Sep. 14–21 interview digest: a Google round with no code, Microsoft's senior bar, and a $620K Anthropic offer

The week in one read

• A Google phone screen can spend the entire round on a proof and never ask for code. An engineer currently at Amazon described a Google L4 phone screen where the question only became solvable after proving a mathematical constraint. Proving it took about 30 minutes and required a hint. The candidate offered to write a brute-force version first and was turned down, then offered to code the proven approach and was turned down again, and was asked a theoretical follow-up question instead. Five to ten minutes remained on the clock. 2
• Late Google onsite rounds can hang on a single data structure. A candidate from Salesforce interviewed for a Staff SRE/SWE role, cleared the initial system design and behavioral rounds, and reached an onsite where the first system design round went well and the final two rounds asked about random shuffling and KD trees with roughly 30 to 40 minutes each. The candidate reached the right data structure, a KD tree, and ran out of time before the code was complete. He had prepared for five straight months and wrote that he had already cleared loops at several other companies. 3
• Microsoft's senior loop can end one level below the role you applied for. A senior engineer currently at Meta interviewed through a Microsoft hiring event. Across four onsite interviews, none of which were system design rounds, the hiring manager probed details after coding questions. The recruiter's verdict: below the Senior SWE bar, with a hire signal for SWE 2, and team matching would begin. The candidate asked whether anyone at Microsoft would run one or two system design rounds for a Senior SWE L63 opening. 4
• The same senior signal gap down-leveled an 11-year engineer. A candidate describing himself as a Senior SWE with 11 years of experience reported being down-leveled to L4 after system design rounds, and wrote that the same thing happened at a second company. Blind's generated summary of the post's 10 comments, dated Sep. 21, records one commenter reporting a similar outcome after a design interview and several suggesting more behavioral-interview work. 5
• A passed final round at Google can stall on recruiter turnover. A candidate at Microsoft reported passing an in-person Google final round, then watching his recruiter be fired the day after promising to send feedback. Her email address began bouncing. A replacement recruiter sent what the candidate called a vague AI-generated introduction with no next steps, and 10 days passed without an update despite two follow-up emails. 6

Evidence at a glance

Report cards

The question bank you can practice now

1. When a phone screen stops being a coding round

"I can prove the constraint or I can write a working implementation with a stated assumption. Which one do you want on the board first?"

2. Rounds that turn on one data structure

• Spatial indexing. Implement a KD tree build, nearest-neighbor query, and range query in one sitting, and be ready to say where the structure loses to a grid or an R-tree: dimension count, update frequency, and how skewed the data is. KD trees serve proximity search, which is where the interviewer took the question.
• Shuffling. Write Fisher-Yates, then answer the follow-ups that follow from it in an interview: why the naive sort-with-random-comparator is biased, how to shuffle a stream you cannot hold in memory, and how to sample k items from n without replacement.

3. Senior signal in the system design round

1. Ask for the scale target in the first five minutes, and write the number on the board.
2. Name two candidate architectures, then pick one out loud and say what you gave up.
3. Close with the first thing that breaks at ten times the load, and what you would monitor to see it coming.

4. Google's L5 AI/ML domain round

After the loop: process risk

• Recruiter turnover can silently reset a decision. The passed Google final round went 10 days without an update after the original recruiter's departure; the replacement's boilerplate introduction contained no next steps. 6 Escalation paths that work here are the ones that skip the missing recruiter: the hiring manager, the interview coordinator, or anyone named in the loop's scheduling emails.
• Team matching carries its own clock. A candidate relayed his recruiter's terms: the aligned team can close with another candidate, a passing result stays valid for 18 months for team matching, and a failing result triggers a 12-month cooldown. He was weighing a rushed interview in about two weeks against keeping a later date. 8
• A single-team onsite leaves no fallback. Blind's generated summary of the Staff SRE report notes the loop covered one team. A loop that meets one team converts the team's headcount decision directly into the candidate's outcome. 3

Compensation: signed offers and refreshed medians

Signed and reported this week

The signed offer as posted on Blind, read Sep. 21, 2026. Name, date, and location are redacted by the poster; the level is not stated anywhere in the post.

Fresh aggregate anchors, refreshed Sep. 21, 2026

Company read-through

Before your next interview or negotiation

1. Settle the round's output in the first two minutes. Ask whether the interviewer wants a proof, a design, or running code, and say what you will deliver with the time left.
2. Drill spatial indexing and shuffling to a 30-minute clock. A KD tree with nearest-neighbor and range queries, plus Fisher-Yates with its bias follow-ups.
3. Write your senior-signal checklist before the loop. Scale number, rejected alternative, migration path, operational cost, on-call owner.
4. Ask how the level will be assessed when a recruiter describes an onsite without a system design round.
5. Keep an escalation path that skips the recruiter. Colleagues on the loop and the interview coordinator are reachable when a recruiter goes silent.
6. Price the second contract period. Model vesting schedules against start date before comparing headline totals, and treat the refreshed medians as a reference point rather than a target.