How it works
The whole mechanism, in the open
A review site is only worth reading if you can check how it reaches its conclusions. This page is the calculation, the safety scan, the limits of moderation, and an honest account of what anonymity does not cover.
What happens when you post a review
- 1
You write it
Ratings across six axes, what happened in your own words, and red flag tags from a fixed list. Anonymous is the default and stays the default unless you change it.
- 2
The safety scan runs before anything publishes
It looks for personal data about individuals and for accusations of crime stated as fact. If it finds something, you are told what tripped and why, in the composer, before you post.
- 3
Clean content publishes immediately
No queue, no delay, no approval. Most reviews never touch a moderator.
- 4
Flagged content is held, not deleted
It goes to a moderator queue with your text intact. A human reads it and either publishes it as written or tells you what has to change. Nobody edits your words for you.
- 5
Your review joins the company's score
Your red flags and ratings are folded into the danger score with everyone else's, weighted by whether your employment is verified and damped by how many people have spoken so far.
How the danger score is calculated
One number from 0 to 100 per company. It is not an average of stars. It is built from two things — how heavy the red flags are and how bad the ratings are — and then held back until enough people have spoken.
Step one: red flag pressure
Every flag has a fixed weight. Within a single review the heaviest flag counts in full, the second at half, the third at a third, and so on, capped at 2. Tagging everything on the list does not multiply your effect — it dilutes it.
- Wage theft / unpaid wages1.00
- Illegal practices1.00
- Harassment tolerated1.00
- Discrimination0.95
- Retaliation against speaking up0.95
- Unsafe working conditions0.90
- Toxic leadership0.80
- Unpaid overtime expected0.75
- Broken hiring promises0.70
- Chaotic layoffs0.70
- No work-life balance0.60
- Micromanagement / surveillance0.50
- Revolving-door turnover0.45
- Misleading interview process0.45
- No growth or promotion path0.40
Step two: whose review it is
A reviewer with verified employment at that company is weighted 1.5×, in both the pressure and the total it is divided by. Unverified reviews still count in full at 1×. Verification is a bonus for proof, not a gate on speech.
Verify your employment with a work email or evidence a moderator checks.
Step three: the mix
62% of the raw score is red flag pressure. 38% is how far the average overall rating sits below five. Specific, structured warnings count for more than a mood.
Step four: the confidence ramp
This is the part that stops the site being a weapon. Whatever the raw pressure comes to, a company only carries a fraction of it until enough people have said the same thing. One person cannot brand an employer. Twelve consistent people can.
One review carries 40% of the score it would otherwise produce. Five carry 62%. At twelve the damping is gone. A company cannot be branded high risk by one furious night.
Step five: where the number lands
- Looks safe
- 0–21
- Mixed signals
- 22–44
- Caution
- 45–69
- High risk
- 70–100
The band is what a job seeker actually reads on a company page and next to every job listing that company posts.
Worked example
A company has one review. It is from a verified employee, it rates the place 2 out of 5 overall, and it tags wage theft (weight 1.00) and micromanagement (0.50).
- Flag pressure for that review: 1.00 + (0.50 ÷ 2) = 1.25 out of a possible 2.
- Rating pain: (5 − 2) ÷ 4 = 0.75.
- Raw score: (0.62 × 0.625) + (0.38 × 0.75) = 0.67, so 67 out of 100 before damping.
- Confidence at one review: 40%.
- Published danger score: 27.2 — mixed signals, not high risk.
If eleven more people report the same pattern, the damping is gone and the same pressure publishes as roughly 67 — high risk. That gap between 27 and 67 is the difference between one bad night and a pattern, and the site is built to tell them apart.
What the safety scan holds, and why
The scan is not there to soften what you say about a company. It exists for two reasons: it keeps private people from being doxxed, and it keeps true reports from being knocked down on a technicality. Every finding is shown to you in the composer before you publish, with the exact text that matched.
Email addresses
HeldAn email address identifies a person. Posting one is the fastest way to get a true review taken down and to hurt someone who is not the employer.
Links to personal profiles
HeldA link to someone's LinkedIn or Instagram turns a workplace complaint into a pile-on aimed at an individual.
Bank or account numbers
HeldFinancial identifiers, yours or theirs, do not belong in a public review under any circumstances.
A named individual in a role
HeldPhrases like "my manager Jane Doe" name a private person. Their role carries the same weight for a reader and keeps the legal risk off you.
An accusation of crime stated as fact
HeldCalling someone a thief is a claim you would have to prove. Describing what you saw — "I was not paid for 62 hours of overtime in March" — is stronger and far harder to attack.
Something that looks like a phone number
WarnedThis one is noisy, so it only advises. You are shown what matched and you decide.
Threats of violence
BlockedThis is the one thing that does not publish at all. This place exists to warn people, not to threaten them.
Held means held, not deleted
What a moderator can and cannot do
Moderator powers are deliberately narrow, because the obvious way to ruin a site like this is to give someone a quiet button that makes criticism go away.
Can
- Publish something the scan held, exactly as you wrote it.
- Hide or remove content that breaks the community rules, and tell you which rule.
- Approve or reject an employment verification after checking your evidence.
- Warn or ban an account for brigading, astroturfing, or targeting individuals.
- Dismiss a report from an employer without touching the review.
Cannot
- Edit a word of what you wrote. There is no function for it.
- Reveal who wrote a review to an employer, a recruiter, or another member.
- Take a review down because a company asked, complained, or offered money.
- Set, nudge, or override a danger score by hand. It is computed, not decided.
- Act without leaving a record — every moderation action is logged against the moderator who took it, and the author is notified.
What anonymity protects, and what it does not
Most people reading this are still employed, still owed money, or still need a reference. You deserve the real answer rather than a reassuring one.
What it does
- Your name is not attached. Reviews render as “Anonymous employee” unless you deliberately choose otherwise.
- No employer feature exposes reviewer identity. Claiming a company page gets an employer a public reply box and their own numbers, nothing more.
- Your verification evidence — work email, document, note — is visible only to moderators, never on your profile and never on the review.
- Private and secret groups are filtered in the database query, not hidden with CSS. A secret group does not appear to people outside it at all.
What it does not
- It does not make you unrecognisable. If you were the only night-shift supervisor in a team of six, the details will identify you no matter what name sits on top.
- It is not encryption. Accounts exist in a database, and a lawful order compelling disclosure is a real thing that can happen to any site.
- It does not protect a screenshot. Anything published can be copied by anyone reading it, including your employer.
- It does not cover what you post elsewhere. Repeating your review on a public profile ties the two together for anyone looking.
If your team is small, write differently
The two rules on money
No salary, no listing
Job posting here requires a minimum and a maximum. The form rejects an empty salary with “Salary is mandatory here” and there is no way around it, for any employer, at any price. Every listing also carries the employer’s danger score.
Open jobs →No median from thin data
Under three reports for a role we refuse to print a median and show the raw entries instead. A confident-looking average built on two data points is how people end up negotiating against a number that was never real.
Pay reports →Still want to check something
The community rules cover what makes a review hold up. The employer page covers what a company can do when it finds one it does not like.