Week 12 quietly repeats week 3

Say a plumber starts a fortnightly piece in March. Week 3 is a hot water changeover in a 1970s house, with the cost, the time on site and why the old unit went where it did.

Week 12 is a hot water changeover in a 1970s house.

Nobody copied anything. He did that job 40 times in between. It’s the work.

Repeating yourself is what a specialist looks like

Say a tradie does about 6 kinds of job over and over. That is the pool a fortnightly piece is drawn from, and it is the reason the writing is worth reading in the first place. Depth comes from repetition.

So the second piece on the same kind of job isn’t a mistake. The mistake is the second piece arriving with nothing in it that the first one didn’t already have. Same fault, same explanation, same order, a different day.

Two pages then split one job between them. Neither is wrong. Neither is complete. And 18 months later nobody involved remembers enough about either job to merge them.

Three versions of “we check it before it goes up”

Ask whoever writes for you what gets checked when a new piece is published. There are only 3 answers.

• Nothing. The piece gets read for typos and pushed.
• The new piece against the last one or two. Better, and it feels thorough, because the last 2 pieces are the ones everybody remembers.
• The new piece against every page on the site, old ones included.

The middle answer is the one that produces week 12 repeating week 3. At a fortnightly cadence, remembering the last month is remembering 2 pieces. Week 3 dropped out of the window in May and nothing has compared it to anything since.

And you won’t catch it either, unless you re-read your own site.

The arithmetic nobody does by eye

Similarity belongs to a pair of pages, not to a page. So the real question is how many pairs a site has, and that number does not grow the way people expect.

Pages on the site	Pairs to compare
12 (a build, before any writing)	66
20 (about 4 months in)	190
26	325
38 (a build plus a year of fortnightly)	703
64 (two years)	2,016

The rule underneath it is that every page gets compared to every other page, so a site of 38 pages has 38 times 37, halved. Pages go up in a straight line. Pairs go up like the area of a square.

The line that matters is the last one added. Publishing page 38 onto a 37-page site does not create 1 comparison. It creates 37, and 36 of them involve pages written before anyone had a reason to think about them.

That is the whole argument for scoring the site rather than the piece. It isn’t thoroughness for its own sake. It’s that 703 comparisons is a second of arithmetic and 0 comparisons is a guess, and there is nothing sensible in between.

Where the two pages actually meet

It is rarely copy and paste. It’s usually one of these 3.

The same job type comes round again. The pool is 6 jobs and the year is 26 pieces, so by definition some job types get written up 4 or 5 times.

Your best explanation gets used twice. You have one way of explaining why a fault happens, worked out over years of saying it in driveways. It is genuinely the best version. It goes into piece 4 and then, phrased slightly differently, into piece 19.

The standing questions get asked again. Any honest interview process comes back to the same prompts when a fortnight was quiet. Same prompt, same year, similar answer.

None of that is dishonest and none of it is fixed by trying harder. It is fixed by comparing. How the comparison is actually computed, and why the fail line sits at 0.45, is a separate piece of arithmetic and worth reading if you want to audit the number rather than accept it.

What happens when a pair comes back

The check runs before the piece is published, across the whole site plus the draft. If any pair sits at 0.45 or above, both pages in the pair come back, named, with the score printed next to them.

Then a person decides, and the fix is almost never a rewrite:

1. Pick a different job. The recorded call usually has 3 jobs on it. Use the one that isn’t fighting with anything already published.
2. Narrow the second piece to the difference. If both jobs were the same changeover, the second piece is about the part that wasn’t the same: the reason the first fix didn’t hold, the different price, the thing that only happens in that pocket of housing.
3. Don’t write it. Sometimes the honest answer is that the ground is covered. The fortnight’s material becomes something else, and the piece that would have duplicated never exists.

The cheap moment is before publishing. Nothing we publish lands on top of a page you wrote, and anything that needs taking down afterwards is a person doing it by hand. There is no undo button and we’re not going to pretend there is.

The version you can run for nothing

If nobody is checking your site and you’re not paying anyone to, keep an index. One row per piece, 5 columns:

• Date it went up
• Job type, in your own words
• Suburb
• The number in it — a price, or a time on site
• One sentence: what’s in this piece that isn’t in any other

Sort it by job type once a quarter. Any job type with 3 rows against it is exactly where your next repeat is going to land. Before you write the next piece, read that one column, not the whole site.

For a site under about 20 pages, a 5-column spreadsheet does most of the work. Past that, the pairs get away from you, which is the point at which the arithmetic above stops being interesting and starts being the reason you’d pay someone.

What we do and don’t run

A person runs this. The check gives a pass or fail on each pair, and somebody reads it, decides what to do about a failing pair, and publishes. Nothing here happens unattended.

We also cap how much goes up at once. Left uncapped, the output becomes churn, and churn on a site you rely on looks a lot like something breaking.

The site-wide check matters most when new pages keep arriving: a page a fortnight on Hold Your Patch at $1,997 a month, or a page a month on Stay Found at $1,297.