# AWS event-driven fintech interview preparation

This package is for a senior backend fintech interview. It is a reasoning and
design system, not an AWS trivia guide: start from repository evidence,
requirements, authority, and failure boundaries; then select services, size the
system, and explain how it is recovered and operated. It covers the reviewed
sample repository, event-driven AWS architecture, financial correctness,
capacity/cost, operations, and interview practice. It does not promise a
production design without workload discovery, experiments, security review, and
game days.

## Reader index

| ID | Entry | Outcome |
|---|---|---|
| IDX01 | [Repository case study](01-repository-case-study.md) | Diagnose concrete dual writes, acknowledgement gaps, unsafe money, projection, and recovery behavior in the sample. |
| IDX02 | [Event-driven foundations](02-event-driven-foundations.md) | Choose event vocabulary, asynchronous boundaries, identities, and guarantee scope before services. |
| IDX03 | [AWS messaging decision guide](03-aws-messaging-decision-guide.md) | Select messaging, streaming, workflow, and compute from explicit requirement axes. |
| IDX04 | [Kinesis deep dive](04-kinesis-deep-dive.md) | Reason about partition keys, order, lag, poison records, resharding, and safe replay. |
| IDX05 | [DynamoDB deep dive](05-dynamodb-deep-dive.md) | Design from access patterns, physical lanes, exact money, authority, indexes, and rebuild. |
| IDX06 | [CQRS, event sourcing, and sagas](06-cqrs-event-sourcing-and-sagas.md) | Place CQRS on a spectrum and close atomicity, projection, saga, and economic boundaries. |
| IDX07 | [Reliability and correctness](07-reliability-and-correctness.md) | Assign retry ownership, quarantine, reconciliation, headroom, and tested recovery objectives. |
| IDX08 | [Fintech trading architecture](08-fintech-trading-architecture.md) | Separate order, execution, ledger, clearing, settlement, and projection authorities. |
| IDX09 | [Cost, performance, and capacity](09-cost-performance-and-capacity.md) | Recompute capacity, complete cost, unit economics, sensitivities, and infeasible branches. |
| IDX10 | [Observability, security, and operations](10-observability-security-and-operations.md) | Turn business outcomes and authority gaps into SLOs, alarms, runbooks, evidence, and exercises. |
| IDX11 | [Interview drills](11-interview-drills.md) | Practice requirement-first answers, failure analysis, quantitative work, and design narration. |
| IDX12 | [Quick reference](quick-reference.md) | Recall deciding ratios, correctness boundaries, recovery duties, formulas, and last-day prompts. |
| IDX13 | [Sources](sources.md) | Trace each used source through the authoritative ledger to claims and consuming chapters. |

## Dependency-aware order

Use this dependency chain rather than treating the filenames as equal-weight
reading: **repository diagnosis → foundations → service selection → correctness
and recovery → fintech authority → cost and operations → drills**.

1. Chapters 01–02 establish the observed problem and vocabulary. Without those,
   a service choice tends to hide the actual acknowledgement or business-effect
   boundary.
2. Chapter 03 is the branching service guide; Chapters 04–05 supply the Kinesis
   and DynamoDB mechanics needed to defend its most important branches.
3. Chapters 06–07 put correctness before architecture: authority, atomic
   publication, idempotent effects, rebuild, retries, headroom, and DR.
4. Chapter 08 applies those controls to order, fill, ledger, clearing, and
   settlement responsibilities; it must not be learned as a service diagram.
5. Chapters 09–10 test whether the design is feasible, economic, observable,
   secure, and recoverable.
6. Chapter 11 and the quick reference are retrieval practice after the reasoning
   chain, not substitutes for it.

## One-day path

Use this only when one elapsed eight-hour day is available. Seven 60-minute
study sessions are separated by six 10-minute breaks.

<!-- one-day:start -->
Study **420 min** + breaks **60 min** = elapsed **480 min**.

| Session | Minutes | Activity | Exact assignment |
|---|---:|---|---|
| ON-S1 | 60 | read | repository Chapter 01; write the five highest-risk boundaries; read foundations Chapter 02 decision sequence |
| ON-S2 | 60 | read | messaging Chapter 03, then Kinesis Chapter 04 mental model, partition key, consumer, and recovery sections |
| ON-S3 | 60 | read | DynamoDB Chapter 05 access-pattern, transaction/idempotency, authority, amplification, and rebuild sections |
| ON-S4 | 60 | skim | CQRS Chapter 06 spectrum and outbox/rebuild/saga contracts; reliability Chapter 07 retry, quarantine, reconciliation, and DR |
| ON-S5 | 60 | skim | fintech Chapter 08 selected RSP01/RSP03/RSP04/RSP06/RSP08–RSP10 and INV01/INV03/INV05–INV10 rows (≤1,500 visible words); cost Chapter 09 “Optimization sequence”, “Six fintech unit economics”, and “Decision table” (≤800 words); operations Chapter 10 “Failure recovery cost and poor-fit analysis” plus IQ01/IQ06/IQ11 (≤1,000 visible words); write one failure/cost/security/recovery sentence |
| ON-S6 | 60 | recompute | drills Chapter 11 QE01–QE04; write units, zero/no-spare branches, and one sensitivity for each |
| ON-S7 | 60 | answer aloud | drills Chapter 11 SC01–SC10, FA01–FA06, and SD01–SD04 selection; use quick-reference cards; review misses against detailed chapters |
<!-- one-day:end -->

## Three-day path

This is the recommended default. Each day has three 70-minute study sessions and
two 15-minute breaks: 210 study minutes + 30 break minutes = 240 elapsed minutes.
Dependencies accumulate; do not start Day 3 as isolated memorization.

