Six Assumptions About Kakobuy Spreadsheets That Do Not Hold

The assumption is that a saved row points at a seller who is still trading. A working list of 195 Weidian entries carries that assumption 195 times over, and a single snapshot cannot confirm one of them. The row records what was true on the day it was copied: a title, a price band, a photo set, a category label, a brand tag. It does not record reachability, and it does not record whether the listing was ever reachable in the first place.

Why it fails: the sheet stores a description, not a connection. Opening the file refreshes nothing. The pool was extracted on 2026-09-29, so every price in it is a reading from that date, and a seller who stops trading afterwards leaves a row that keeps its title, its price and its position in the sort order. Rows do not decay visibly, which is precisely why the error survives review. A dead entry looks exactly like a live one until something is attempted against it.

What holds instead: treat every row as a dated claim rather than a stored fact. Keep an extraction date column and a last checked column, and let the distance between them rank rows for attention. The method page at /method/ applies the same discipline to the whole workbook: a row is only as current as its newest stamp, and a row with no stamp is the first to re-check rather than the last. A date column costs one field and converts an unanswerable question into a sortable one.

What it costs: an entry that stopped trading between extraction and consolidation consumes a warehouse intake cycle, a QC slot and a reserved place in the parcel. International freight is quoted at $45 per parcel by default, not per surviving item, so a dead row inside a consolidated box raises the effective freight carried by every delivered item. The loss is not the item price. The loss is the empty space that price occupies and the calendar time spent discovering it.

The assumption is that price ranks quality inside a category. Headwear in the pool spans $7.94 to $32.12 across 24 entries, and Accessories spans $5.17 to $451.10 across 24 entries. Both categories hold two dozen rows, and the Accessories band is roughly eighty-seven times wider at the top than at the bottom. A single quality ladder cannot produce a spread that large across the same shelf of objects.

Why it fails: within-category spread is produced by object class, material count and photograph count rather than by grade. The category bands overlap heavily. T-Shirts run $13.61 to $45.36 with a median of $22.23, while Hoodies and Sweaters start at $12.13, which places the cheapest hoodie in the pool below the median T-shirt. Pants and Shorts run $16.07 to $69.20, and Jackets run $19.06 to $156.27, so the top of one band sits inside the middle of another.

What holds instead: compare a row against its own category band and quote the band alongside the median. A headwear row at $18.40 is above the headwear median of $16.22 and still inside a band that stops at $32.12. That statement is checkable. A statement that $18.40 is good quality is not. The category band and the median are the two numbers that make a row legible, and neither of them claims anything about what will arrive.

What it costs: the assumption runs in both directions. Buying up to a tier that does not exist raises the item subtotal, and every fee row that is percentage-based follows it, including the agent commission at 5 percent and the payment surcharge at 3 percent. Skipping a row because it looks cheap removes a legitimate option. Both errors are silent, because neither produces a number that looks wrong on its own line.

The assumption is that list size measures list quality. The current pool holds 195 entries across 8 categories and 20 brands, and it carries 2806 listing images in total, a median of 14 per row and a maximum of 46. Of those 195 rows, 158 carry at least five images and 37 do not. The five-image line is the index threshold, so 81 percent of the pool enters the index layer and 19 percent stays outside it.

Why it fails: growth is cheap and coverage is not. Adding a row is a copy operation that takes seconds and costs nothing. Establishing that a row deserves trust takes a photo count, a category placement and a look at the price band. A list that grows from 150 rows to 300 rows without a matching change in coverage has doubled its overhead and left its usable fraction where it was, and nothing in a plain row count will show it.

What holds instead: measure the list by coverage. Report the verified share, the count below the gate, and the median image count, and treat the row count as context rather than as a score. The note at /field-notes/measuring-a-spreadsheet/ sets out how those three figures are read together; the short version is that a list is as strong as the fraction of its rows that survive its own gate. A count above the gate is a claim, and a count below it is a queue.

What it costs: an unverified row does not announce itself in a sort. It sits in the same column, in the same font, at the same row height as a verified one. When the list is passed to somebody else, the unverified fraction transfers with it, and the receiving sheet inherits 37 open questions that look like 37 answered ones. That transfer is where a coverage gap turns into somebody else making a decision against a number that was never established.

