The full inventory of what this book will cover, extracted from the source
document and prioritised. It is the build's work queue, and it doubles as a
study checklist : read the list, mark anything you could not explain to a peer
for two minutes, and that marked set is your syllabus regardless of whether the
page exists yet.
Where a topic is still [todo], the source document (§ references are given at
the foot of each chapter index) carries the outline version. The expansion adds
the mechanics, the worked example, the production evidence and the follow-up
answers.
Inventory of every distinct term, drill, design question and claim extracted from
the source document, grouped by target chapter, prioritised, and tagged with
status. This file is the work queue. Never re-plan from scratch; take the next
P1/P2 items from here.
Status: [todo] not written · [wip] in progress · [done] at the full
ten-point quality bar.
Priority:
P1 appears in the source's drill bank (§25, §34, §38.8, §39.7) or its
design-question lists (§5.2, §6.6, §15.10)P2 tagged 🔴 (outline only) in the source's Gap Register (§51)P3 tagged 🟡 (partial) in the Gap RegisterP4 polish and relocation of material already at target depth (✅)
Totals: 341 topics · 90 done · 251 remaining. (Session 0 reported 354; that
was an arithmetic error in the per-chapter tally, corrected in session 2. See
state/DECISIONS.md D9.)
P Status Topic
P4 [todo] What gets scored: the six signals, senior vs staff on each P1 [done] The staff-level design interview structure, minute by minute P2 [todo] The rounds nobody prepares for: design doc review, code review, debugging, incident sim, take-home P4 [todo] Company loop formats: Google, Meta, Amazon, Stripe, Netflix, Databricks, Uber, Shopify, AI labs P1 [todo] The architecture deep dive on past work, at three zoom levels P4 [todo] Above-and-beyond differentiators: the artifact, the 30/60/90, the follow-up P1 [todo] Executive communication: BLUF, altitude calibration, signposting, bounded uncertainty P1 [todo] Handling what you do not know (drill 56) P1 [done] Scar-tissue stories and where to deploy them (drill 60) P3 [todo] Whiteboard and remote mechanics; interview-day stamina
P Status Topic
P4 [todo] The latency numbers and the derived facts P1 [todo] Back-of-envelope capacity math: a worked search-service sizing P1 [done] Little's Law and the connection-pool death spiral P1 [done] Queueing theory: why the knee is at 70 percent P4 [todo] The Universal Scalability Law and the coherence term P1 [done] The tail at scale: fan-out amplification, hedged and tied requests P3 [todo] Benchmarking discipline and what microbenchmarks lie about P4 [todo] Paxos: the two phases and the Phase 2a constraint P4 [todo] Raft: state, the five safety properties, pre-vote, membership change P1 [todo] The consistency ladder and the session guarantees P1 [done] Isolation anomalies and write skew under snapshot isolation (drill 28) P3 [todo] Time: Lamport, vector clocks, TrueTime, hybrid logical clocks P3 [todo] CRDTs: the catalog, real deployments, the invariant they cannot save P1 [todo] Consistent hashing, jump hash, Maglev, rendezvous, resharding P1 [done] Idempotency keys and exactly-once as at-least-once plus dedupe
P Status Topic
P1 [done] Cassandra tombstones and the range-scan timeout (drill 25) P4 [todo] LSM trees vs B-trees: the three amplifications P3 [todo] Compaction strategies: STCS, LCS, TWCS, and how to pick P3 [todo] RocksDB tuning: block cache, bloom filters, write stalls P1 [done] DynamoDB single-table design over five access patterns (drill 24) P3 [todo] GSIs vs LSIs, sparse indexes, hot partitions, write sharding P3 [todo] Cassandra data modelling, query-first, worked P3 [todo] Tunable consistency, repair mechanisms, gc_grace_seconds P1 [done] MongoDB shard key selection and the ESR index rule (drill 26) P4 [todo] PostgreSQL MVCC, bloat, autovacuum, XID wraparound P3 [todo] Reading a Postgres query plan P3 [todo] PgBouncer transaction pooling and what it breaks P1 [todo] Online schema change: expand-contract, CONCURRENTLY, gh-ost (drill 66) P1 [done] OpenSearch shard sizing, heap rules, zero-downtime reindex (drill 27) P3 [todo] Iceberg table lifecycle: snapshots, hidden partitioning, small files P1 [todo] Choosing a database: the decision walkthrough
P Status Topic
