monitor drift/calibration, and shadow new models before promotion.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Model Monitor from also owning the contract and failure behavior of operators and repair tooling.	Keep this behavior inside an adjacent service. Prefer that simpler option until independent scale, ownership, policy, or failure isolation is measurable. Here that simpler choice would couple “New model shifts false-positive rate” to Model Monitor.	Costs an interface, deployment, telemetry, ownership, and failure mode. Keep it in-process first; extract when independent scale or reliability pays for the boundary. Measure the latency, saturation, lag, or mismatch at the Shadow Deployment boundary before extracting or scaling it.

Decision 01

Chosen boundary vs simpler monolith

Compute online features from keyed streams and join them with durable profile features under strict time budgets.

A single deployable with one transactional store is the right starting point for a small fraud detection pipeline workload. The proposed boundary becomes justified when compute online features from keyed streams and join them with durable profile features under strict time budgets. Separating it contains the named failure “Online feature missing at decision time” without placing unrelated work on the critical path.

Decision 02

Correctness vs availability

Run deterministic rules alongside models; record feature values, versions, and reasons for every decision.

The design coordinates only where each accepted fraud detection pipeline operation has one stable result and retries do not repeat its business effect. Everywhere else it accepts bounded staleness or asynchronous propagation. The alternative—strong coordination for every projection—simplifies the mental model but increases tail latency and turns partitions into broad write unavailability.

Decision 03

Operational cost and evolution

Use shadow deployment, delayed labels, and human-review outcomes to monitor drift and retrain safely.

This choice adds dashboards, ownership, schema compatibility, capacity planning, and recovery runbooks. Start with the minimum topology that supports “Combine outputs through policy thresholds and return allow/challenge/review/deny.” and extract another service only when measured throughput, blast radius, data lifecycle, or team ownership changes independently.

Why this design: the synchronous path contains only the decisions required to protect each accepted fraud detection pipeline operation has one stable result and retries do not repeat its business effect. Durable projections, notifications, indexing, analytics, and repair are isolated so their load or failure cannot corrupt the acknowledged operation.

Correctness boundary

Compute online features from keyed streams and join them with durable profile features under strict time budgets.

Scale boundary

Run deterministic rules alongside models; record feature values, versions, and reasons for every decision.

Operational boundary

Use shadow deployment, delayed labels, and human-review outcomes to monitor drift and retrain safely.

↑ top

5. Critical flow and algorithm

Resolve entity/account/device keys and fetch streaming/profile features in parallel.

Evaluate deterministic rules and model under a fixed feature/model version budget.

Combine outputs through policy thresholds and return allow/challenge/review/deny.

Persist feature snapshot, reasons, and versions before asynchronous case routing.

Join delayed labels, monitor drift/calibration, and shadow new models before promotion.

Retry-safe mutation skeleton

identity = authenticate(request)
prior = idempotency.get(identity, request.key)
if prior: return prior.result

current = authoritative_store.read(partition_key)
next_state, events = domain.transition(current, request)
authoritative_store.compare_and_set(current.version, next_state)
idempotency.commit(request.key, next_state.result)
outbox.append(events)
return next_state.result

Exactly-once business effect: the network can duplicate requests. Stable identities, conditional state transitions, immutable event IDs, and reconciliation make the user-visible effect happen once.

↑ top

6. Caching, hot keys, and backpressure

Cache policy

Add a cache only for a measured repeated read; do not place mutable correctness state in a disposable layer.

Hot partition or tenant

Measure per-key and per-tenant skew. Replicate safe reads, isolate the offender, salt only when ordering permits, and enforce budgets.

Backpressure

Bound queues, buffers, connection pools, and concurrency. Shed low-priority work and scale from lag/oldest-item age rather than CPU alone.

Stampede and retry storms

Use single-flight/coalescing, jittered expiry, bounded retries, exponential backoff, and circuit breakers so one dependency failure does not multiply traffic.

Cache outage rule: never allow every miss to fall through simultaneously. Use local protection, request coalescing, circuit breakers, bounded pools, and controlled fallback.

↑ top

7. Partitioning and consistency

partition = stable_hash(primary_or_aggregate_key) → shard owner
write = conditional_update(expected_version)
read_projection = versioned and rebuildable

Strong boundary

each accepted fraud detection pipeline operation has one stable result and retries do not repeat its business effect. Use a uniqueness constraint, serializable/conditional write, expected version, or fencing token at this boundary.

Eventual boundary

Analytics, notifications, search, read models, and operational dashboards may lag when freshness is observable and replay exists.

Operation	Consistency	Reason
Authoritative mutation	Conditional / serializable	Protect uniqueness, ownership, money, inventory, order, or legal state transition.
Read after own write	Primary, session token, or cache warm	The caller must observe the result it just committed.
Ordinary read projection	Bounded stale where safe	Replicas and caches improve latency when version/freshness is visible.
Analytics and operations	Eventual + replay	Freshness is less important than isolating the online path.

Route every mutation for one aggregate to one owner/partition epoch.
Use cursors rather than offsets for changing ordered collections.
Retain tombstones until replication and repair windows close.
Move shard ownership through versioned maps and dual-read verification.

