The Interview Report Standard, Each Report Graded Admit, Hold, or Reject

This standard is for a candidate who has an interview booked and a pile of review-site entries, forum threads, and third-party prep guides about one employer. It sets the bar a single report has to clear before it earns prep hours: seven written criteria, three verdicts (Admit, Hold, Reject), and a rule for which claims inside an admitted report make it onto your prep sheet. If two people grade the same report with this document, they should reach the same verdict and the same list of admitted claims.

What counts as an admitted interview report?

An admitted report passes six report-level gates: exclusion, dated, role family, level, business unit and location, and provenance. It also has at least one claim backed by the employer's own process page or by a second independent report. Anything else is either Hold, which makes it worth a question to the recruiter, or Reject, which keeps it off the prep sheet entirely.

A few terms, each defined once:

• Report: one public account of one interview loop. That covers a review-site entry, a forum post, or a section of an aggregator prep guide.
• Claim: one checkable statement inside a report, such as "four loop interviews" or "a system design round".
• Structure claim: a claim about rounds, their order, durations, or format.
• Content claim: a claim about question topics, specific questions, or tools used inside a round.
• Official baseline: the employer's own prep page for your role and level, plus your invitation email.
• Independent: two reports are independent only if they share no distinctive wording and don't trace back to the same original thread.

The verdicts work at two levels:

Only claims from Admit reports go on the prep sheet. Hold claims, from any report, become questions for the recruiter.

The seven admission criteria, and what each one proves

Each criterion below says what it proves, when it passes, how it can lie, and which verdict a failure triggers. Criteria 1 to 6 grade the report. Criterion 7 grades each claim.

1. Exclusion: not confidentiality-covered or assessment content

• Passes when nothing in the report is pasted from an assessment, a case pack, or other material the author received under a confidentiality agreement.
• Proves that you can use the report at all.
• Lies when a "real questions" post presents itself as a recollection but reproduces a prompt verbatim, with formatting, constraints, and sample inputs no one would remember.
• Fail verdict: Reject the excluded content. If the excluded part is the whole report, reject the report.

2. Dated: the interview month is stated

• Passes when the report gives the month of the interview, not just the date it was posted.
• Proves that you can apply a recency window.
• Lies when the posting date stands in for the interview date. Moderation alone can delay a Glassdoor review by up to 48 business hours, plus up to 24 hours before it appears, and authors often post long after the loop ended.
• Fail verdict: Reject. If you can't date it, you can't window it.

3. Role family match

• Passes when the report covers the same family of work as your posting, for example software engineering for a software engineering loop.
• Proves that the rounds were built to test the same skills.
• Lies when a broad title such as "Engineer" hides a different family, like hardware or support engineering.
• Fail verdict: Reject.

4. Level match

• Passes when the stated level matches the level on your posting.
• Proves that the loop length and the screen type apply to you. At Amazon, the SDE II page describes an online assessment and then four 55-minute interviews. The SDE III page describes a 60-minute phone screen with a senior leader and then five 55-minute interviews.
• Lies when the report title names one level and the body describes another, or when the author gives their own pre-interview level instead of the level they interviewed for.
• Fail verdict: Reject if the level is stated and different. Hold if it isn't stated.

Grade against the posting you actually applied to, not the title you remember. If you tailored the application in Refolk, the posting and the resume version you sent are stored together, so the role family and level you're matching against are on record.

5. Business unit and location match

• Passes when the team or business unit and the location match your loop.
• Proves that local variation, such as a regional hiring process or a team-specific technical round, isn't hiding in the report.
• Lies when a company-wide page groups reports from units that run different loops under one employer name.
• Fail verdict: Hold if either one is mismatched or unstated. Structure is often company-wide, so this isn't a Reject trigger on its own.

6. Provenance: first-person, aggregator, or likely AI-written

• Passes when the author describes their own loop in the first person.
• Proves that a single observation sits behind the claims.
• Lies when an aggregator guide is written in the first person, or AI-polished text reads as personal.
• Labels and verdicts:
  • First-person: pass.
  • Aggregator: Hold. An aggregator can't serve as an independent second report, because you can't see what it was compiled from.
  • Likely AI-written: Reject. A report gets this label when it shows two or more of the three markers in the failure modes section.

7. Corroboration (graded per claim)

• Passes when the claim matches the official baseline, or appears in a second report that is independent and has itself passed gates 1 to 6.
• Proves that the claim isn't one person's memory or one thread's echo.
• Lies when two reports share phrasing or link to the same original post.
• Verdicts: Admit when corroborated and inside its window. Hold when uncorroborated. Reject when the official baseline or the recruiter contradicts it.

How old can an interview report be and still count?

That depends on what the claim is about. Structure claims stay valid longer than content claims, and startup reports go stale faster than reports from large employers. No public source gives a month count, so you write your own two windows down once, before grading, and apply them the same way to every report.