P1 [done] Kafka exactly-once, end to end (drill 10) P1 [done] flatMap vs concatMap in Project Reactor (drill 11) P1 [done] Why an L4 load balancer breaks gRPC (drill 14) P2 [todo] ISR, min.insync.replicas, acks=all, unclean leader election P2 [todo] Log compaction vs retention; compacted topics as state P2 [todo] Consumer group rebalancing: eager, cooperative sticky, static membership P2 [todo] Consumer lag as the health metric; KEDA lag-based autoscaling P2 [todo] Schema Registry compatibility modes as a design decision P1 [done] Flink event time, watermarks, allowed lateness (drill 15) P2 [todo] Flink state backends, checkpointing, savepoints P2 [todo] Kafka Streams vs Flink vs Spark Structured Streaming P2 [todo] CDC with Debezium and the dual-write problem P2 [todo] Dead letter queues, poison pills, retry topics with backoff tiers P2 [todo] Reactive Streams: Publisher, Subscriber, request(n), backpressure strategies P2 [todo] publishOn vs subscribeOn, and the schedulers P2 [todo] Never block the event loop: the number-one WebFlux production bug P1 [done] Virtual threads vs reactive: the defensible position (drill 12) P1 [done] GraphQL N+1 and DataLoader, in full (drill 13) P2 [todo] Apollo Federation v2 and Netflix DGS P2 [todo] GraphQL caching, persisted queries, depth and complexity limits P2 [todo] Protobuf wire format and compatibility rules P2 [todo] Deadline propagation and cancellation across a call tree P1 [todo] The API decision matrix P1 [todo] Sagas vs two-phase commit, expanded
P Status Topic
P1 [done] Prefill vs decode (drill 1) P1 [done] Diagnosing confidently wrong RAG (drill 2) P3 [todo] The transformer, walked through with tensor shapes P3 [todo] RoPE and ALiBi explained, not named P3 [todo] Tokenisation, BPE, and the multilingual token-count penalty P1 [done] KV cache math, GQA, PagedAttention, prefix caching P1 [done] Continuous batching and a scheduler implementation P3 [todo] FlashAttention, speculative decoding, quantisation, parallelism P3 [todo] The fine-tuning ladder with a worked LoRA example (drill 3) P3 [todo] Sampling strategies with demonstrated outputs P3 [todo] Structured output: JSON mode, constrained decoding, grammars P1 [done] Chunking strategies and why fixed-size is usually wrong P1 [done] Hybrid retrieval and reciprocal rank fusion P1 [todo] Cross-encoder and LLM reranking, and the latency budget P1 [done] Separating retrieval eval from generation eval P1 [done] LLM-as-judge and its three biases P1 [done] Agent evaluation as task-level success rate (drill 4) P1 [todo] Regression gates in CI P1 [done] Cost engineering: levers ranked, with the math (drill 6) P1 [done] OWASP LLM Top 10 and indirect prompt injection (drill 5) P3 [todo] Classical ML you still need: calibration, leakage, drift
P Status Topic
P2 [todo] Context engineering vs prompt engineering P2 [todo] The context pipeline, six stages P1 [done] A context-budget worked example on a 128k window (drill 9) P2 [todo] Lost in the middle, and context rot P1 [todo] Compaction: rolling, hierarchical, structured state extraction (drill 7) P2 [todo] Sub-agent isolation as cost and pollution control P2 [todo] Prompts in git, prompts with tests P2 [todo] ReAct, Plan-and-Execute, Reflexion, router, supervisor, handoff P2 [todo] Tool registry design; tool descriptions as prompt surface P2 [todo] MCP as the tool-exposure standard P1 [todo] Agent failure modes: loops, injection, hallucinated calls, non-idempotent retries P2 [todo] Agent memory: short-term, long-term, episodic, semantic P2 [todo] Tracing, replay and checkpointing P2 [todo] LangGraph-style state machines, with code P1 [done] When a knowledge graph beats a vector index (drill 8) P2 [todo] GraphRAG: extraction prompts, communities, local vs global search P2 [todo] Entity resolution
P Status Topic
P4 [todo] The inverted index, mechanically P3 [todo] Lucene segments, refresh vs flush vs merge P4 [todo] BM25 derived from first principles P3 [todo] Analyzers per language: stemming, lemmatisation, CJK, transliteration P1 [todo] Multilingual index topology: shared vs per-language vs cross-lingual P1 [todo] HNSW internals: M, ef parameters, the memory formula, delete cost P3 [todo] IVF-PQ vs HNSW vs DiskANN vs ScaNN P1 [done] Filtered vector search and the recall cliff P1 [todo] Billion-scale sharding, routing and cold start P1 [todo] The multi-stage ranking funnel P3 [todo] Learning to rank: features and a worked feature set P3 [todo] Judgment collection and a training-data pipeline P1 [todo] NDCG, MRR, recall@k, offline-online correlation P1 [done] Interleaving, position bias and inverse propensity scoring P1 [todo] Two-tower retrieval and feature store parity P3 [todo] Cold start, exploration and bandits P3 [todo] Popularity bias and feedback loops P3 [todo] Embedding freshness and index rebuild without downtime
