Things I want to dive deeper

The standing backlog of topics the vault surfaced but hasn’t gone hands-on with yet. Each entry: what it is, why it earned a slot, and where the vault already touches it. Entries leave by being dived into (a proper note/hands-on lands elsewhere and the entry becomes a link), never by deletion.

Backend concepts list (Puneet Patwari, Principal Engineer Atlassian)

The anchor capture (thread, 1k❤/45.6k views): “13 years in backend. Concepts I wish I’d learned in year 2 instead of year 8 — none of them are hard, every one would have saved me from tonnes of headaches.” The ten, each mapped to vault state:

• p99 latency, and why averages lie — tail thinking; kin to Performance engineering (measure first, elimination ladder)
• Connection pooling — unexplored hands-on; relevant to the Postgres lanes (Embedded databases’ server cousins)
• Backpressure — queue/signaling discipline; unexplored
• Idempotency keys — already vault doctrine in the agent lane (CLIs for agents rule 4: “agents retry autonomously, so operations must be safe to repeat”); the payments/distributed-systems depth is the dive
• Reading query plans (EXPLAIN ANALYZE) — partially covered (Kurs administracji PostgreSQL lane); hands-on gap
• Write amplification — storage-engine concept (why LSMs and B-trees trade differently); unexplored
• Consistent hashing — distribution primitive; unexplored
• Change data capture — already a full note (Change Data Capture); the dive = hands-on with a real pipeline (duckle CDC lane)
• The outbox pattern — reliable event publication; the natural next step past CDC
• Tail latency amplification — why one slow call in a fan-out ruins the page (the p99 entry’s deeper why); unexplored

Open questions

• Which of these earn a labs-lane session first? Candidates by homelab leverage: EXPLAIN ANALYZE (Postgres everywhere), CDC + outbox (the duckle lane), backpressure (queue work).

pstack (poteto) — hands-on with the plugin itself

USER note at capture, verbatim: “I have to dig deeper into poteto pstack.” Surfaced by DmitriyLeybel’s tweet (403❤/21.6k views, captured): an Opus 5.5-generated deep dive calling pstack “a masterclass in agentic engineering” (Claude artifact — login-walled, not captured; the vault’s own pstack note is already the deeper primary source: playbooks/principles/skills anatomy, 2,500 PRs/month, verification-first delegation). The dive to do is hands-on: install the plugin, run a real task through /poteto-mode end-to-end, and check its finish-condition/verification loop against the vault’s own capture ETL and the jnsahaj process-skills set.

Flowchart runbooks (CFE format generalized)

USER: “worth to generalize and try to present how-tos or runbooks or devops pipelines — really dense and information packed” — the Cooking-for-Engineers recipe format applied to ops prose. Full mapping sketch in Design inspirations; the dive = render one real oncall runbook + one deploy pipeline as Mermaid flowchart LR with ingredient-style input columns, and check whether incident-mode readability actually improves over numbered prose.

Distributed systems: the evergreen reading (aphyr, 6.824)

vivekgalatage (637❤/28k views; tweet + QT captured): “The evergreen resource” — distsys-class (OSS) (10.1k★) — Kyle Kingsbury’s (Jepsen) “An Introduction to Distributed Systems” lecture series, captured in full: nodes/networks → causality diagrams → clocks → availability → consistency → tradeoffs → “avoid consensus wherever possible” → consensus → transactions → characteristic latencies → common distributed systems → a pattern language → production concerns. The QT adds the companion course: MIT 6.824 (schedule at nil.csail.mit.edu, HTTP-only, noted) — GFS lecture included.

Placement: this is the spine for the backend-concepts lane above — consistent hashing, tail-latency amplification, the outbox pattern, and Change Data Capture all have their theory chapters here; “Characteristic latencies” is the numbers-table backing the p99/tail-latency entries. A Pattern Language + Production Concerns are the two chapters to read first for ops leverage.