<!-- three-day:start -->
Study **630 min** + breaks **90 min** = elapsed **720 min**.

| Session | Minutes | Activity | Exact assignment |
|---|---:|---|---|
| TH-D1S1 | 70 | read | repository Chapter 01 end to end; record authority, dual-write, acknowledgement, exact-money, and recovery findings |
| TH-D1S2 | 70 | read | foundations Chapter 02 and messaging Chapter 03 requirement axes; explain every rejected alternative |
| TH-D1S3 | 70 | answer aloud | drills SC01–SC10; use quick-reference service cards; review misses in Chapters 02–03 |
| TH-D2S1 | 70 | read | Kinesis Chapter 04: key/order, producer/consumer, lag, poison, reshard, replay and recovery |
| TH-D2S2 | 70 | read | DynamoDB Chapter 05: APs, physical lanes, transactions, durable receipts, exact money, authority, global recovery |
| TH-D2S3 | 70 | recompute | cost Chapter 09 Kinesis/DynamoDB and drills Chapter 11 QE01–QE02; include hot-lane and no-safe-drain branches |
| TH-D3S1 | 70 | skim | CQRS Chapter 06 “CQRS is a five-step spectrum” plus “Architecture contract shared by all variants” (≤2,100 words); reliability Chapter 07 “Four identities, four stopping boundaries” plus “Ordering and concurrency repair” (≤1,700 words); fintech Chapter 08 selected RSP04/RSP06/RSP08–RSP10 and INV03/INV06–INV10 rows (≤1,200 visible words); note one authority, duplicate, rebuild, recovery, and cost boundary |
| TH-D3S2 | 70 | skim | operations Chapter 10 evidence/runbooks; recompute cost Chapter 09 CQRS/compute/unit economics and drills Chapter 11 QE03–QE04 |
| TH-D3S3 | 70 | review misses | answer aloud drills Chapter 11 FA01–FA06, SD01–SD04, and RW01–RW06; finish with quick-reference checklist and correct weak answers in source chapters |
<!-- three-day:end -->

**Recommended default: three-day path.** For this senior backend fintech
interview, protect the repository diagnosis, messaging/Kinesis, DynamoDB, CQRS,
reliability, trading correctness, quantitative work, and drills in that order.
If time is lost, skim secondary comparisons; never skip authority, financial
invariants, recovery arithmetic, or answer-aloud practice.

## Seven-day path

Each day has a 50-minute session, a 15-minute break, and a 55-minute session:
105 study minutes + 15 break minutes = 120 elapsed minutes. Every day ends with
retrieval practice without notes. The two named design/drill sessions are the
minimum; repeat weak ones.

<!-- seven-day:start -->
Study **735 min** + breaks **105 min** = elapsed **840 min**.

| Session | Minutes | Activity | Exact assignment |
|---|---:|---|---|
| SE-D1S1 | 50 | read | repository Chapter 01 evidence and repair levels |
| SE-D1S2 | 55 | answer aloud | retrieval practice: narrate repository authority, dual-write, acknowledgement, money, and projection failures |
| SE-D2S1 | 50 | read | foundations Chapter 02 boundary vocabulary and guarantees |
| SE-D2S2 | 55 | answer aloud | retrieval practice: messaging Chapter 03 questions and drills Chapter 11 SC01–SC03 |
| SE-D3S1 | 50 | read | Kinesis Chapter 04 key, order, producer, consumer, poison, and recovery |
| SE-D3S2 | 55 | recompute | retrieval practice: drills Chapter 11 QE01 shards, hot key, spare, drain, and no-safe-drain |
| SE-D4S1 | 50 | read | DynamoDB Chapter 05 access patterns, lanes, exact money, authority, and rebuild |
| SE-D4S2 | 55 | recompute | retrieval practice: drills Chapter 11 QE02 rounding, transaction/GSI work, lane feasibility, and replica topology |
| SE-D5S1 | 50 | read | CQRS Chapter 06 spectrum plus reliability Chapter 07 retry/reconcile/DR |
| SE-D5S2 | 55 | answer aloud | retrieval practice and design/drill session 1: drills Chapter 11 SD01 plus CQRS/reliability follow-ups |
| SE-D6S1 | 50 | read | fintech Chapter 08 order/fill/reservation/ledger/settlement authorities and operations Chapter 10 evidence |
| SE-D6S2 | 55 | answer aloud | retrieval practice and design/drill session 2: drills Chapter 11 SD03–SD04, security, RTO/RPO, and reconciliation |
| SE-D7S1 | 50 | recompute | cost Chapter 09 and drills Chapter 11 QE03–QE04: fan-out work, complete-cost CQRS, compute, unit economics |
| SE-D7S2 | 55 | review misses | retrieval practice: quick-reference last-day checklist; answer weak drills aloud and route each miss to its chapter |
<!-- seven-day:end -->

## Vocabulary and notation

- **Claim/source IDs:** `C…` is controlled wording, `F…` is stable foundational
  evidence, `A…` is dated mutable first-party AWS evidence, and `R…` is a
  non-authoritative local cross-reference. Chapter-local IDs such as `AP`,
  `RSP`, `INV`, `FSR`, and `SLO` route to reviewed contracts.
- **Planning assumptions:** explicit workload, SLO, topology, team, and price
  inputs used to test a decision. They are not vendor limits or production
  promises.
- **`Inference:`** a conclusion derived from named premises; it is not an AWS
  guarantee. Challenge it when the premises change.
- **Authority versus projection:** authority accepts and adjudicates a command or
  financial fact; a projection is derived, freshness-bounded, disposable, and
  rebuildable. A projection cannot authorize money, an order transition, or
  settlement.
- **Boundary-scoped exactly-once:** name the operation, identity, scope, time
  window, and side effect. Broker deduplication, a transaction, or a checkpoint
  does not by itself prove one externally visible business effect.