1. Lock and evidence

Adarsh Vishwakarma, SDE I AUTA APJ, Job 10454435, ADCI Karnataka, Java, Zoom + Amazon Live Code. R1 done 13 Aug 2026. R2 done 18 Aug 2026. This notebook is the third live and a possible fourth. Still no public IE names this Job ID for a live-round question. Unnamed stays unnamed. Labels: IE-asked Resume-derived Standard CS.

Your loop vs other people’s numbering

Amazon University Talent Acquisition loops are not one shape. Your process, as locked in Question-Research-BIBLE.md §0 and §3:

OA / coding test / workstyle     ← not a round. Do not study it as R3.
live R1  (13 Aug 2026)           ← done
live R2  (18 Aug 2026)           ← done
live R3  (next 60 min)           ← this notebook
possible R4                      ← often Bar Raiser; sometimes HM if R3 was already tech

When a blog says “Round 3” they often counted the OA as Round 1. Their Round 3 is then the second live — the analogue of the hour you already finished. Those questions are still useful as pattern reuse. They are not evidence of what a third live after two lives contains until you remap.

Example. GFG Off-Campus 2025: R1=OA, R2=Count Complete Subarrays, R3=Connect Sticks + Rotten Oranges. Mapped: their R2 → your already-done R1 analogue; their R3 → your already-done R2 analogue. igreaper Medium is the same numbering. LC 6800853 AUTA grid paths + digit-subtract sits in that trap unless the post proves two lives already happened before that slot.

The OA-as-R1 trap

Put this in your head before you read any new “R3 experience”:

IE heading	Maps to for you
OA called Round 1, then “Round 2” coding	First live — you already did this
Their “Round 3” / Tech Interview 2 after that	Second live — you already did this, if the loop was two lives after OA
OA + two lives + a later HM / BR / Fluency	Closest to your R3 / R4
OA + coding + LLD + HM (three interviews)	Not UTA-default two-DSA. Rate Limiter lives here (count = 1)

Closest first-hand shapes for a true third live already in the research bible:

LC 7563011 AUTA Bengaluru — two DSA lives, then HM (complex problem, POCs, multiple solutions).
LC 8362604 AUTA APAC 2026 — two DSA same day, then HM (Anagrams variation + Combine Garlands), then BR.
LC 6806195 AUTA — R1/R2 then BR (Nice Subarrays + named LPs).
LC 7406809 AUTA — recruiter said LP-only; interviewer still asked DSA. BR later with a tree path.
LC 7623949 — slot labeled GenAI Fluency was Distance K DSA, O(n), no parent map.
LC 7850431 — R3 GenAI prompts plus dynamic k-th largest DSA.
LC 7724048 — R3/HM LLMs + Next Greater.

Receiving a third invite does not mean the loop is over. Several AUTA accounts schedule Bar Raiser after the third live. It also does not mean the third live is the Bar Raiser. The invite text you have is not a round-type label unless it says Bar Raiser or Hiring Manager.

Conflicts in secondary research

You pasted several research dumps. They disagree. The research bible with opened bodies wins. The rest is Class C until a URL is opened.

Claim in a dump	What the bible actually supports
R3 is always Bar Raiser	Unsupported. AUTA IEs put BR after a third live (6800853 recruiter wording; 7563011 HM; 8362604 HM then BR).
R3 is always GenAI	Unsupported. Fluency can be Distance K (7623949). Rank A two-DSA UTA often named no GenAI-primary prompt.
R3 is always LLD / Parking Lot	Parking Lot is a real SDE I ask (GFG fresher R2). AUTA Rank A is mostly two-DSA. Arijit AUTA Tech 2 was unnamed SD. Prep both class design and DS-design. Do not spend the week on Amazon-retail HLD.
SDE I never gets OOD, only Trie+Heap	Conflicts with Parking Lot / Locker / dog check-in / bookstore OOD rows in bible §5.C. Prep both.
Login Tracker was asked in UTA R1	Unverified. Mentor prep (Ravi) is not a first-hand IE.
Rate limiter is a standard UTA R2	Independent SDE I live count = 1, OA+3 system design, not UTA two-DSA.
Java local-variable “trap” in a 2026 AUTA BR	The Java fact is true (locals do not default). A first-hand Amazon URL for that exact trap is not in the bible as a stamped 2026 AUTA BR. Treat as Standard CS.
Celebrity Problem as Jan 2026 Hyderabad AUTA BR	Celebrity is an old Amazon/GFG classic. The specific 2026 AUTA BR citation in one dump had no opened URL in the bible. Include the algorithm as Standard CS; do not quote a fake IE.
“412 URLs” / “500 sources” in a paste	Do not launder. Honest quotas live in Question-Research-BIBLE.md §1. The notebook miner was a targeted pass: 68 seeded, ~33 opened IE bodies — far under 500.
OA cluster Minimum Shipments / Make Array Positive	Class C / flagged ghostwriting cluster in your own earlier research. OA is not a round anyway.

Still unverified

Login Tracker (new_login / get_oldest_login) — mentor only. If they ask it, it is LRU with insert + peek-oldest. Do not say “Amazon asked this in Job 10454435.”
Job 10454435 live question list — still none. LC 8413215 is “Post not exist.” CampusToCareer “Tech 2 = APIs/caching” is Class C invention.
Ylogx SEO 403 / noindex — prep-only. Do not lead Dive Deep with it. Lead Redis + RLS (resume numbers).
GiftedBooks GitHub currently describes AegisAI. Resume wins.
Haar cascades as Argus first attempt — not on the resume. Do not say it unless you confirm it happened.
A missed ERC or a missed 99.9% SLA — not on the resume. Do not invent a miss for Earn Trust.

Java vs production — say once

Live Code is Java. Ylogx and IQVIA shipped Python (FastAPI, LangGraph) and TypeScript (NestJS, React). Horizon was ROS2. Same data-structure ideas; different syntax. Do not pretend the internships ran on the JVM.

If they ask why Java in the interview and Python on the resume: Amazon’s loop language is Java. Production chose the stack the job used. I can write the same HashMap, heap, and graph in either.