The assumption is that the price written next to a row is the price of the row. The pool spans $5.17 to $451.10 with a median of $31.76, and those are item prices. The worksheet that carries them also carries five further lines: domestic shipping at $1.40 per item, agent commission at 5 percent and adjustable, payment surcharge at 3 percent and adjustable, international freight at $45 and adjustable, and destination tax that applies only above a threshold.

Why it fails: the item price is one row out of six, and the other five move without it. A single item at the $31.76 median attracts $1.40 of domestic shipping, $1.59 of commission and $1.04 of surcharge under the default settings, which puts the parcel at $35.79 before freight. Add the $45 freight line and the same item lands at $80.79, with freight taking 56 percent of the landed total. The saved price describes 39 percent of what the parcel costs.

What holds instead: store the landed total and keep the item price as an input to it. Every shelf, every comparison and every budget decision in a worksheet should be made against the landed column, because that is the column the parcel will actually settle at. The questions collected at /field-notes/kakobuy-spreadsheet-questions/ keep returning to the same point from different angles: which line is being quoted, and is it the line that decides anything.

What it costs: a spreadsheet that stores item prices converts every budget into an estimate that drifts upward at the moment of payment. The drift is proportional, so it grows with the haul. A five-item parcel with a $200.78 subtotal carries $7.00 of domestic shipping, $10.04 of commission and $6.53 of surcharge, which is $23.57 before freight and before any destination tax. None of that appears in a column that stores item prices only.

The assumption is that a rich photo set is evidence about the item. The pool holds 2806 images across 195 rows, with a median of 14, a maximum of 46 and a minimum of 1. Five images is the index threshold, and 158 rows clear it. Fourteen images is not a small set. It is also entirely produced by the seller, before anything has been bought, shipped to a warehouse or handled by anybody other than the person who wants the sale to happen.

Why it fails: the two photo sets answer different questions. Listing photos describe what the seller chose to show, and their count measures sales effort rather than fidelity. A warehouse intake photo is taken of the object that arrived, at the address that received it, and it is the only image in the workflow that exists on the buyer side of the transaction. A row with 46 listing images and no intake photo has more evidence about the listing than a row with 6 images, and the same amount of evidence about the delivery.

What holds instead: use the photo count as triage and the intake photo as evidence. The five-image gate is a filter that removes rows nobody bothered to document, which is a real and useful signal, and it is not a promise. A row that clears the gate has earned a place in the index layer and nothing more. Treating the gate as a quality mark converts a sorting rule into a claim it was never designed to make.

What it costs: an item that arrives off specification after passing a 46-image review consumes a QC decision, a return or dispute path, and a slot in a parcel whose freight has already been quoted. In a consolidated parcel the freight is shared, so a single rejected item also raises the effective freight on the items that were accepted. The photo set did not cause the loss. Trusting it as proof delayed the moment when the loss could still be avoided.

The assumption is that a list copied from somebody else carries the same meaning in the new file. The pool used here is Weidian only, one source marketplace with one set of listing conventions, and it was read against a six-country destination table: US $800 at 0 percent, UK £135 at 20 percent VAT with an £8 fee, Germany €150 at 19 percent with a €6 fee, Poland €150 at 23 percent with no confirmed fee, Canada CAD 20 at 13 percent with a CAD 9.95 fee, and Australia AUD 1000 at 10 percent.

Why it fails: defaults travel silently. Commission sits at 5 percent, the surcharge at 3 percent, freight at $45, and the destination assumptions are whatever the original author happened to need. The checked-row counts behind the destination table differ by a factor of seven between the largest and the smallest non-zero reading: Germany carries 264 checked rows, the United States 196, Canada 138, Australia 39, the United Kingdom 35, and Poland none at all. A shared sheet does not display which of those columns it was built against.

What holds instead: re-derive the fee block before trusting any total, and record the destination the sheet was written for. The failure mode described in /field-notes/why-spreadsheets-rot/ is exactly this one at scale: a workbook that was correct for its original question keeps answering that question long after the question changed. Two sheets that produce the same landed total for the same row are comparable. Two sheets that produce the same item price are not.

What it costs: a shared list makes cross-checking look free and makes it expensive. Two people comparing subtotals conclude that a haul is affordable, then discover at the tax line that one sheet assumed Canada with a CAD 20 threshold and a CAD 9.95 handling fee on the first dollar, while the other assumed the US $800 line and zero tax. The disagreement appears at the end of the process, after the items are bought, which is the most expensive possible moment to discover it.