Reading our snapshot table: what changed between quarters

The snapshot series now holds twenty-six rows across four quarters, and the counts are the most useful thing in it. Eighteen rows are recorded as scope changes and eight as flat. Scope changes outnumber flat rows in every single quarter, which is a pattern rather than an accident, and it says something specific about what moves in this category: the way a line item is charged changes more often than the figure attached to it.

The table is built to record that distinction. A row carries a previous value, a current value, and a direction, and the direction that matters most is the one labelled as a scope change. A scope change means the basis moved: what the charge is calculated on, what unit it is charged per, or which side of a classification boundary a parcel falls on. A rate that was republished at a different number is not a scope change and is deliberately not logged. The consequence of that choice is a table that under-reports movement in the figures and over-reports structural movement, which is worth knowing before drawing conclusions from it. A quarter in which every carrier republished a rate card at higher numbers would read as quiet here, because nothing about the basis moved.

Three rows currently read Not verified rather than carrying a figure forward, and they are worth as much attention as the eighteen that moved. Two of them are fee bases whose structure appeared to vary by category, and one is a transit basis for which we do not have a defensible sample. Carrying an old figure in those rows would have made the table tidier and the table would have been wrong.

One row was retired outright in the first quarter of the year and stays in the series as retired rather than being deleted. Keeping retired rows visible is the point of a versioned table. A reader who wonders why a category they remember is no longer tracked can find the entry and the reason rather than concluding that it never existed. The same rule applies to any row we cannot re-derive: it gets marked rather than deleted, and the marking carries the reason. Deletion would make the series look cleaner and would destroy the only evidence that a category, a unit or a basis was ever recorded at all, which is precisely the information a reader needs when a total stops matching their expectation.

The moves cluster into three families, and the families are more informative than the individual rows. The first is basis changes, where the thing being charged stayed the same and the rule for calculating it changed. The clearest example is the service line, which moved from a single tier to tiers by order size, so the effective share falls as an order grows. Nothing was republished to announce that; it shows up as a different number on a large order than on a small one.

The second family is unit changes, where the same charge is now counted per something else. The domestic leg moved from per order to per seller shipment, which changes the arithmetic on any haul assembled from several sellers, and the restricted goods list moved from per carrier to per service, which changes the answer when two services under one brand differ. The divisor belongs here too: recording one value per carrier was too coarse, and the correct unit turned out to be per service, with published values of 5000, 6000 and 8000 observed.

The third family is classification changes, where a figure did not move but which parcels it applies to did. A category boundary for coated outerwear was redrawn without the underlying threshold changing, and the point at which a tax identifier is requested moved to before departure on some lines. Both are invisible in a table of rates and both change what a buyer has to do.

What held is as informative as what moved. The storage free window, the payment channel structure and the published destination thresholds were all reviewed with no scope change across the period. A quiet row is evidence that the last reading is still current, which is the only defence a reader has against a page that looks maintained but has quietly stopped being checked. Quiet rows also carry practical value: they are the handful of things you can plan around without re-reading anything, which is why we keep them in the table instead of removing them as uninteresting. When one of them finally moves, it moves into the first family and gets the attention it deserves, and until then it is doing its job by staying still.

The first reading is that the market moved: a platform genuinely restructured what it charges on, and the row records a real change that a buyer will feel in a total. The service tiering and the per-seller domestic leg are both in this category, and both are visible in a bill without any measurement effort. When a row of this kind appears, the right response is to re-check your own estimate rather than to wait for confirmation.

The second reading is that we changed how we measure, and the market did not move at all. The divisor row is the honest example: carriers were not charging differently, we were recording at the wrong level of granularity. Rows of this kind usually leave a trace, which is a new finer unit, a split into two rows, or a row moved to Not verified because the old recording cannot be re-derived.

The third reading is that the observation was a seasonal artefact. A structure seen during a congested quarter can look like a permanent change and then revert, and we have no clean way to separate that from a real change inside one quarter of data. Where the two cannot be told apart, the row stays as it is and the next quarter decides. That is a slow method and it is the reason the table is versioned rather than edited in place.

Telling the three apart is quicker than it sounds, and it starts with one question: does the change show up in a total. If it does, treat it as a market change and re-run your own estimate. If it only shows up in how a row is recorded, treat it as a method change and go back to the rows that depend on that recording. And if it appeared in the quarter after a congestion event, hold it open for one more quarter before acting on it, because a structure observed under load has a decent chance of being a structure produced by load.

We expect the pattern rather than a specific number, which is the only kind of expectation this series can support. On the evidence of four quarters, basis and unit changes should continue to outnumber figure changes in our table. The observation that would falsify that is a quarter in which no row changes scope while rate rows move, and if that quarter arrives we will record it rather than explain it away.

We expect granularity to keep increasing in one direction: per carrier becoming per service, per order becoming per seller shipment, single tier becoming tiered. The trigger that would count as a reversal is a quarter in which a charge is consolidated back to a coarser unit. That has happened once in the series, when two domestic lines charged on identical bases were merged into one row, so the direction is a tendency and not a law.

We expect the Not verified count to grow before it shrinks. Sea and rail consolidation are the areas where our sample is thinnest, and the honest alternative to a thin sample is an empty cell rather than a borrowed figure. What would change that is a defensible community sample with dates attached, at which point the row becomes a range with a stated basis. The test we would apply is the one we apply to any range on this site: reports we can date, from more than one route, spanning more than one season, and described in enough detail that a second reader could classify them the same way we did. Until those conditions hold, an empty cell is the more useful entry, because an empty cell tells a reader to go and check while a borrowed figure tells them not to bother.

None of this forecasts your costs. What the series forecasts is where to look when a total changes unexpectedly: at the basis of the charge first, because that is where the movement has been, and at the figure second. A reader who checks in that order will find the reason faster than one who assumes every change is an increase.