How the fifty were picked
The sheet was not chosen for being representative. It was chosen because it was still reachable on 2026-09-29, still carried a wrapper column, and still had rows nobody had tidied up. Fifty consecutive rows were taken from the top and copied out in order. No row was skipped for looking broken and none was added for looking healthy. The picking rule was position, not condition. Anything else would have meant choosing the sample after seeing the results, which is a different exercise with a different answer.
Scaled against the published window, fifty is a large slice. The 2026-W39 table carries ninety-three records in total, split between thirty-one wrapped addresses from sheet pages and sixty-two source addresses from a directory listing. Thirty-one of the fifty rows read here are those sheet records. The rows that carried no readable product number are not in the table either, and the reason is the same for all of them: no rebuild could be written, so no status could be published. A row that cannot be identified is not the same thing as a row that failed, and the totals keep the two apart.
Nothing was opened twice and nothing was opened from a stored copy. Each row got one attempt at the wrapper, and where a rebuild existed, one attempt at the rebuilt address. Retrying a dead host five times would have produced a bigger number and told us less, because the extra attempts measure patience rather than availability. One attempt per address is the rule this site applies to the public sample too, and the checking method page sets out where that line sits.
Picking rows from one sheet answers a narrow question well. It tells you what an unmaintained sheet looks like when somebody finally opens it. It does not tell you what a maintained sheet looks like, because this one had not been touched in months. The directory side of the window covers the opposite case, and the two are worth reading next to each other. One is a record of neglect and the other is a record of housekeeping, and neither one replaces the other.
What happened on each attempt
Every wrapped address failed in the same way and at the same stage. The connection never completed, so no page could be read, no product name could be confirmed and no error message from the platform was available to interpret. That is the least informative kind of failure and also the most consistent one. Fifty rows, fifty identical outcomes at the wrapper, no partial reads and no ambiguity to argue about. The wrapper is what stopped; nothing behind it had been tested yet.
The second attempt is where the rows separated. Thirty-one rebuilt into a source address and every one of those thirty-one answered normally on the first try, landing on the host the marker named. The rest never got that far, because no product number could be read from them. No marker was guessed for those rows; guessing would have produced workable addresses for the wrong listings. A wrong address that opens is harder to catch than a dead one, and it travels further once it is in a sheet.
Not one rebuilt address returned a missing page. Across the thirty-one there were no 404s, no takedowns, no verification screens and no login walls. Nothing was fenced off and nothing had been pulled. Whatever else the shutdown did to the platform, it did not touch the listings these rows pointed at, at least not on the afternoon they were opened. That result is about survival, not about quality, price or whether the seller is still trading. A listing that still loads is a listing that still loads, and that is the whole of the finding.
Timing deserves a line of its own. All fifty were read inside one sitting, which keeps the comparison fair but narrows it to a single point in time. A row read in the morning and a row read at night would already be two different measurements. The whole set shares one afternoon in the 2026-W39 window, and that is the strongest claim the numbers can support. Anything longer would need a second sitting, and a second sitting is a different sample.
The numbers that came out
Thirty-one rebuilt addresses answered. Thirty-one wrapped addresses failed at the network level. The remaining rows were never testable, because nothing in them identified a listing. Those numbers account for every row in the set, and they are the only results this exercise produced. Nothing was classified as moved, nothing was fenced off and nothing was left sitting in an unknown pile waiting for a second look. Zero unknowns is unusual, and it is a consequence of the picking rule rather than of the sites involved.
The published window tells a slightly bigger story with the same shape. Ninety-three records, ninety-three resolutions, zero moved, zero gone, zero fenced off, zero unrecorded. The directory half of that window produced the same clean sheet of results, on the same day, from a list that is actively maintained by somebody else. Two different sources of addresses, one afternoon, and not a single missing page between them. That agreement is the most useful thing the period produced.
Exclusions were the only real subtraction. One hundred and thirteen directory addresses were dropped during collection after the host stopped accepting connections, and they were left out of the table rather than written down as unknown. Excluding them makes the sample smaller than the reading list and keeps the numbers honest, because a refused connection says something about the collection window, not about the listing. The same rule kept rows that identified nothing out of the loss column. A smaller honest set is worth more than a larger set with a guess folded into it.
Read the totals as a description of one day. Ninety-three out of ninety-three is not a claim that the shops behind these listings are healthy, solvent or still trading. It is a claim that the pages answered when asked. Everything else people usually want from a sheet, including price, stock and whether an order would actually be fulfilled, sits outside what was measured. A page that answers is a page that answers; it is not a promise of anything after that.
Where this sample misleads you
Fifty rows from one sheet is not fifty rows from the internet. This sheet was public, indexed and old enough to have been copied, and sheets like that are the ones that survive long enough to be read at all. A sheet deleted in July, or one kept behind a login, has no chance of appearing here. The sample describes survivors and should be read that way. Survivorship is not a flaw in the reading, but it is a hard ceiling on what the reading can say.
The rows that identified nothing are doing quiet work in these totals. They are not failures and they are not unknowns; they are rows this site declined to guess at. Read the fifty as thirty-one measured and the rest untouched by the reading, and do not let a total of fifty imply that fifty verdicts were reached. A verdict needs a rebuild, and a rebuild needs a marker this site has observed with its own eyes. Anything past that point is arithmetic dressed up as evidence.
One afternoon is one afternoon. All fifty rows answered inside a single sitting, so the set can say nothing about how stable those answers are. A listing that answered on 2026-09-29 might have gone quiet the next morning, and this reading would not notice. Repeat visits are the only thing that turns a snapshot into a trend, and repeat visits are what the weekly snapshots page exists for. One column of one week is a starting point, not a chart.
The most common misreading is the cheerful one. A sheet where everything resolves looks like a sheet where everything is fine, and those are different sentences. Resolution says the page was there. It says nothing about the seller, the item, the price you were quoted last spring, or whether anybody would still ship it to you. Keep the two apart and the sheet stays useful for longer. A dead wrapper and a dead listing are not the same failure, and only one of them was measured here.