RungDesk
Rung 3 · Edge cases⑱ 版本差异型KB-2026-11· 2026-09-301695 words

Last year versus this year: what actually changed

The phrase CNFans spreadsheet 2025 2026 is usually typed by someone still holding links from last year and wondering what applies. Almost everything that changed was a change in how figures get read, not in what the table is.

How a 2025 table was read, and what that assumed

A year ago the working assumption was that a link table is one file with one set of numbers in it. Currency was judged by whether the page looked recently touched, transit was a single figure per destination, the volumetric divisor was one number applied to everything, and a size label was treated as the size. Each of those assumptions was reasonable at the time and each one has since been shown to be a simplification that costs money.

The costs were not dramatic individually. A single transit figure made every late parcel look late, because there was no range to sit inside. A single divisor meant two readers could weigh the same box differently and both be right, since consolidated air lines do not all use the same one. An undated duty percentage looked precise and could be two years stale without anyone noticing.

What changed is less about the platform and more about the discipline of reading a number at all. The comparison below is the one the desk applies to its own entries, and it is the same test a reader can apply to any table they are still using, whichever site published it.

What the desk changed on its own side this year

The desk opened eleven review batches between January and September 2026, and the changes they recorded were mostly about how figures are carried rather than what the figures are. Every destination band was re-dated, every window was re-cut with a stated sample, and the volumetric divisor set was expanded from one value to four with the mode each belongs to recorded beside it.

The most useful of those changes was the least glamorous one. A whole-desk pass in late September re-read every entry against its own stated basis, removed eleven entries that had drifted into repeating another entry without adding a figure, and re-filed nine entries onto a different difficulty rung after readers reported opening them too early or too late. Nothing in that pass added a new claim; it removed claims that could not be checked.

That is the honest shape of a year-over-year change on a reference site. The headline is not a dramatic revision, it is a slow removal of numbers that were never sourced, and a corresponding increase in how often a figure arrives with a date attached to it.

The divisor question, a year apart

The volumetric divisor is the clearest example of a simplification that had to go. Last year the working figure was one divisor treated as the rule, which made every other result look like a mistake. The desk now records four divisors, 5000, 6000, 7000 and 8000, together with the mode each one belongs to, because they belong to different services rather than to different opinions.

The second change in the same batch was the rounding step. A half-kilo rounding switch was added as an input rather than baked into the arithmetic, after it turned out to change the chargeable figure on 31 of 88 sampled parcels. That single finding is why two quotes for an identical box can differ by a step and neither one is an error, and it is the reason the desk now treats a small discrepancy as a question about method before treating it as a dispute.

For a reader holding last year notes, the practical consequence is simple. A chargeable weight copied from a year-old note cannot be trusted unless the divisor and the rounding step are written next to it, because the same dimensions produce different billable weights under a different pair of settings.

Windows re-cut with larger samples

The first crossing windows in January 2026 came from an initial set of 240 community-reported arrivals. By May the desk had enough reports to re-cut every window with an explicit sample count beside it, which changed the shape of the answer more than the numbers themselves. The United States sea window widened from 22 to 34 days into 25 to 42 days on a larger sample, and a choke-point column was added so that a wide window could be explained rather than merely displayed.

Sample counts now sit inside the windows rather than in a footnote. The United States air-economy lane carries 214 reported arrivals and the New Zealand lane carries 29, which is why one of those windows can be used to rule a parcel late and the other cannot. Lanes under 45 reports are labelled thin on the gauge, and a thin lane widens the expected window instead of justifying an escalation.

A reader comparing last year to this year should therefore compare the sample counts, not just the day ranges. A window that grew is not a service that got slower; it is usually a window that was too narrow to begin with, drawn from too few reports.

Dated thresholds replacing undated figures

Destination rates changed shape in August. Every point rate was replaced by a band with a stated reason the band is wide for that category, because a single percentage for apparel implied a precision that customs practice does not support: rates move with fibre and construction, so a shirt and a coat are not the same line even when both are apparel. The state or provincial override stayed as a required input wherever the rate is not national.

In September all eight destination bands were re-dated after being checked one by one against a public notice, and the desk treats a figure without a date as unread. That is a small rule with a large effect: a reader can immediately see whether the number in front of them is three weeks old or three years old, and can decide how much weight to give it.

The practical test for any table is the same one. Pick a number that matters to your order, and look for the date it was checked. If there is no date, the number is not wrong, it is simply unusable, and it should be re-derived rather than relied on.

