Seller Records: How We Keep Them
Why records matter
A link list decays, and the seller behind a row is usually the part that changes without notice. A row can keep its code, its price band and its photograph while the direction supplying it is replaced twice in a season. Records exist to give that change somewhere to show up, in a form a reader can compare against what the page said last time.
The alternative is memory. Buyers who follow lists for a year accumulate impressions of which directions were steady and which were not, and those impressions are unusable: they carry no date, no source and no count. A record replaces an impression with four checkable things — what the direction is known for, which class of source the observation came from, how many observations were read, and when.
Records also make refusal possible. A desk that keeps no records can answer a question about supply only with an opinion, and an opinion about a named shop is exactly what this site does not publish. Keeping the note at the level of a direction, with the count and the date attached, is what allows a page to say something specific without making a claim it cannot support.
The scope is deliberately small. Twenty records exist at the current check date, covering eight category directions, and nothing here is a ranking. A record is not a recommendation, a warning or an endorsement; it is a dated summary of what public sources show. Readers who want a shortlist will not find one, and readers who want to know how a summary was built will find every step of it.
Inclusion evidence standard
Nothing enters the roster without a public source that can be re-read later. A direction qualifies when at least three separate public observations describe the same construction features, and when those observations can be tied to a category rather than to a single product. One vivid thread is not enough; the desk waits for repetition, because repetition is the only quality control available without buying anything.
Every record is capped at the third-party evidence level or the platform level below it. That ceiling is structural rather than provisional. This desk does not purchase from the directions it records, does not inspect their output and holds no transaction data, so a higher level could not be claimed honestly even if a reader supplied private material. Where the only available source sits below the cap, the record is not created.
Sample size is stored as the count of observations an editor actually read, not as an estimate of how common something is. A record resting on thirty-seven readings says thirty-seven, and a record built from the desk's own analysis says zero, which is printed as such rather than hidden. The distinction matters because a count of readings can be audited by anyone and an estimate cannot.
Dispute fields follow the same rule. Where a disagreement type has been publicly described often enough to be recognisable, the record names the type and stops there: a size run that ships narrower than the chart, plating that fades at a contact point, adhesive that releases along a sole edge. No record names a shop, quotes a message thread, or describes a case in enough detail for a reader to identify a party.
Review cadence design
Each record carries a due date, set ninety days out from its most recent reading. The interval is chosen against how quickly the material underneath it moves: a direction's construction features change when its supply changes, which is a seasonal event rather than a weekly one. Ninety days also matches the shortest span over which a supply change becomes visible in public writing rather than in a single complaint. A shorter window would produce churn without producing new information, and a longer one would let a replaced supply run under a stable description for most of a year.
Re-reading a record means opening the same class of source again and asking three questions. Do the observations that produced the record still exist where they were found? Do they still describe the same construction features? Has a dispute type appeared that the record does not mention? A record answering yes to the first two and no to the third is confirmed in place, and only its date changes.
Sourcing a record a second time is not always possible, and the cadence treats that as a finding rather than a failure. Where a source has been withdrawn or a direction can no longer be identified, the window closes without a confirmation. One missed window keeps the record live with a note attached to it; two missed windows end its current status, which the next section describes.
The cadence also sets the order of desk work. Records approaching a due date are read before new proposals, because a stale record misleads more than a missing one. A reader who wants to know when a specific record was last touched can read the check date on the roster page, which prints the date and the length of the window beside every entry.
Removal triggers
Four conditions end a record's current status, and each is a property of the evidence rather than a judgement about the direction. The first is a source that cannot be reached again. The second is a direction that has stopped being identifiable, usually because listings of the kind the record described are no longer published. The third is a dispute type whose public reporting has changed materially. The fourth is a second review window closing without a re-reading.
None of the four can be triggered by a request. A direction that objects to its record cannot have it removed, for the same reason a direction that admires it cannot have it extended. The same symmetry protects a direction from being promoted on the strength of a request. That rule is stated in advance rather than applied case by case, because a record whose life depends on the preference of its subject is not a record of anything.