P Status Topic
P1 [done] CPU limits and CFS throttling (drill 20) P2 [todo] Lambda cold start anatomy and the levers P1 [done] Serverless vs containers cost crossover, with the math (drill 21) P2 [todo] Namespaces and cgroups v2, hands-on P2 [todo] Container security primitives: seccomp, capabilities, read-only root P2 [todo] runc vs gVisor vs Firecracker vs Kata P2 [todo] The reconciliation loop as the Kubernetes mental model P2 [todo] Requests, limits, QoS classes and eviction order P2 [todo] HPA, VPA, Cluster Autoscaler, Karpenter, KEDA P2 [todo] kube-proxy modes and why iptables degrades at scale P2 [todo] NetworkPolicy and service mesh authorization P1 [done] A full pod-debugging transcript (drill 19) P2 [todo] The ndots:5 DNS latency classic P2 [todo] Memory: page cache, faults, huge pages, NUMA, the OOM killer P2 [todo] I/O models: blocking, epoll, io_uring, zero-copy P1 [done] fsync, write barriers, and what "the write returned" means (drill 22) P2 [todo] TCP: TIME_WAIT, accept queues, Nagle, BBR, conntrack exhaustion P1 [done] eBPF for observability, networking and security (drill 23) P3 [todo] USE and RED, and the tooling for each P2 [todo] JVM in a container: heap sizing, GC choice, async-profiler
P Status Topic
P1 [done] Cache stampede on a hot key (drill 16) P1 [done] SSE vs WebSockets for token streaming (drill 18) P2 [todo] WebSocket scaling: sticky routing vs a pub/sub backplane P2 [todo] Connection resource math, ulimit, ephemeral ports, conntrack P2 [todo] Reconnect, resume by last-event-id, client-side dedupe P2 [todo] Cache-Control decided per asset class P2 [todo] stale-while-revalidate and stale-if-error P2 [todo] ETag, Last-Modified, and validator semantics P2 [todo] Vary and the cache-key design that destroys hit rate P2 [todo] CDN tiered caching and origin shield P1 [done] Purge strategies: hard, soft, surrogate keys (drill 17) P2 [todo] Edge compute: what belongs at the edge P2 [todo] Caching a personalised page: shell plus fragments, ESI, streaming SSR P2 [todo] Cache poisoning and unkeyed input P2 [todo] The six cache layers and their patterns P2 [todo] Redis eviction policies, hash slots, hot-key mitigation P2 [todo] Redis persistence, and why Redis is not a database P2 [done] Invalidation: TTL vs event-driven vs versioned keys P2 [todo] Negative caching
P Status Topic
P1 [done] PKCE and the authorization code flow (drill 29) P2 [todo] The OAuth grants that are alive, and the two that are dead P2 [todo] Access, refresh and ID tokens, and what each is not for P2 [todo] JWT validation implemented, with every check justified P1 [done] Token revocation with stateless JWTs (drill 30) P2 [todo] Refresh token rotation with reuse detection P2 [todo] DPoP and mTLS-bound tokens P2 [todo] Token exchange for on-behalf-of chains P2 [todo] The OAuth vulnerability catalog P2 [todo] RBAC to ABAC to ReBAC, with Zanzibar tuples P1 [done] Zero trust explained in 60 seconds, plus the sequencing (drill 32) P2 [todo] Workload identity: SPIFFE/SPIRE, IRSA, no static credentials P2 [todo] SSRF and cloud metadata, exploitation and defence P2 [todo] A STRIDE threat model worked on a real system P2 [todo] Supply chain: SBOM, SLSA, sigstore, dependency confusion P2 [todo] Secrets: dynamic credentials, rotation, scanning history P2 [todo] The security pipeline: SAST, SCA, DAST, IaC, container, admission P1 [done] CVSS vs EPSS vs KEV and the patch SLA (drill 33) P1 [done] The zero-day response runbook, narrated (drill 31)
P Status Topic
P2 [todo] React reconciliation, walked through P2 [todo] Keys, and why index-as-key breaks lists P2 [todo] Hooks rules and the useEffect dependency traps P2 [todo] Concurrent React: startTransition, useDeferredValue, Suspense P2 [todo] Server Components and the "use client" boundary, worked P2 [todo] The state ladder P2 [todo] Redux thunk vs saga vs observable P1 [done] Migrating a Redux store to TanStack Query (drill 34) P2 [todo] Rendering strategy matrix P1 [done] Core Web Vitals and an INP debugging transcript (drill 35) P1 [done] Micro-frontends and Module Federation: the skeptical case (drill 36) P2 [todo] Frontend security: XSS, CSP with nonces, third-party script risk P2 [todo] Accessibility as a design input P2 [todo] Testing ratio: unit, component, E2E, visual regression P2 [todo] Streaming AI UIs: token rendering, thinking states, citations, undo
