Withdrawal Conditions: When We Drop a Record
Signals before inclusion
The cheapest withdrawal is the one that never has to happen, which is why most of the filter sits before inclusion. Entry asks for corroboration: a direction has to be described by three or more unconnected public accounts that agree on how the thing is built and that can be attached to a category rather than to one listing. A single thread, however specific, waits outside until something else corroborates it.
Two pre-entry signals end a proposal outright. One is gated material, because anything needing an account, a login or a purchase to open is unusable when the desk has to be able to return to it later. The other is a description that only works for a single product, since a record describes a direction, and a direction has to outlive the listing that revealed it.
A third signal is softer and matters more over time. Proposals resting on writing that appeared in one burst and then stopped are held rather than entered. A burst can be genuine enthusiasm or a coordinated push, and the desk has no way to tell the two apart at the moment of reading. Waiting one review window costs nothing and separates them, because the second kind does not survive a quiet season.
Where a proposal passes the entry filter, what is recorded is deliberately thin: the direction, its known features, the source class, the count of readings and the date. Nothing in that set describes a party, and nothing in it can be invalidated by a later dispute about a particular order. The rules further down this page operate on those fields alone, which is what keeps them applicable to every record in the same way.
Warning signs during tenure
A live record is re-read every ninety days, and the re-reading watches for three changes. One is a source that has disappeared from the web. Another is a direction that has stopped appearing in the sorts of listings the record was built from. The third is public reporting that has shifted away from the features the record lists and towards a dispute type it does not mention. All three concern the state of the public record rather than anybody's conduct.
A vanished source is the most common warning and the least dramatic of the three. Threads are deleted, accounts close, forums reorganise their archives, and a link recorded eighteen months ago can stop resolving without anything being wrong with the direction it described. When the observations behind a record cannot be opened again, the record keeps its history and loses its future: nothing can confirm it at the next window, and a second missed window ends its current status.
A direction going quiet is ambiguous in a way the desk cannot resolve. Fewer listings can mean a line has stopped, or that it has moved to a channel that leaves no public trace. Both are recorded as not re-readable rather than as an improvement or a decline, because the instrument cannot distinguish them and a record that guessed would be worse than one that admits the gap.
The third warning is the one that leads to escalation. Where public writing about a direction has moved from construction features to a repeat dispute type, the record is not immediately withdrawn; it is re-read with that dispute in mind. Where the dispute type holds up across several independent observations, the record's disputed field is updated and a note is added, which is a change to the record rather than an end to it.
Escalation ladder
Escalation runs in four steps, and each one is triggered by evidence rather than by how serious a report sounds. Step one is a note: the record stays current while a specific claim is checked. Step two is under-review, used when a re-reading cannot settle a question at the next window. Step three is a correction to a field. Step four is withdrawal. No step can be skipped, and no step is reached by repetition alone.
Notes and under-review status appear on the record itself, which is the point of having them at all. Somebody opening a record during step two sees the status, the date it changed and the question that is open. Hiding an unresolved question inside an internal queue would make the published status meaningless, and the desk would rather show a record in doubt than a record that looks settled.
A correction is the outcome of most escalations, which is worth stating because removal gets more attention than it deserves. Where a dispute type appears that the record did not carry, the field is added. Where a direction's described features have changed, the description is rewritten and the earlier wording is kept in the archive beside the new one. Both outcomes leave a live record behind, which is the ordinary result of a re-reading that found something to fix.
Withdrawal is the last step and the narrowest. It applies when the evidence cannot be re-read, when the direction cannot be identified, when a dispute type has changed the picture materially, or when two windows have closed without a check. Every one of those conditions rests on the state of the record's evidence, and none can be created by a party asking for an outcome in either direction.
Removal and notice
Removal moves a record to retired and leaves it in place. The entry keeps its earlier wording, gains the reason in the terms of the rule, and keeps the date the change was made. The identifier, the category and the evidence level stay as they were, so an old note about the record still resolves to the right entry. A reader arriving after the change sees a retired record with an explanation rather than a gap where a record used to be.
Notice is published rather than sent, and it appears in three places at once. The status changes on the roster page, the batch entry in the change archive records what moved and why, and the affected record keeps the reason attached to it. No notification goes to a party named in the record, because no record names one. The audience for a withdrawal is the readership, and the entry has to make sense to somebody who never saw the earlier version.
A withdrawn record keeps its identifier, which is what makes a later reinstatement legible. Where a record is re-entered after its sources become readable again, the page shows the earlier retirement beside the new check date rather than presenting a fresh entry. A reader can therefore see that a direction was once unverifiable, and can weigh that alongside whatever the current reading found.
The retired entry is not hidden from search or from the roster listing. Anyone who wants only live entries can filter them out, and anyone who wants the reason something disappeared can open it. Keeping retired records visible also blocks a quiet form of drift, in which a lightly sourced record is removed and later re-entered as though it had never been questioned. That is the practical reason the retired state exists rather than a delete function.
Reinstatement bar
Reinstatement is not automatic. A retired record comes back only if the conditions that pushed it out can be satisfied a second time, and satisfying them means a new reading with a new observation count and a new check date, taken from material anybody can open. Elapsed time on its own changes nothing, and a direction that stays unreadable across three consecutive windows stays retired rather than reverting by default.
The bar is higher for a record withdrawn over a dispute type. A fresh reading has to show that the public picture has changed, not merely that no new complaints appeared. Silence after a dispute is not evidence of repair when the instrument is public writing, and the desk will not read an absence of posts as an improvement. Something has to be present that was not present before.
The return is recorded with the same three fields as the withdrawal: the record, the date, and the reason the evidence now supports a current status. Somebody comparing two versions of the same record can see the whole path, including the period when the desk could say nothing about it. That path is the most useful part of a long-lived record, because it shows what the method did when it had no information at all.
One limit closes the rule. Reinstatement is not a prediction, and it does not upgrade a record's evidence level. A direction that returns to current sits at the same third-party ceiling it occupied before, with the same inability to say anything about the next order placed with it. The status describes the state of the public record and nothing beyond it.
Withdrawal conditions and the action each one triggers
Rule table. The conditions below are the ones that end a record's current status, and each has been applied at least once across the three most recent archive batches, which publish the count of records affected. No condition can be triggered by a request from a party named in a record, and none is applied without a dated entry in the change archive. The supporting reading covers 20 records re-read against their sources in the current window. Last checked 2026-09-27.
| Item | Reading |
|---|---|
| Source no longer resolvable | Retire; state the date the source was last reachable |
| Direction no longer identifiable | Retire; the earlier wording stays visible beside the change |
| Dispute type materially changed | Move to under review and re-read before any current status returns |
| Second review window missed | Retire once the second window closes without a re-reading |
| Evidence falls below the record level | Retire; a third-party-only record has nothing left above the floor |
| Direction splits into two lines | Split into two narrower records instead of one widened one |
| Reader report upheld | Correct the field named, log the date, keep the record live |
Sources and checks
- Editor analysis — The four withdrawal conditions and the escalation ladder; written by the deskno sample
- Reader report — Roster records re-read against their own sources during the current window, each counted oncen=20
Last checked · next review due · status current.