The evidence for splitting the window comes from interviewing.io's interviewer survey of 67 interviewers, 52 of them at FAANG or FAANG-adjacent companies:

• At large employers, the rounds stayed the same but the questions moved. 58% of FAANG interviewers had changed the kinds of algorithmic questions they ask, 21% ask harder questions, and 11% use AI-detection software.
• 67% of startup respondents said AI had meaningfully changed their interview process. At FAANG the figure was 0%.
• One limit: the startup sample covered only about five companies. Treat the startup gap as a direction, not a measured rate.

Three rules follow from this:

1. Set the content-claim window shorter than the structure-claim window.
2. Set both windows shorter for a startup than for a large employer.
3. When the current official baseline confirms a structure claim, the claim's age stops mattering. The official page is the source now, and the report has only pointed you to it.

The platforms won't do this for you. Glassdoor says it does not remove or archive reviews based on their age, so a loop from several years ago still shows with full weight next to last month's.

A platform that never ages out a review has handed the expiry date to you.

Are Glassdoor interview reviews accurate enough to prep from?

They're verified as coming from a real account, and they're not verified as true. The headline scores on an interview page mix every role and level, so on their own they can't support any claim about your loop.

What Glassdoor discloses about its own process:

• Verification: it requires email verification from a permanent, active email address or a valid social networking account for each user who submits content.
• Moderation: a two-step process that can take up to 48 business hours.
• Age: reviews are not removed or archived for being old.
• Accuracy: Glassdoor states that it does not act as a fact-checker in disputes.
• Difficulty score: the survey asks candidates to rate the difficulty of a recent interview from 1 to 5.

What isn't public: how the positive-experience percentage and the average process length are calculated or weighted. Don't treat either as a measurement.

The displayed scores also blend roles:

The Glassdoor Intern page shows 64% positive against a company average it cites as 74.9%. That's 10.9 points lower inside one employer. The company page in the table shows 75.6%, which reminds you that these figures move and shouldn't be quoted as fixed. If you use a displayed score at all, use the role-filtered page and read it as atmosphere, not as a claim.

Selection bias pushes in a known direction. Chip Huyen's analysis of Glassdoor interview reviews found that candidates who receive offers are more likely to write reviews than those who don't, and that junior candidates review more often than senior ones. That inflates "positive %" and leaves senior loops thinly described. No published measure of a negative-experience skew in interview reviews exists, and Huyen's sample size isn't stated, so don't correct for the bias with a number. Correct for it with the level gate.

Can you trust a company-specific interview prep guide?

Use it as a lead, not as a source. Third-party guides disagree with the employer and with each other on basic structure, mostly because they define "round" differently. Grade every aggregator guide Hold, and admit its claims only where the official baseline backs them.

Here is the same Amazon SDE loop from the employer and from three third-party guides:

A candidate who preps from the six-round figure expects two extra loop rounds that are really the online assessment and the phone screen. Amazon's SDE II page describes the assessment separately: 90 minutes for two technical questions, 20 minutes of system design scenarios, and an 8-minute multiple-choice work style survey. The page also says the outcome comes within 5 business days of the loop. When a guide's numbers don't match the employer's, check what the guide counts as a round before you decide which is wrong.

Check the author as well. Titles that sound like expertise are noisy in public profile data. In Refolk's index, 51 people in the United States match "Interview Coach", but in a 25-profile sample only 5 hold that exact current title. The rest include airline captains and a pilot, and two of the sampled profiles list Delta Air Lines as employer.

For scale, the people who actually run these loops are far more numerous and easier to identify by title:

The US count of Technical Recruiters is 30.0 times the UK count. The employer lists come from 25-profile samples only, so read them as illustrations, not rankings. The practical point is that your own recruiter is a better source than an anonymous guide's byline, and that's why Hold claims go to the recruiter.

Which "real interview questions" are off limits?

Anything pasted from an assessment, or from material covered by a candidate confidentiality agreement, is excluded before any other criterion runs. Even legitimate topic claims are weak evidence, because certified assessments rotate their questions.

Why the line sits where it does:

• Confidentiality scope: template interview NDAs advise that confidential information can cover everything learned about the company, including the interview questions.
• Copyright: CodeSignal states that it holds copyright on all content on its platform, and that it can replace leaked questions and request DMCA takedowns.
• Rotation: CodeSignal says its General Coding Assessment includes over 628,000 variations to minimize cheating, and it supports 45 programming languages. A posted question is unlikely to be the one you get.
• Employer rules: Microsoft's hiring page says candidates should demonstrate their own skills without outside assistance unless explicitly permitted.
• Consequences: no employer I could find has formally stated a penalty for using leaked question banks. The only case on record is a public interviewer thread describing a candidate confronted mid-interview after confessing to memorizing leaked questions. Treat that as an anecdote, not a policy.

What survives exclusion is the topic. "A graph traversal problem in the coding round" can be admitted if it comes from a non-excluded report, falls inside the content window, and is backed by something independent. A literal problem statement can't.

