Content review checklist (2026-08)¶

Full-curriculum accuracy pass: factual mistakes, capacity math, contradictions, overly dense or missing explanations. Items are done unless marked remaining.

Distributed systems¶

[x] fundamentals.md — clock skew: fast clock expires early (not slow)
[x] fundamentals.md — fencing token, not client-side TTL
[x] fundamentals.md — vector clocks: merge then increment once
[x] fundamentals.md — isolated Raft leader still believes it is leader; cannot commit
[x] fundamentals.md — Raft term diagram winner matches labels
[x] fundamentals.md — Lamport receive = max(local, msg) + 1
[x] fundamentals.md — Redlock is best-effort; Kleppmann / fencing
[x] index.md Q1 — PACELC: no-partition case is still L vs C
[x] replication.md — DynamoDB is per-partition leader, not leaderless Dynamo
[x] consistency-models.md — R+W>N is overlap, not linearizability
[x] raft.md — left as gold standard (own-term commit / joint consensus)

Foundations / math¶

[x] math.md — 99.95% series downtime via 8760 × (1 − 0.9995^5)
[x] math.md — 1 MB sequential RAM ~100–250 µs
[x] requirements-estimation.md — average vs peak QPS labels
[x] api-design.md — idempotency race / gateway vs resource

Databases¶

[x] redis.md — Cluster does automatic failover (async data-loss caveat)
[x] redis.md — node counts aligned; ZRANGE ascending
[x] redis.md vs consistent-hashing.md — hash slots, not a pure ring
[x] consistent-hashing.md — add ≈ K/(N+1), remove ≈ K/N
[x] dynamodb.md — GSI Query needs equality on PK; time-range bucket+SK
[x] dynamodb.md — provisioned price order-of-magnitude + hot-key takeaway
[x] cassandra.md — ⌊N/2⌋+1; batch types; secondary-index scatter-gather
[x] cassandra.md — quorum overlap ≠ “strong reads”
[x] sql-vs-nosql.md — BASE is not family-wide
[x] ddia-concepts.md — isolation = anomalies then MVCC vs 2PL; hash ≠ hot-key
[x] ddia-concepts.md — LWW register vs CRDT counters
[x] sql-deep-dive.md — IN subquery, TABLESAMPLE estimator, FROM-order myth, index column order
[x] databases/index.md — vendor/SQL pages Draft / needs review, not Coming soon

Messaging / reliability / performance¶

[x] kafka.md — lag = offset gap, not rate difference
[x] kafka-internals-pulsar-comparison.md — acks=all vs min.insync.replicas
[x] kafka-internals-pulsar-comparison.md — Kafka EOS ≠ local DB + commit_async
[x] kafka-internals-pulsar-comparison.md — broker add ≠ group rebalance; rebalance times; session.timeout.ms 45s
[x] pulsar-primer.md — cost units, partitioned topics, write vs ack quorum, pulsar://
[x] messaging/index.md — redelivery IDs are not Kafka offsets
[x] cache-stampede.md — Redis SET NX across pods
[x] cache-strategies.md — write-through diagram = DB then cache
[x] rate-limiting.md — leaky bucket section; INCR/EXPIRE race
[x] Timeouts — request budget shrinks inward; idle: LB closes first
[x] Retries — 3 retries = 4 attempts; default full jitter

Networking / security / cloud / observability / K8s¶

[x] HTTP/2: no HTTP HOL; TCP HOL remains
[x] QUIC first flight 1-RTT; 0-RTT is resumption
[x] BBR ≈ BDP; CUBIC ×0.7
[x] ALB 100k RPS cost not ~$18/month
[x] NLB is not a Postgres writer/reader router
[x] limit_req_zone in http{}; no Istio kind: EgressGateway
[x] Refresh-token theft mints access tokens
[x] JWT signed (JWS), not encrypted (JWE)
[x] OIDC ID token = user, not the client/valet
[x] Env vars not in ps aux; GCP Assume Role ≠ SA keys
[x] cProfile tottime/cumtime; four golden signals include saturation
[x] Contract tests: additive JSON is not breaking
[x] Liveness vs readiness vs dependency checks
[x] CPU affinity needs LockOSThread; GOGC via SetGCPercent

System-design exercises¶

[x] Video streaming — Little’s law concurrent + 36 PB/day egress
[x] Collaborative editor — ~28K concurrent docs; OT vs CRDT is a choice
[x] Autocomplete — 1.2 trillion lookups/month; timeout vs SLO
[x] URL shortener — 1000:1 R/W; 7 hex chars tiny keyspace; CDN recosted
[x] Pastebin — no CDN for burn-after-read / private
[x] Payments — outbox ≠ PSP charge; PCI reduced scope; webhook ack; V1 called out
[x] API gateway BFF — Order then User (not same par)
[x] E-commerce / food / ride-hail / recs / social / logs / metrics / Instagram NFRs internally consistent
[x] Notification self-assessment matches crash-after-provider
[x] EDA — charging on OrderPlaced framed as antipattern
[x] Video calling — NACK/RTX/FEC, not “no retries”
[x] KV store — Dynamo paper vs DynamoDB; GET 200 + values[]
[x] Hotel/flight typo + catalog units

LLD¶

[x] Splitwise remainder → payer
[x] Logger FileHandleRegistry + non-blocking shutdown sentinel
[x] Parking lot duplicate plate
[x] Task scheduler priority among due tasks

DSA¶

[x] KMP mismatch 'c' vs 'd' + real fallback
[x] HyperLogLog ~12 KB
[x] Bucket sort 0.42 in [0.4, 0.6)
[x] Aho-Corasick failure links, not merged "he" node
[x] heapq.heapify in place
[x] Variable window assumes non-negative; viz HUD no bogus Target
[x] DP / Dijkstra viz vs snippet called out
[x] Union by rank ≠ union by size
[x] Pattern-recognition links; word-ladder complexity
[x] dsa/index.md does not claim every page has Step/Go

Status / how-to-study / reference¶

[x] README — rest of curriculum is shipped, not stubs
[x] project-status.md — LLD all 15 Complete; no “remaining 11”
[x] how-to-use.md — Learn/Interview/Staff tabs not wired
[x] Story count = 6 on behavioural + how-to-use
[x] Production incident — reversible mitigation, then RCA
[x] Playgrounds / home — 15 priority sims + DSA visualizers
[x] Glossary DSA terms + CQRS link
[x] Cheat sheet Dijkstra space, Timsort/Java, consistent hashing add vs remove
[x] Roadmap — start DSA patterns in week 1

Remaining (not this pass)¶

Interview-mode tab UX as a reusable component (still Planned)
Capstone + Go example parity
Vendor DB pages stay Draft / needs review (facts patched; not gold-template)
Extra V1-first rewrite on leftover “finished diagram first” exercises
Visualizers that still simplify the algorithm (Dijkstra O(V²), DP loop order) are labeled, not rewritten