Four-Leaf has now written up seventeen company interview processes, each built the same way: only what the company itself publishes, or what a named person said on the record, with the widely repeated claims that have no source named and refuted rather than passed along.
Doing seventeen of them surfaced something the individual guides can't show. The companies differ far less in how they interview than in how much they'll tell you about it, and the gap between the two is where nearly all bad interview advice lives.
This page puts them side by side.
What each company publishes
| Company | How much it publishes | The round or rule that decides |
|---|---|---|
| Salesforce | A lot, split by function | Publishes its HackerRank grading order: does it run, then design, then scale, then readability |
| Uber | A lot, including durations | Recruiter and panel debrief against named competencies and values |
| Nvidia | Rules, not stages | Required in-person onsite, and outside AI tools disqualify |
| Amazon | Its principles and its process roles | The Bar Raiser, a trained interviewer outside the team |
| Its attributes, not its loop | A hiring committee, not your hiring manager | |
| Microsoft | Six competencies, no round list | A senior final interviewer, no published program behind it |
| Netflix | Its culture, exhaustively | Culture memo alignment, and it invites disagreement |
| Stripe | An old but detailed guide | A published communication rubric with scoring bands |
| Anthropic | Its values and AI-use stance | A values round that drives a lot of rejections |
| OpenAI | Fragments | A point of view on AI, threaded through every round |
| Apple | Almost nothing | The hiring manager, who decides |
| Meta | Fragments | A hiring committee, with level set inside the loop |
| Bloomberg | Almost nothing | A code review round that catches strong coders |
| SpaceX | Almost nothing | A project presentation to the team you would join |
| Robinhood | Almost nothing | Two system design rounds in one onsite |
| Tesla | How to behave, not what happens | Thought process over correct answers, by its own instruction |
| Airbnb | Almost nothing | Unknown, and the company will not say |
The right-hand column is the useful one. Read down it and the loops that look identical from the outside turn out to be answering different questions, and to be decided by different people.
The four things worth knowing across all of them
Who decides varies more than what they ask. Google and Meta run committees. Apple's hiring manager decides. Amazon gives a trained outsider real weight. Uber describes a panel debrief with a recruiter and no single gatekeeper. This changes your preparation more than the question bank does. If a committee that never met you reads a written packet, the interviewer's notes are the artifact being judged, and making them easy to write is a real strategy. If your hiring manager decides, the conversation with that one person carries the weight.
Almost nobody publishes an AI policy. Nvidia states plainly that unapproved outside tools such as ChatGPT will disqualify you. Anthropic addresses candidate AI use in its own guidance. Beyond that, published policies are scarce, and prep sites fill the silence with confident assertions of bans that no company has stated. Ask your recruiter. It's a normal question and the answer is genuinely useful.
In-person is returning, quietly and unevenly. Nvidia requires an onsite before it will consider an offer, in writing. Most companies leave the format open. Several careers pages describing everything as virtual have not been touched since 2020, and at least one still carries a live pandemic banner. Format claims age badly and this is the detail that costs you a flight.
The most repeated claims are the least sourced. Bar raiser rounds get attributed to companies that never used the term. Fixed round counts appear as fact while three sites give three different numbers. Acceptance rates get quoted with no filing behind them. In seventeen write-ups, the pattern was identical every time: the more specific and confident a claim, the less likely it had a source.
The unpublished-process problem
Five of the seventeen publish almost nothing, and they include some of the most searched employers in tech. That vacuum is where the prep-site industry makes its living, and the content that fills it isn't lying so much as guessing and then presenting the guess in a table.
Airbnb is the clearest case. Nearly every description of its process, including the claim that non-technical employees run values interviews with veto power, traces to a single Medium post from October 2019 by a named former executive. He was describing a company of just over five thousand people. Airbnb now reports around 8,200 and went through a 25 percent staff cut in between. The source is real. It's also nine years old and nobody quoting it says so.
The workable response is to ask your recruiter the process questions directly, in one message, early, rather than hunting for a better prep site. What are the stages, how long is each, who is in the room, is it virtual or in person, and what is the AI policy. Recruiters answer this readily. It is their job, it costs them nothing, and the answer is current in a way no article can be.
Where the published processes agree
Strip out the branding and the same instruction appears across every company that says anything at all.
Uber's engineers say ask questions before you build, and say it at three different levels of seniority. Salesforce says narrate your decisions on the whiteboard. Tesla writes that its interviewers may value a compelling thought process more than a correct answer, and that some questions are vague deliberately. Stripe publishes a communication rubric whose top band is about supporting general claims with specific details. Nvidia's sessions run 30 to 60 minutes, which only works if you get to the substance fast.
That's five companies with nothing in common structurally, all describing the same behavior: think out loud, ask before assuming, be specific under follow-up. It's the thing that transfers across every loop in this table, and it happens to be the one thing that reading about it will not teach you. Our guides to preparing for technical interviews and structuring behavioral answers cover the mechanics, and practising out loud is what closes the gap between knowing your answer and delivering it under pressure.
How these guides are built
Every guide linked here follows the same rules, and they are worth stating because most interview content follows none of them.
Company-owned sources first: careers pages, engineering blogs, newsrooms, and SEC filings. Then named individuals on the record, or reporting from a named publication. Aggregator and prep sites are never cited as evidence, only used to identify claims that need checking.
Where a company publishes nothing, the guide says so instead of filling the gap. Where a claim is widely repeated but unsourced, the guide names it and explains what is actually known. Where a source is real but old, the guide dates it. Numbers get pulled from the filing rather than from another article about the filing.
That approach produces shorter, less satisfying answers in places. It also means that when one of these guides tells you a company requires an in-person onsite or grades correctness before elegance, there is a sentence on a company page behind it, and you can go read it yourself.
The seventeen guides
Publish the most. Salesforce, Uber, Nvidia, Amazon, Netflix.
Publish some of it. Google, Microsoft, Stripe, Anthropic, OpenAI, Meta.
Publish almost nothing. Apple, Tesla, Airbnb, Bloomberg, SpaceX, Robinhood.
If you're interviewing at a company not on this list, the method transfers. Find the careers page, the engineering blog, and the most recent annual report. Read what the company says about how it hires and what it says about its own headcount and turnover. Then ask your recruiter the five process questions. You will know more than any article about that company, including this one.