Holding links from last year

In August 2026, community reports described the platform as no longer taking new orders; this desk does not repeat that conclusion, and nothing in this note depends on it. What follows is the check you can run yourself on any link you are still holding, whoever published it and whenever you saved it.

Open the link and look at what comes back rather than at whether it loaded. A page can return a success response and still show nothing: an empty product shell, a placeholder image, a title with no description. Then compare the item to what you originally saved, and write down the date of the comparison beside the original link text. If the item is gone or changed beyond recognition and money has already moved, the question is no longer about the link but about the order.

For anything already paid, the rule is the same in every case: contact the party that holds the order, ask one specific question in writing, and keep the reply. Ask whether the item is still the item that was ordered, and if not, what the options are. Keep the order reference, the date the order was placed, the original link text, a dated capture of what the link shows now, and the reply itself. If a balance or an unsettled order is involved, ask the same question of the same party rather than of a community thread, and keep the answer in writing as well.

What has not changed, and why that matters

The arithmetic did not change. Chargeable weight is still the greater of gross and volumetric weight, duty is still assessed on a landed value rather than on a goods price alone, and a return still costs postage in one direction or two. A year of revisions moved the inputs and the confidence attached to them, not the shape of the calculation.

The evidence rule did not change either. A photograph of a fault, a measurement of an item, and a warehouse receipt are still the three things that make a request answerable in one pass, and no revision to any table changes that. Sizes are still not self-describing, which is why the measurement runs exist: 46 footwear rows measured as insole length and widest point, then 38 garment rows measured as chest width and body length.

What that means for someone holding notes from last year is unglamorous and useful. Keep the notes, but re-date them. Re-check the three or four figures your order actually depends on, write the check date beside each, and treat everything else as background rather than as an input.

The same seven items, read a year apart
#ItemHow it was read in 2025How the desk reads it in 2026What it changes for a reader
01Transit windowOne figure per destinationA range per destination and mode, with report counts beside itA parcel inside the range is not late, so fewer escalations
02Volumetric divisorA single divisor treated as the ruleFour divisors recorded, with the mode each belongs toTwo readers can weigh the same box differently and both be right
03Rounding stepNot mentioned at allA half-kilo step kept as a switchable inputQuotes a step apart stop looking like errors
04Destination rateOne percentage per categoryA band, with the reason the band is wide stated beside itThe estimate admits what it cannot know
05Currency of a figureUndatedEvery band carries the date it was checkedAn undated number is treated as unread rather than as wrong
06Size labelTaken as the sizeRecorded separately from the measured figureA label argument becomes a measurement argument
07A link that no longer worksTreated as a table-wide problemTreated as one link event, checked on its own firstThe cost is bounded instead of open-ended
∑Every row is a change in how the desk reads figures, not a statement about any platform policy.

A figure without a check date is treated as unread; re-derive it before relying on it.

What this note comes down to

  1. 01The year-over-year change is mostly in how figures are carried: ranges instead of points, samples instead of assertions, dates instead of memory.
  2. 02Four volumetric divisors replaced one, and a half-kilo rounding step moved the chargeable figure on 31 of 88 sampled parcels.
  3. 03Every transit window was re-cut with a stated number of reported arrivals, so a wider window usually means a better sample rather than a slower service.
  4. 04Point duty rates became bands because a single percentage implied precision customs practice does not support.
  5. 05A figure without a check date is unread rather than wrong: re-derive the three or four numbers your order depends on.
  6. 06Community reports about platform status are transcribed and not adopted; the check you can run yourself is the part that matters.

The data point behind this note

This comparison uses the desk own batch record: eleven review batches ran between January and September 2026, and the closing whole-desk pass re-dated all eight destination bands to September 2026, removed eleven entries that repeated another entry without adding a figure, and re-filed nine entries onto a different rung.

KB-2026-11· 2026-09-30 Batch ledger · Challenge a figure

Notes in the same cluster

Adjacent reading
#NoteWhat it arguesWords
01A refund that was actually worth takingA refund is worth taking when the money coming back beats the money the return itself removes. Three completed files show where that line actually fell, and how close two of them came to falling the other way.1625
02What a rejection letter gets you, and what it does notA rejection notice answers one question well and leaves four others untouched. This is a composite exchange, followed from the first question through to the point where the answer stopped changing.1591
03Five ways a parcel gets held, and how each one endsA parcel that stops at a border is usually stopped by paperwork rather than by suspicion. Five holding events, scored on how likely each is and how much of the order it can consume, with the first move for every one.1856

Related entries and gauges