Skip to content

Quality matrix

This is a quality audit, not a page-count. A green status in project status is only honest if the page teaches the way a senior engineer actually works.

How we audited

We cannot close-read 150 files in one pass. The method:

  1. Deep-read the first-release vertical slice (README + the first table in project status): methodology, CAP/sharding/hashing/Raft, Kafka/stampede/breaker/tail latency, the four flagship designs, the three DSA pattern pages, the two behavioural stories, the debugging playbook, K8s, and the first-release cloud/reliability pages.
  2. Sample every other Complete pillar (one or two pages): LLD (OOP + Parking Lot), security (OAuth vs privacy), architecture (EDA vs batch/ETL), messaging extras (Pulsar), networking extras (gRPC/K8s), growth-mindset, remaining design exercises (Ad Serving).
  3. Grep the rest for Predict / V1 / bottleneck / trade-off / exit / failure / Mermaid or sim embed / estimates. Hits are not automatic ✓ — a word in passing is ~.
  4. Conservative symbols. If a heading exists but the section does not actually do the job (a "Key Takeaways" that restates definitions, a diagram with no prediction), mark ~ not ✓.
  5. Applicable columns only. Behavioural pages n/a simulation and BOE. DSA n/a production ops and design-V1 (brute force counts as V1 when it is actually written). Catalogs (failure library, cheat sheets) n/a evolution. Process pages (ADRs) n/a capacity math.

Demotion rule (CROSS-6): if a module marked Complete has 3+ applicable ✗, it is Needs review — finished prose is not Complete.

Symbols: present and doing the job · ~ partial / implied · missing · not applicable.

Column What “present” means
Req Opens on a problem / requirements, not a definition
Pred Predict-before-reveal, reasoning exercise, or “predict the log line”
# Back-of-envelope or concrete numbers that change the design
V1 Simplest viable system (or brute force, for DSA)
Fail Named bottleneck or failure that V1 cannot survive
Mech Mechanism introduced because of that failure, not because it is fashionable
Viz Mermaid and/or a working sim where it would actually help
Prod Production caveat, debugging, or “this lie will page you”
TO Explicit trade-off defense (table or A vs B with a decision)
Exit Exit criteria, self-assessment, or “you can do X when…”

First-release vertical slice (deep)

Module Req Pred # V1 Fail Mech Viz Prod TO Exit
Design methodology ~
Requirements & estimation ~ ~ ~
CAP theorem ~
Sharding ~ ~ ~ ~
Consistent hashing ~ ~
Raft ~
Kafka consumer groups ~
Cache stampede ~ ~ ~ ~
Circuit breaker ~ ~
Tail latency ~ ~
URL shortener
Rate limiter (exercise)
WhatsApp ~
Payments
Sliding window ~ ~ ~
BFS / DFS ~ ~
Dynamic programming ~ ~
Technical disagreement ~
Production incident
Debugging playbook ~ ~
Kubernetes ~ ~
Docker ~ ~
Terraform ~ ~ ~
CI/CD ~ ~ ~ ~
Deployment strategies ~ ~
IAM & managed services ~ ~ ~
FinOps ~ ~ ~ ~
Model serving ~ ~ ~
Microservices communication ~ ~
Stateless vs stateful ~ ~
Single points of failure ~ ~ ~
Consistency models ~ ~
Replication ~
Indexing ~
SQL vs NoSQL
Engineering mathematics ~ ~ ~
Queue patterns ~
Cache strategies ~
Failure library ~ ~
Distributed cache (ex)
Load balancer (ex) ~
API gateway (ex) ~ ~
Notification system (ex)
Web crawler (ex) ~ ~ ~
Autocomplete (ex) ~ ~
Distributed KV (ex) ~ ~ ~
Social feed (ex) ~

First-release notes

  • Gold standard is the four flagship exercises plus CAP / Raft / circuit breaker / cache strategies / queue patterns. Those already are the method.
  • Sharding, consistent hashing, Kafka have working sims and production sections but no predict-before-click task on the host page (the sim is used as a demo). That is ~ / ✗ Pred, not a demotion by itself.
  • Terraform, CI/CD, SPOF, microservices communication miss 3+ applicable columns → demoted. They read as finished notes, not a reasoning loop.
  • SQL Deep Dive is labeled Complete in the old tracker and Draft on the databases hub. Draft wins.