P Status Topic
P1 [done] Burn-rate alerting (drill 43) P3 [todo] SLI selection: measure closest to the user P3 [todo] SLO targets derived from tolerance P1 [todo] The error budget policy as a signed contract P1 [todo] Composite SLOs and dependency availability math P1 [todo] Why you cannot average percentiles across instances P3 [todo] OpenTelemetry: API, SDK, Collector, OTLP, semantic conventions P3 [todo] A Collector pipeline config, annotated P3 [todo] Head sampling vs tail sampling P1 [done] Cardinality as the cost model, and the policy fix (drill 44) P3 [todo] Exemplars and the modern debugging workflow P3 [todo] Structured logs, trace correlation, retention tiers P3 [todo] Wide events and the observability 2.0 argument P3 [todo] Continuous profiling as the fourth signal P1 [done] Chaos engineering: hypothesis, fault menu, game days (drill 51) P3 [todo] Load testing: open vs closed models, coordinated omission P3 [todo] Incident management: roles, severity triggers, mitigate before diagnose P3 [todo] Blameless postmortems and repeat-incident rate P1 [todo] SLA vs SLO vs SLI: the contractual layer (drill 55) P3 [todo] Per-tenant SLAs and the observability requirement they create
P Status Topic
P1 [done] Canary vs A/B testing (drill 61) P3 [todo] The strategy matrix: recreate, rolling, blue/green, canary, shadow, rings P3 [todo] Traffic shifting and consistent cohorting P1 [todo] Automated canary analysis with a worked scoring example (drill 62) P1 [todo] Bake time, minimum detectable effect, and the ritual canary P3 [todo] What canaries cannot catch P1 [todo] Shadow traffic and side-effect containment (drill 63) P1 [done] Mobile rollout: N-2 support and the kill switch (drill 65) P1 [done] Expand-contract across a three-version window (drill 66) P1 [done] Cache and message compatibility across versions (drill 64) P3 [todo] Draining long-lived connections during a deploy P3 [todo] Rolling stateful services with quorum awareness and PDBs P3 [todo] Trunk-based development and the merge queue P3 [done] Feature flag types, lifecycles and flag debt policy P3 [todo] Testing strategy: pyramid vs trophy, with a position P1 [done] Consumer-driven contract testing, end to end (drill 45) P3 [todo] Property-based and mutation testing P1 [todo] Experimentation: randomisation unit, power analysis, guardrails (drill 49) P3 [todo] Peeking, novelty effects, interference, switchback tests P1 [done] Deploy vs release, and the pipeline as a system (drill 46)
P Status Topic
P4 [todo] SOLID with the mature caveats P4 [todo] The GoF subset that appears P1 [done] Repository, Unit of Work, Specification, and the repository critique (drill 53) P4 [todo] Hexagonal, Clean and Onion as the same idea P1 [done] CQRS: the adoption ladder and where to place a team (drill 54) P4 [todo] The five fixes for CQRS eventual consistency P4 [todo] Event sourcing: store schema, concurrency, snapshots, upcasting P3 [todo] Crypto-shredding for GDPR, and why Kafka is not an event store P4 [todo] Sagas: orchestration vs choreography with compensation code P4 [todo] The transactional outbox with SQL and Debezium config P3 [todo] Vertical slice architecture P3 [todo] The anti-pattern catalog P3 [todo] Refactoring vocabulary: strangler fig, branch by abstraction, seams P3 [todo] Bounded contexts, ubiquitous language, the anti-corruption layer P3 [todo] Aggregates as consistency boundaries P3 [todo] Event storming worked on a real domain P1 [done] Modular monolith vs microservices, and the extraction forces (drill 48) P1 [done] Team Topologies and splitting a team by cognitive load (drill 47) P3 [todo] Decision machinery: one-way doors, DACI, ADRs, C4 P4 [todo] The standards reference card, with critiques
P Status Topic
