11. What to prepare / what not to

Highest-value work for a third live and a possible fourth. Not a Job 10454435 forecast. Not “grind everything.” Dates in pasted research dumps were predictions — use the round map, not a calendar panic.

Highest-value items

1. Genuine STAR hour. Scan chapter 12, speak 05-star-br.md: all 16 with the spread, why-chain on Dive Deep / Argus / hybrid / costmap. Failure = Argus 73% mAP (no invented manager quote). Deadline = no fake miss — Horizon 60 FPS + live costmap, flag stale occupancy early. Conflict = technical hybrid vs vector or RLS-in-DB vs app-only, then commit. Do not invent a miss. Never speak chapter 13 as yours.
2. Why-stack hour. Scan chapter 14 then chapter 15; speak 06-tech-why-scale.md and 07-ylogx-horizon-live.md. Discarded option plus failure mode plus resume metric. 10× is more ranker / RLS tests / cache / mAP gate — not shard, Kafka, EKS, or a bill.
3. Stay in DSA posture even if HR said LP-only. LC 7406809 AUTA. Write one Java problem brute → optimal out loud the morning of.
4. Distance K without a parent Map plus a heap (k-th largest) plus a grid DP (count, then max reward). Fluency can be the tree. Intern Fluency on LC 7643209 was an unnamed min-heap, not Distance K. HM can be the heap. AUTA 6800853 is OA-as-R1-flagged but the DP is still the right family. BR coding: Maximum Sum Linked List (6881058) — switch at common nodes, not LC 962.
5. IQVIA 10x / POC answer. LC 7563011 shape. Parallel scrapers vs one; hybrid+graph vs vector; LangSmith as stop. Azure, not Qdrant.
6. GenAI two minutes. Should: cited RAG, SQL behind RLS, GiftedBooks PDF. Should not: RLS, costmap, PPE boxes, dashboard hot path. If they paste a graph, code first.
7. Java locals vs fields, static, HashMap equals/hashCode. Standard CS. Do not fake Kafka.
8. One LLD skeleton on paper: Locker or Parking Lot classes plus “full” / “wrong pin.” Plus Trie+count if they want DS-design instead.
9. Mental model: R3 may not be last. AUTA posts schedule BR after a third live. Do not celebrate so hard you skip a coding warm-up for R4.
10. Resume critique. Skills listed vs evidenced, GitHub mismatches, imprecise %. Chapter 17. Horizon/IEDC depth: chapter 16. New paste: remap in chapter 18 before grinding it as your third.

What not to overprepare

• Amazon retail HLD, Kafka-as-if-you-ran-it, EKS, segment trees, Fenwick.
• Fabricated missed SLA, manager fight, mentee count, AWS bill — chapter 13.
• OA question clusters (shipments / make array positive) — OA is not a round; that cluster was flagged Class C in your own research.
• Login Tracker as “confirmed UTA R1.” Unverified. LRU is enough.
• Rate limiter as “every UTA R2.” Count = 1, not your default. Know token bucket at high level anyway.
• Sixteen new STAR scripts. You already have chapter 11 of the study book. Later rounds want depth on failures, not a new product.
• Word-perfect 2,000-word LP HTML. Shape + metrics + why-chain.
• InstaRecon internals. Ethics line, redirect.
• Mixing GiftedBooks with AegisAI, or inventing Haar, Qdrant, missed 99.9%, student headcount, AWS bills, cache TTLs.

Morning-of scan

Intro 90s (ch. 02)
Dive Deep Ylogx −35% + 3-tier RLS
Deliver Horizon 17th  — or Ylogx 40%/99.9% if rover already used
Argus 73% as mistake
IQVIA hybrid POC
GenAI should / should-not
Distance K  |  Unique Paths DP  |  min-heap k  |  Kahn
Java local vs field
One question for them (ch. 04)

Open this folder, not the old dumps

Secondary research pastes disagree and sometimes skip OA-as-R1. The lock is chapter 01 plus Question-Research-BIBLE.md. This notebook does not replace those files and did not edit them.