Skip to content
openproduct
Search

What we removed, and how often we were wrong

Every act of moderation on openproduct is written to an append-only record before it takes effect. This page is generated from that record — not written by hand from it — so what is published and what was done cannot drift apart.

21 August 2026 to 19 November 2026.

Why a rolling window
The last 90 days, rolling. Not a running total since launch: a cumulative figure only ever goes up, which makes a bad quarter permanently diluted by every good one before it. That is a comfortable property for us and a useless one for you.
Acts taken against something
0
Acts undone
0
Undone as a share
—

A dash means there is nothing to take a share of — not that the share was nought.

Which number to read
The number worth reading is the second one. Any estate can publish how much it removed; the figure that says whether its judgement is any good is how much it put back.

Every act, by kind

Including the decision to do nothing.

None in this window.

Why the decisions to do nothing are counted
An estate that publishes only its removals is publishing a boast.

Appeals

Filed by the people the acts were about, refusals counted beside upholds.

None in this window.

Why refusals are counted, and why the two figures can differ

A page showing only the appeals we granted would be the same boast in a different section.

An act sometimes gets undone by somebody else while the appeal is waiting, and upholding it then changes nothing except the record.

Reports

What people told us about listings, by where each one got to.

None in this window.

Why the waiting ones are shown
A growing tail here is a fact about our capacity, and leaving it out would hide exactly the failure worth knowing about.

Upvotes not counted

Upvotes that did not reach a listing’s public total, by the shape that produced them — never by listing or by person.

None in this window.

A discounted upvote is still recorded, and a later review can restore it.

Why never by listing or by person
A per-listing figure is an accusation about a specific product’s supporters, made in public, by arithmetic.

Automatic checks

What the checks at /docs/checks raised, and what a person decided about each. None of them can remove anything.

None in this window.

Why the dismissed share matters
That share is the checks' false-positive rate measured against real listings rather than against a corpus, and it is the number the published one has to answer to.

Where these come from

Acts
The append-only moderation record. Every handler that changes anything writes here first and acts second, so an act that was carried out but not recorded is not a gap in this page — it is a bug in the code that would show up as an unexplained state.
Reports
The report queue. Counted by where each one got to, including the ones a person read and decided needed nothing — an estate that publishes only its removals is publishing a boast.
Appeals
The appeals table, by where each one got to, with refusals counted beside upholds for the same reason. The reversal figure is counted separately and on the date it was decided rather than the date it was filed, so a reversal appears in the window it actually happened in.
Endorsements not counted
The endorsement integrity record, by reason only. Never by listing and never by person: a per-listing figure is an accusation about a specific product's supporters, made in public, by arithmetic.
Automatic checks
The flag queue, by kind and outcome. The dismissal share is the production false-positive rate of the checks published at /docs/checks.

A zero here means we did not act, not that we did not look. Everywhere else on this site a missing number is shown as missing rather than as nought, because an unobserved thing and an empty thing are different. This page is the exception, and it earns it: these are records of our own acts, written by our own code before the act takes effect.

What this page will not do

Nothing on this page names anybody — not a listing, not a person, not a number small enough to work one out. A specific act is visible at the address it happened to, which is where the people affected by it are looking.

Naming a listing we acted on
A list of removed products is a permanent public accusation, readable by people who will never see the case for it, about makers who mostly are not here to answer. The act is already visible at the listing's own address, dated, to the people actually looking at it.
Breaking the numbers down far enough to identify anybody
Three removals in a category with four listings is a name. Every figure here is estate-wide for that reason, and staying coarse is the cost of being able to publish anything at all.
Publishing the reasons we gave
A reason is written to be read by the person it is about and by whoever reviews the decision later. Published in bulk it becomes a corpus of judgements about individuals, none of whom agreed to be a case study.

If one of these acts was about your listing, you can appeal it, and the outcome is counted here whichever way it goes. It is offered beside every decision, in the studio for the listing it was about. The checks that run without a person are published at /docs/checks.