↑ top

8. Failure drills and recovery

Failure	Runtime behavior	Recovery proof
Online feature missing at decision time	Stop unsafe progress, serve only explicitly safe degraded behavior, and recover from durable state.	Reconciliation returns zero invariant gaps; lag/replica health and user-visible SLO return to baseline.
Adversary probes threshold	Stop unsafe progress, serve only explicitly safe degraded behavior, and recover from durable state.	Reconciliation returns zero invariant gaps; lag/replica health and user-visible SLO return to baseline.
Manual-review backlog grows	Isolate the key/tenant/priority, apply admission control, and preserve durable work.	Reconciliation returns zero invariant gaps; lag/replica health and user-visible SLO return to baseline.
Chargeback label arrives months later	Stop unsafe progress, serve only explicitly safe degraded behavior, and recover from durable state.	Reconciliation returns zero invariant gaps; lag/replica health and user-visible SLO return to baseline.
New model shifts false-positive rate	Stop unsafe progress, serve only explicitly safe degraded behavior, and recover from durable state.	Reconciliation returns zero invariant gaps; lag/replica health and user-visible SLO return to baseline.

Recovery checkpoint

# Recovery checkpoint
scenario: Online feature missing at decision time
operation_id: op_01J...
aggregate_key: fraud-detection_123
last_durable_version: 18
projection_checkpoint: partition-07@offset-482901
retry_policy: same_identity / bounded_backoff
reconciliation_status: pending
owner: Shadow Deployment

Every dependency needs a timeout, retry budget, circuit policy, and explicit fail-open, fail-closed, stale, or read-only mode. “Highly available” is incomplete until degraded behavior is named.

↑ top

9. Security, regions, and observability

Security and abuse

Authenticate at ingress and authorize the exact resource/tenant on every operation.
Encrypt transport and durable data; tokenize or isolate sensitive material.
Rate-limit by tenant, actor, resource, and network; detect enumeration and anomalous fan-out.
Redact secrets and personal data before logs or analytical storage.
Audit privileged mutations, support access, policy changes, and forced recovery.

Signals that prove correctness

User-visible success rate and p50/p95/p99 latency.
Timeouts, circuit state, saturation, and replica health by dependency.
Queue lag, oldest-item age, retries, and dead-letter growth.
Conflict, duplicate-suppression, and reconciliation mismatch counts.
Business drill: Online feature missing at decision time — alert before the user-visible invariant is violated.

Regional evolution

Begin multi-zone with tested failover, backups, and reconciliation.
Add global stateless ingress and regional read projections.
Keep one write owner per aggregate or partition where coordination is expensive.
Move ownership with epochs/fencing; avoid silent simultaneous writers.
Use expand/contract schemas, canaries, shadow reads, and reversible cutovers.

↑ top

10. Interviewer follow-ups

What happens if feature storage is down?

Compute online features from keyed streams and join them with durable profile features under strict time budgets. Walk the concrete lifecycle step “Resolve entity/account/device keys and fetch streaming/profile features in parallel.” Then test it against “Online feature missing at decision time”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do you explain a decision?

Run deterministic rules alongside models; record feature values, versions, and reasons for every decision. Walk the concrete lifecycle step “Evaluate deterministic rules and model under a fixed feature/model version budget.” Then test it against “Adversary probes threshold”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do chargebacks become labels?

Use shadow deployment, delayed labels, and human-review outcomes to monitor drift and retrain safely. Walk the concrete lifecycle step “Combine outputs through policy thresholds and return allow/challenge/review/deny.” Then test it against “Manual-review backlog grows”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do you deploy a new model?

Compute online features from keyed streams and join them with durable profile features under strict time budgets. Walk the concrete lifecycle step “Persist feature snapshot, reasons, and versions before asynchronous case routing.” Then test it against “Chargeback label arrives months later”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do adversaries affect features?

Run deterministic rules alongside models; record feature values, versions, and reasons for every decision. Walk the concrete lifecycle step “Join delayed labels, monitor drift/calibration, and shadow new models before promotion.” Then test it against “New model shifts false-positive rate”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

Answer pattern: state the invariant, choose a mechanism, walk one failure, name the recovery signal, and explain the cost of the alternative.

↑ top

11. How to present this in 45 minutes

Time	What to cover
0–5 min	Clarify scope, workload, SLOs, and the correctness invariant.
5–10 min	Estimate throughput/storage and define concrete APIs and records.
10–20 min	Draw online, durable-write, and asynchronous/repair paths.
20–30 min	Explain the exact lifecycle, partition key, and concurrency control.
30–38 min	Discuss caching, hot keys, replication, and backpressure.
38–43 min	Run failure drills, security, observability, and regional evolution.
43–45 min	Summarize the five most important choices and trade-offs.

Final decision summary

Invariant	each accepted fraud detection pipeline operation has one stable result and retries do not repeat its business effect.
Partitioning	Stable aggregate/key hash with versioned ownership and measured hot-key isolation.
Commit	Expected version or uniqueness at the authoritative state boundary.
Async work	Durable outbox/log, idempotent consumers, checkpoints, and replay.
Recovery	Same operation identity, durable truth lookup, and business reconciliation.

Closing summary

CLOSING “The design protects each accepted fraud detection pipeline operation has one stable result and retries do not repeat its business effect. The online path stays bounded, authoritative state is versioned and partitioned, and asynchronous work is replayable and observable.”

↑ top

Transaction API

Decision Gateway

Streaming Feature Engine

Online Feature Store

Complete System Design Tutorial

1. Requirements and scope

Functional requirements

Non-functional requirements

2. Capacity assumptions

3. API and data model

Authoritative records and indexes

Command and event envelopes

4. Why this architecture and what it costs

Chosen boundary vs simpler monolith

Correctness vs availability

Operational cost and evolution

5. Critical flow and algorithm

Retry-safe mutation skeleton

6. Caching, hot keys, and backpressure

7. Partitioning and consistency

Strong boundary

Eventual boundary

8. Failure drills and recovery

Recovery checkpoint

9. Security, regions, and observability

Security and abuse

Signals that prove correctness

Regional evolution

10. Interviewer follow-ups

11. How to present this in 45 minutes

Final decision summary

Closing summary