The risk list
Put the evidence risk first, because it is the one this site can show rather than argue. Thirty-one wrapped addresses taken from public sheet pages returned nothing during the 2026-W39 window; the source addresses rebuilt out of the same rows all answered. Your order is not in that sample. Your file may still be shaped like it. A dead entry point does not cancel a purchase, it removes the column you would have quoted. The order continues regardless, which is the part that surprises people who assume the link is the order.
Second on the list is money that left before anything was ordered. The public timeline is blunt about how that happens: purchasing and redemption were suspended from 2026-06-12, and the notice that preceded it on 2026-06-01 already mentioned some cancelled orders. A paid row whose order never opened is not a shipping problem, a quality problem or a patience problem. It belongs in its own column, and it is the one category where the money question is live rather than theoretical.
Third is quality, which is the risk people prepare for and still lose. A flaw can be visible in a photo set, invisible in it, or invisible until a parcel is opened in another country. This site runs a QC scorecard for exactly that reason: not to grade a seller, but to make a photo set answer a fixed set of questions so two items can be compared and one decision can be recorded. A score with no date is an opinion with a number attached.
Fourth is the row that quietly stops meaning what it meant. A seller swaps a variant, a listing is replaced by a similar one, an identifier still answers while the page behind it has changed. Fifth is the long tail: a parcel that outlives the service carrying it, a question that never gets an answer, a deadline nobody wrote down. Two of those five are yours to prevent. The rest are yours to detect early. Detection is not passive. It means opening the file on a schedule and reading the stage rather than the mood.
Probability against impact
This site does not publish a probability for any of those five, because none has been measured. Inventing a percentage would be the easiest thing in this article and the least useful, since a made-up frequency gets quoted back as fact within a week. What can be ordered honestly is how likely it is that the damage would still be visible when you go looking, set against how much of it could be recovered afterwards. Two questions, one page, and not a single invented percentage anywhere in the answer or beside it.
The evidence risk sits in the awkward corner of that grid. It is close to certain if nothing is done, and complete when it lands, because what is lost is not the item but the ability to prove anything about it. The money risk is rarer and just as total, which is why a written record of the stage each row reached is worth more than a confident memory of it. Quality problems sit lower on impact and higher on frequency, and they are the ones a checklist can shrink.
Ranking by likelihood alone produces bad decisions, because the things people fear most are usually the things they can most easily check. Ranking by impact alone produces paralysis, since the worst case in any haul is a total loss and no plan survives that framing. The useful order is the one you can act on: what is still fixable today, what needs a message sent this week, and what is already outside anybody's hands. That order changes as the weeks pass, which is why the ranking belongs in the file with a date on it.
One discipline keeps the grid honest. Every risk on the list needs a signal that would tell you it has happened, and that signal has to be something you can read without asking anybody. A status field with a date is a signal. A feeling that something seems off is not. If a risk has no readable signal, it is not a risk entry, it is background noise, and it will sit in the list forever without ever being resolved.
Responses and stop-losses
The response to an evidence risk is duplication, not insurance. Rebuild the wrapped addresses into source addresses, keep the wrapped column untouched, and store the file somewhere that is not the same account you bought with. It costs nothing, takes one pass, and removes an entire class of argument from the haul. The legacy link resolver does the rebuild; a dated copy of the sheet does the rest. Duplication also settles the worst argument in a stalled haul, which is whose version of the row is the real one.
The response to a money risk is a written stage, not a hopeful message. For each paid row, record whether an order was placed, whether it was confirmed, and whether anything was dispatched. That single classification tells you which rows are waiting on somebody else and which are waiting on nothing at all. It also stops the most common error in a stalled haul, which is sending good money after a problem that was never yours to fix.
Quality gets a checklist and a cut-off. Inspect the photo set against the scorecard's fixed questions before a parcel is consolidated, decide from the answers rather than from the mood of the day, and write the decision next to the date. Then set the stop-loss while you are calm: the point at which you stop adding services to a row and treat it as a settled loss. A stop-loss written in advance is a decision. One written during a bad week is a reaction.
Questions that keep returning are a stop-loss of a kind as well. Most of what arrives with a stalled haul has already been asked and answered on the open questions page, grouped by intent rather than by keyword, with a one-line conclusion under each entry. Reading there is faster than asking a stranger, and it produces a record rather than a conversation. That matters at the point where an answer has to be defended. A sentence you can point at on a page is worth more than a screenshot of somebody agreeing with you in a thread.
The fallback plan
A fallback plan is a set of sentences written before they are needed, and it fits on one page. Which service the rows would move to, what gets rebuilt before the move, which rows would be abandoned rather than carried, and what each of those decisions costs in work rather than in money. The plan is not a prediction. It is a way of not making four decisions in one afternoon while a notice sits on the screen. Write it while nothing is wrong, and keep it where the sheet lives.
The move itself has one requirement and no shortcuts. A wrapped address issued by a stopped service has to become a source address before any other service can use it, because no other service was ever handed the wrapper. The platform ledger tracks whether each object accepts carried-over addresses and records the day that field was read, which is the single entry worth checking before a migration starts. Choose the destination on that field and the date beside it, rather than on the reputation of a name or the confidence of a stranger.
Some rows should be left behind, and naming them in advance is part of the plan. A row with no identifier, a row whose address nobody can rebuild, a row that only ever existed as a screenshot: none of those become more useful on a new service. Carrying them across adds noise that the next reader has to sort through, and the next reader may be you in a year. A shorter file that can be trusted beats a longer one that has to be re-explained every time it opens.
The last line of the plan is the exit condition. Write down what would have to be true for the haul to be closed: a parcel delivered, a refund recorded, an order confirmed dead with a date beside it. An exit condition turns an open wound into a task with an end. Without one, a stalled haul keeps its own hours, and the file stays open long after the order stopped moving. Closing a row is not the same as winning it, and only one of those two is available on demand.