What We Changed in Our Numbers, and Why

A pond liner calculator was quietly understating water volume by 30%

A pond dug with sloped or irregular sides holds less water than its outer footprint and maximum depth alone would suggest, and the calculator corrects for that with a shape factor — 0.7, say, meaning the basin averages seven tenths of its deepest measured point. The code was applying that correction twice: once to shrink the maximum depth down to an average, and again to the volume worked out from that already-shrunk depth, which compounds a 0.7 adjustment into roughly 0.49 and understates a typical pond's real volume by about 30%. It was caught while writing the page that defines each term in the formula for a reader — the field's own description names a single depth correction, and that didn't match what the calculation was actually doing twice over.

It's fixed to apply once now, and a regression test locks the corrected figure in at both the calculator's default inputs and a second case, so the same double-application can't quietly return in a future change. Because a pond's water volume is what pump turnover and chemical dosing both get sized from, a low reading here wasn't a cosmetic rounding issue — it would have propagated into an undersized recommendation everywhere downstream of it. The pond liner size calculator now reflects the corrected figure.

A casing calculator said a stick shorter than the piece being cut would still work

The door and window casing calculator sorts every cut piece a job needs into whichever standard board length is being bought, so a shopping list comes out in sticks rather than in loose inches of trim. The function counting how many pieces fit on one stick rounded that count down to a minimum of one, even where the piece being cut was longer than the stick itself — at the calculator's own default inputs, a door leg needing about 84.44 in was being packed onto 84 in (7 ft) stock, a stick 0.44 in too short to physically hold it, and the calculator reported that as one piece successfully cut from it anyway.

It's fixed at the source: the calculator now steps up to the shortest stock length that actually accommodates the longest piece required, states which length it chose, and reports the margin by which that length clears the requirement rather than silently assuming any stock length works. Regression tests were added at both edges — a piece just barely too long for a given stock length, and one that just barely fits it — so this specific class of off-by-a-fraction error has a test in the way of it returning unnoticed. The door & window casing calculator carries the corrected packing logic now.

Titles and descriptions are now measured, not silently trimmed

Page titles and descriptions used to have their length limits enforced by silent truncation — a title running too long simply had part of it, including the site's own name, dropped without anything flagging that it had happened. Replacing that with an explicit check that stops the build instead of quietly cutting text caught fourteen individual page titles over their intended length, a homepage title at 62 characters, and a site-wide description at 175 characters, all of which had been live. None of the underlying pages were wrong — only the metadata describing them in a search result or a browser tab, which is exactly the part most likely to go unnoticed without a check that actually stops on it.

A page's description is now written for the reader, not repurposed from planning notes

Every content page on this site carries a short internal note describing what that page needs to cover — a planning aid for whoever writes it, never meant to be read by a visitor. That internal note was being reused directly as the page's actual meta description in search results, and separately published in full inside the site's machine-readable llms.txt file, on any page where a real description hadn't been authored yet. The description a visitor or a search engine actually sees is now its own field, written for that purpose specifically and kept separate from the planning note, so the two can no longer end up published as the same text.

Topdressed compost now accounts for what a surface application actually loses

The compost amendment calculator originally treated a tilled-in application and a topdressed, surface-spread one as needing the same volume to raise soil organic matter by the same amount. A topdressed layer sitting on the surface oxidizes and is only partially worked down into the soil by the time the organic-matter target it's meant to hit gets measured, so it takes more volume than a tilled-in application to land the same amount where it counts. The calculator now adds roughly 15% more volume for a topdressed application than a tilled-in one closing the identical gap, which the compost amendment calculator applies automatically based on which application method is selected.

Why this gets published rather than fixed quietly

Why two online calculators give you different answers argues that a calculator publishing its formula and naming its sources is checkable in a way a plain output number isn't. A record of what's turned out to be wrong here, and exactly how it was corrected, is the same argument applied to this site's own history instead of to a single page's assumptions. It isn't a closed list — it gets added to whenever a number here needs correcting again, rather than quietly rewritten as though it had always been right.