The grading procedure, report by report

Grade in this order. Exclusion comes before matching so you never read excluded content closely, and corroboration comes last because it needs the full log. Sources disagree on whether to contact the recruiter before or after collecting reports. I put it second, so the reply can arrive while you grade.

Use one log for the whole employer so a second grader can repeat your work:

POLICY
Employer:
Employer type: large / startup
Structure-claim window (rounds, order, durations): __ months
Content-claim window (topics, tools): __ months
Official baseline: <role-level prep page URL>, saved on <date>
Recruiter confirmation: <reply date> / no reply as of <date>

REPORT
ID:
Source type: review site / forum / aggregator guide
Interview month:
Role family: ______ | match / mismatch / unstated
Level: ______ | match / mismatch / unstated
Business unit or team: ______ | match / mismatch / unstated
Location: ______ | match / mismatch / unstated
Exclusion: pass / struck (reason: ______)
Provenance: first-person / aggregator / likely AI-written (markers: ______)
Report verdict: Admit / Hold / Reject

CLAIMS
- <claim> | structure / content | corroborated by: baseline / report ID / none | in window: Y / N | Admit / Hold / Reject

How report grading goes wrong

Most grading errors share one mechanism: a report looks better than it is because nothing on the page marks its age, its level, its sources, or its origin. Each failure below names what it looks like and the check that catches it.

Stale reports shown at full weight

• Looks like: a detailed, well-written loop description near the top of the page.
• Why it happens: Glassdoor never ages reviews out, so an old loop displays exactly like a new one.
• Check: compare the stated interview month against your window, then compare the structure against the current official page.

Level drift

• Looks like: an SDE II report used to prep for SDE III.
• Cost: you expect four loop rounds instead of five, and an online assessment instead of a 60-minute phone screen with a senior leader.
• Check: read the level in the report title and in the body, and reject on a stated mismatch.

Round-count inflation from aggregators

• Looks like: "6 rounds" in a guide, where the employer lists four.
• Why it happens: the guide counts the online assessment and the phone screen as loop rounds.
• Check: find the guide's definition of a round before comparing numbers.

AI-written reports that pass as personal

• Looks like: fluent, confident, verbose text.
• Scale: a company that sells an AI-text detector reported that AI-flagged employee reviews on S&P 500 company pages rose from 6.18% to 29.45% over two years. That covered employee reviews, not interview reviews, and the company measured it with its own tool. No detection study of interview reviews specifically has been published.
• Why not use a detector: the best detector in one peer-review benchmark caught only 70% of fully AI-generated reviews, and research finds detection works better in aggregate than on an individual text.
• Check by hand: three markers drawn from the peer-review literature (verbosity, generic language, lack of specificity), made concrete:
  1. No team, business unit, or location named.
  2. No round-by-round sequence.
  3. Language that would fit any employer unchanged. A report with two or more markers is labelled likely AI-written and rejected. The label is about specificity, not about authorship you can prove.

Blended display scores

• Looks like: "this company is 75% positive, so the interviews are fair."
• Why it misleads: the company-wide figure mixes roles. At Glassdoor itself, the intern figure sits 10.9 points below the cited company average.
• Check: use role-filtered pages only, and never put a display score on the prep sheet.

False corroboration

• Looks like: two reports making the same claim.
• Why it misleads: one may copy the other, or both may summarize the same original thread.
• Check: look for shared distinctive phrasing and shared links.

Leaked "real questions"

• Looks like: a verbatim prompt posted as a recollection.
• Why it misleads: it may be copyrighted assessment content, and certified tests rotate variants, so the question may never appear in your loop.
• Check: reject literal assessment questions under the exclusion gate, and keep only a topic that something independent backs up.

Keeping the prep sheet current through the loop

The prep sheet changes whenever the official baseline or your recruiter tells you something new. Re-grade every claim that new information touches, and resolve Hold claims by asking rather than guessing.

Send the Hold list as one short, specific message. Recruiters can answer structure questions without revealing content, and this message only asks for structure:

Hi [name],

Thanks again for scheduling my [role, level] interviews. To prepare properly, could you confirm a few logistics?

1. How many interviews are in the loop, and roughly how long is each?
2. Is there an online assessment or phone screen before the loop, or is that already behind me?
3. Will any round cover [system design / a domain topic] for this level?
4. Which tools or languages am I allowed to use, and is any outside assistance permitted?

Happy to work from whatever you can share. Thank you.

To keep the work current:

• When the reply arrives: promote confirmed Hold claims to Admit and reject any the reply contradicts.
• When the employer's page changes: re-check every structure claim against the new version and update the saved date in your log.
• After your own loop: note which admitted claims held up. That's the only accuracy check you'll ever get on this employer's reports.

Prep hours are limited even after grading. Refolk's fit score against the posting you applied to shows where your record is thinnest, which tells you which admitted topics deserve the most drill time.