The most recent check
The 2026-W39 snapshot is the only published period on this site so far. It carries 93 records: 31 wrapped addresses collected from public sheet pages, and 62 source addresses taken from a directory list. The check ran on 29 September 2026, and the weekly snapshots index is where each period lands as it is published.
Results were flat in the least interesting way. The 31 wrapped addresses all failed at the wrapper itself, and the addresses rebuilt from them all answered. The 62 directory addresses all answered too. Status counts came out at 93 resolved, with moved, gone, shielded and unrecorded sitting at zero across the period.
The one before it
There is no previous period. This is the first snapshot this site has published, and the comparison column stays empty until a second one exists. Filling that column with numbers nobody measured would be worse than leaving it blank, because a fabricated baseline makes every later period look like it moved.
What the slot will hold is narrow and specific: for each recurring address, the status it carried last period and the status it carries now, with a note when the two differ. Addresses that were never checked before are listed as new rather than as changed. The checking method page sets out how a status gets assigned before any of that comparison happens.
What flipped between them
Nothing flipped, because there is nothing yet to flip against. The honest reading is that one period describes a state, not a direction. Two periods describe a movement, and only after that does the word trend mean anything here. A single period is a photograph, not a slope.
The method itself is versioned, so a later change to how addresses are collected or how a status is assigned can be seen rather than inferred. If the sampling rule changes, the comparison column will say so, since a movement caused by a rule change is not a movement in the sheets. That distinction is the whole point of publishing a method version beside a count.
Reading the direction of travel
The test to watch is not whether counts match week to week. It is whether the counts move at all. Flat counts across two periods say the source listings are holding. A nonzero gone column would be the first real signal that listings behind the wrappers are starting to disappear.
The pattern already visible is the one worth tracking: the wrapper breaks while the listing survives. If that flips, and rebuilt addresses start failing too, the advice on this site changes with it. Until a second period exists, treat any claim about the sheet getting better or worse as unmeasured.