Removal is not deletion. A retired record stays on the roster with its earlier wording intact and the reason recorded against it, so a reader who saw the previous version can find out what changed and when. Two records moved to retired in the most recent archive batch, one because its direction stopped being identifiable and one because its only public source was taken down.
A record can also leave current status without retiring. Where a direction has split into two lines that can be described separately, the desk replaces one wide record with two narrower ones, each carrying its own count and its own date. That is a change of shape rather than a withdrawal, and it is logged the same way so the earlier wording stays reachable.
Withdrawal transparency
Every change to a record is written down with three fields: what moved, the date it moved, and the reason. The log lives in the change archive beside the page-level changes, so a correction to a threshold and a withdrawal from the roster appear in one sequence. Nothing is altered quietly, and nothing is removed without a line explaining it. The same log records why a record was split, why a due date was missed, and which batch carried the change.
The reason is published in the terms the rule uses rather than in narrative form. A reader sees that a source could not be re-read, or that a second window closed, rather than a paragraph speculating about what a direction might have done. That restraint is deliberate: the desk holds no transaction data, so any narrative about cause would be invention dressed as reporting.
Readers can push back on a withdrawal the same way they push back on a threshold. A report naming the record, the observation and the source is read in order of arrival, and where it changes a status the change plus its date appears in the archive. Where it does not, the reasoning is recorded and the record stays as it stands. That route is the only one able to change a record from outside the desk, and it is open to anyone.
The archive is also where reinstatement shows up. A record returns to current only when the evidence conditions that ended its status have been met again, which in practice means a fresh reading with a fresh count and a fresh date. There is no queue and no automatic restoration after a period of time; a retired record that nobody can re-read stays retired.
Limits of our method
Four limits apply to every line in the roster. Coverage follows attention, so a direction with no public discussion is missing rather than cleared. Public discussion follows complaint, so a direction whose output is usually fine produces few observations and may never reach the threshold. The level cap means no record claims inspection. And the count of readings is not a rate, so it cannot be read as a proportion of anything.
The desk also cannot tell a change of supply from a change of discussion. A quiet season looks identical to an improved one when the only instrument is public writing. Where the two cannot be separated, the record notes that the direction was not re-readable rather than reporting an improvement nobody observed. Readers are told which of the two happened, because the difference changes what the record is worth.
What the method can do is narrower than what a buyer wants. It can say that a direction has been publicly described in a certain way, on a date, across a countable number of readings. It cannot say what the next parcel will contain. That gap is not a temporary shortfall to be closed with more data; it is the boundary of a desk that reads in public and buys nothing.
Where the boundary matters to a decision, the reader is the one who closes it, using the judging points, the thresholds and the delivered-cost tools that depend on no record at all. A reader looking for a guarantee will not find one here, and the page says so rather than implying it. The roster is one input among several, and it is placed here with its method attached so that its weight can be set by whoever is reading it.
Fields carried on every roster record
Data contract. These are eight of the nine fields the roster dataset stores for each record, chosen because each one can be checked by a reader from the published page; the field names match the dataset. Twenty records carry them at the current check date, and no record stores a shop name, a link, a price or a rating. Inclusion follows the evidence rule on the method page. Last checked 2026-09-25.
| Item | Reading |
|---|---|
| direction | A category direction, never a shop name, a link or a location |
| knownFor | Up to three construction or material features the direction is publicly associated with |
| evidenceLevel | Fixed at the third-party or platform level; no record claims inspection or transaction data |
| sourceType | Which class of public source the observation came from |
| sampleSize | Count of public observations actually read; zero means editorial analysis |
| disputeNotes | A publicly reported dispute type, or null where nothing has been reported |
| lastChecked | The date the record was last re-read against its sources |
| status | Current, under review or retired; a retired record stays visible |
Sources and checks
- Editor analysis — The record fields, the inclusion rule, the ninety-day cadence and the withdrawal conditions; written by the deskno sample
- Public community thread — Public observations re-read across the twenty records during the current windown=214
Last checked · next review due · status current.