P4 [done] Distributed rate limiter P4 [done] RAG with document-level access control P3 [done] LLM gateway P3 [done] Multilingual semantic search at 10k QPS P3 [todo] Agent platform with sandboxed tools P3 [todo] Evaluation pipeline that gates CI P3 [todo] Cost attribution across 40 teams P1 [done] News feed: push, pull, hybrid P1 [done] Notification and fan-out system P1 [done] Distributed job scheduler with exactly-once semantics P1 [done] Metrics and observability pipeline P1 [done] Ad click aggregation with dedupe and late events P1 [done] Payments ledger with idempotency and double-entry P1 [done] Ride-hailing dispatch and geo-indexing P1 [todo] Object storage and file sync P1 [done] Ticketing under contention P1 [done] Multi-region active-active key-value store P1 [done] Personalised recommendation serving on a 50 ms budget P1 [todo] Inference batching for one GPU under a latency SLA
P Status Topic
P1 [done] The toxic code reviewer (drill 37) P1 [todo] Building a story portfolio, not answers to questions P1 [done] SCOR in full, and mapping back to STAR (drill 58) P3 [todo] Unblocking: the four blocker types P3 [todo] Growing people: skill matrix, stretch ratio, delegation ladder, SBI P1 [todo] The underperformance sequence and the no-surprises rule P3 [todo] PR review as a leadership lever: SLA, size, taxonomy, order P1 [todo] The AI-era review queue problem (drill 38) P3 [todo] Capacity math and the three-bucket budget P3 [todo] Forecasting with cycle-time percentiles P3 [todo] On-call design: rotation size, page budget, interrupt shield, toil cap P3 [todo] Product partnership: options with costs P3 [todo] Promotions won two quarters early, and the calibration room P3 [todo] Retention and each person's actual motivator P3 [todo] Hiring: scorecard first, defending the bar P3 [todo] Written-first culture, ADRs, managing up P1 [todo] Metrics a lead watches, and the 2026 DORA caveat P1 [done] Missed commitments, three in a row P1 [todo] A PM promises a date without asking you P1 [done] A director wants a date you cannot commit to (drill 39) P1 [done] Two teams building the same service, no authority (drill 41) P1 [done] The reliability investment case to a product-focused VP (drill 40) P1 [done] Promotion when someone is one level of scope short (drill 42) P1 [todo] Inheriting a team with low morale after a reorg P3 [todo] Managing former peers, and glue work P3 [todo] Legacy modernisation as a narrative
P Status Topic
P1 [done] RTO and RPO, extracted and priced (drill 50) P3 [todo] The DR ladder: backup, pilot light, warm standby, active-active P1 [todo] Active-active conflict resolution P3 [todo] Global load balancing: GeoDNS, anycast, health-check routing P1 [todo] The write path: global strong consistency vs regional writes P3 [todo] Data residency pinning users to regions P3 [todo] Who declares failover, and at what threshold P3 [todo] A full DR runbook example P3 [todo] Failback and split-brain reconciliation P3 [todo] The dependency audit: IdP, DNS, secrets manager P3 [todo] Backup hygiene: 3-2-1, immutable copies, restore testing P1 [todo] Cell-based architecture and blast radius
P Status Topic
P4 [todo] The recruiter screen P4 [todo] The hiring manager P4 [todo] Peers and future reports P4 [todo] Director and skip-level P1 [todo] VP and CTO: 90 seconds with the CTO (drill 57) P4 [todo] Product and design partners P4 [todo] The bar raiser P4 [todo] The three universal closers P1 [done] Reverse due diligence: the red-flag checklist (drill 59) P3 [todo] Debrief dynamics and the follow-up that lands first P1 [done] Level negotiation before compensation negotiation (drill 52) P3 [todo] Component flexibility: sign-on, equity, base, bonus P3 [todo] Equity literacy: RSUs, options, refreshers, 409A, exercise windows P3 [todo] Competing timelines, real versus manufactured urgency P3 [todo] What is negotiable beyond comp P3 [todo] Toronto mechanics: currency, cross-border payroll, termination clauses P3 [todo] References, briefed properly P3 [todo] Post-rejection grace
P Status Topic
P1 [done] AI and LLM drills 1-6, spoken form P1 [todo] Context and graph drills 7-9, spoken form P1 [done] Streaming, reactive and API drills 10-15, spoken form P1 [todo] Caching and edge drills 16-18, spoken form P1 [todo] Compute, Kubernetes and OS drills 19-23, spoken form P1 [todo] Storage drills 24-28, spoken form P1 [todo] Security drills 29-33, spoken form P1 [todo] Frontend drills 34-36, spoken form P1 [done] Leadership drills 37-42, spoken form P1 [todo] SRE, architecture and delivery drills 43-52, spoken form P1 [todo] Patterns and seasonality drills 53-60, spoken form P1 [todo] Deployment drills 61-66, spoken form