How a play bulletin proof orders five cards
Checked: the sort rule for this proof, not a sixth app. Result: one key, the average rating printed on the Netherlands card, highest first. Ties would have kept the order the URLs were supplied. There were no ties on 16 October 2026.
Fields that were allowed to move the order
Only the average. Rating count, file size, age line, update date, and the presence of a price were copied into tables and then ignored by the sort. That is deliberate. Mixing them would hide the rule. A reader who disagrees can reorder the same table. The numbers will not change; the sequence will.
Royal Match shows why the rule matters. It has 79,390 ratings and the lowest average in the set, so it is rank 5. A sort by count would invert the list. The proof does not do that, and the ratings table keeps the count visible so the alternative is obvious.
Fields that were refused
Install totals are not on App Store cards. Writing "1M+" because a title feels popular would fail the protocol. Price was refused for four cards because the lookup returned null. Language count, bundle id, and version were stored as identifiers, not as quality.
What a later pass changes
A new line in the check archive copies the five averages again. If an average crosses a neighbour, the rank changes and the corrections page gets a dated sentence. Until that happens, the order on the home proof is the 16 October 2026 order.
The five averages on 16 October 2026 were 4.85, 4.83, 4.64, 4.59, and 4.54. The first gap is 0.02. The second gap is 0.19. The third is 0.05. The fourth is 0.05. A reader can see that ranks 1 and 2 are close, and that rank 3 is not close to them. The protocol does not add a "cluster" label. It keeps the sequence.
File sizes in the same order are 437.5 MB, 296.3 MB, 987.7 MB, 432.7 MB, and 318.9 MB. If size had been the key, Grindstone would be first and WHAT THE GOLF? last. That list is not the list on the home page. Showing both is the point of a protocol: the unused key stays visible so the choice is checkable.
Age lines do not sort the page either. The two 4+ cards happen to be ranks 1 and 2. That is a result of the averages, not a rule that 4+ outranks 12+. Royal Match is 12+ and last. Grindstone is 12+ and third. A future card with a 4+ line and a 4.10 average would sit at the bottom.
The home page repeats the rule in a timeline, a filter, a ranking, a verdict table, and a rating table. Five views, one sort. The filter does not reorder. It hides cards whose price field does not match the chip. Hiding is not a new rank. With JavaScript off, every card stays on the page, which is why the chips start on "all five".
Worked example from the same morning. Start with the five averages written on a single line: 4.85, 4.83, 4.64, 4.59, 4.54. Mark the names beside them. Cross out every other column. The remaining order is the home page. Put the crossed-out columns back in the tables so a second person can audit the crossing-out. That second person is the reader. The corrections log stays empty until an average on a later card disagrees with one of those five figures. A disagreement about taste does not qualify. A disagreement about a digit does.
The left panel of the home proof repeats the rule in one sentence so the scroll on the right cannot be read as a different method. Pinning that panel on a wide screen keeps the sentence in view while the tables move. On a narrow screen the panel sits above the tables instead, because a fixed side column would cover them. The method does not change with the width. Only the furniture does.
File sizes in rank order are 437.5 MB, 296.3 MB, 987.7 MB, 432.7 MB, and 318.9 MB. A size sort would put Grindstone first and WHAT THE GOLF? last. That is not this page. The unused key stays in the fact boxes so the choice can be checked without opening the store again. Age lines are unused too. The two 4+ cards happen to lead. A later 4+ card with a lower average would sit at the bottom. The rule is the average, written once, applied once.
The same five figures are the whole sort. No second key breaks a tie, because no tie occurred. If two averages land on the same tenth later, the supplied URL order is the tie break, and the corrections log will say that in one sentence. Until then the tie break is unused and written down so it is not invented on the day it is needed.