Sampled Complete pillars

Module Req Pred # V1 Fail Mech Viz Prod TO Exit Verdict
OOP fundamentals ~ ~ ~ ~ ~ Needs review
SOLID ~ ~ ~ ~ ~ Needs review
Design patterns ~ ~ ~ ~ ~ Needs review
Concurrency basics ~ ~ ~ Needs review
Concurrency execution models ~ ~ Complete
Parking Lot (LLD) ~ Complete (sample of the 15)
OAuth2 & OIDC ~ ~ ~ Complete
Session management ~ ~ ~ Complete
Threat modeling ~ ~ ~ ~ Complete (process page)
AuthN/Z ~ ~ ~ Needs review
Zero Trust ~ ~ Needs review
Data privacy ~ ~ ~ ~ Needs review
Web vulns ~ ~ ~ Needs review
Kafka internals / Pulsar ~ ~ Needs review
Pulsar primer ~ ~ Needs review
Event-driven architecture ~ ~ ~ Complete
Batch/ETL Lambda vs Kappa ~ Needs review
gRPC vs HTTP in K8s ~ Needs review
Modern protocols / mesh ~ ~ Needs review
Realtime communication ~ Needs review
API design ~ ~ Needs review
ADRs ~ Complete (process)
Architecture reviews ~ ~ Complete (process)
DDIA concepts ~ ~ Complete
SQL Deep Dive ~ ~ ~ ~ Needs review (already Draft on the hub)
Ad Serving (ex) Complete (V1-first sample of the 21)
Self-respect ~ Complete (judgement, not STAR)
Release engineering ~ ~ Needs review

Grep pass (remaining Complete labels)

Not a close read. Sections found → ✓ or ~. Three-plus applicable ✗ still demoted.

Cluster Typical shape Notes
Remaining design exercises (Pastebin → Video calling, capstone) ✓ Req # V1 Fail Mech Prod TO Exit; Pred ~; Viz ~ Written to the Pastebin “evolves and clicks” bar. Keep Complete. A pedagogy pass on leftover dumps is still listed under Planned.
Remaining LLD exercises (15) ✓ Req V1 Fail Viz; Pred ~; # —; Prod —; Exit ~ 9-step + class diagrams + concurrency. Sampled Parking Lot. Keep Complete.
DSA Interactive pages (heaps → Aho-Corasick) ✓ pattern + viz; Pred ✗ on most Status is Interactive, not Complete — out of this matrix.
DSA written Complete (foundations, two pointers, binary search, pattern recognition) ✓ problem/pattern; Pred ~; Fail/Mech/TO weak Keep Complete as DSA-bar pages (same as sliding-window), not system-design-bar. Pred-before-viz is the remaining gap.
Distributed fundamentals, multi-region DR, CRDTs, stream processing, serverless vs containers, multi-tenancy, microservices vs monolith ✓ Req Fail Prod; Pred/V1 often ~ Keep Complete. Not gold-slice.
Observability: production reliability, testing strategy ✓ Req Fail Prod; TO ~ Keep Complete.
Behavioural remaining 7 + growth-mindset remaining 6 ✓ story/judgement; engineering columns — Keep Complete.
Reference: cheat sheets, glossary, trade-off matrix catalogs Keep Complete as reference, not concept. Exit is weak (no “you can do X”).
Performance fundamentals ✓ numbers/fail; Viz ✗; Mech ~ Keep Complete.

Demoted this pass (Complete → Needs review)

Page Why (applicable ✗)
SQL Deep Dive Tracker said Complete; hub already said Draft. No prediction, V1, visual, production-debug, or failure-driven mechanism.
Terraform No capacity numbers, no smallest-justified V1, no bottleneck chain. State/drift are taught, but not as a reasoning loop.
CI/CD Pipeline catalog. Missing prediction, numbers, V1, bottleneck, mechanism-from-failure.
Release engineering Missing prediction, numbers, trade-off defense.
Single points of failure Missing prediction, numbers, V1. SPOF hunt is the skill; the page does not make you practice it.
Microservices communication Eight patterns listed. No numbers, no V1 (sync REST), no prediction.
Batch/ETL Lambda vs Kappa Missing prediction, numbers, V1, trade-off table.
Kafka internals & Pulsar Internals dump. Missing prediction, numbers, V1, visual, trade-off defense.
Pulsar primer Ops dump. Missing prediction, numbers, V1, bottleneck, visual, trade-off.
Modern protocols & mesh Missing prediction, numbers, bottleneck, mechanism-from-failure.
gRPC vs HTTP in K8s Long production notes; missing prediction, mechanism-from-named-failure, trade-off defense.
Realtime communication Missing prediction, numbers, V1.
API design Missing numbers, V1, trade-off defense.
OOP / SOLID / patterns / concurrency basics Definition pages. Missing prediction, numbers, V1. Execution-models stays Complete.
AuthN/Z Missing prediction, numbers, bottleneck identification.
Zero Trust Missing prediction, numbers, trade-off defense.
Data privacy Missing prediction, numbers, V1, bottleneck, mechanism, production, trade-off.
Web vulnerability classes Missing V1, production debugging, trade-off defense.

Vendor DB pages (postgresql, mongodb, cassandra, dynamodb, redis) were already Draft — left alone.


Remaining gaps (do not pretend these are done)

  • Predict-before-click on host sim pages (Kafka, sharding, hash ring, stampede, Raft). The failure-injection lab is the stopgap, not a substitute for a box on each host page.
  • Exit criteria as a named heading. Most pages have Key Takeaways; few say “you are done when you can…”.
  • DSA visualizers still run as demos. Brute-force-first and “predict the next highlight” are rare.
  • Demoted pages need a quality pass, not more topics.
  • Interview-mode tabs, capstone UX, and Go example parity remain Planned.
  • No new system-design #41. Freeze expansion until Completes on this matrix are honest.