An organization advertising the same role again is telling you something. The mistake — and this is a mistake made here, in this project, and caught by someone else — is being confident about what.
The inference that felt obvious
A repeat posting looks like a failure. The role was filled, the problem was supposed to be solved, and here is the same advertisement again. The reading almost writes itself: the last hire did not make the work go away, so the work was never a headcount problem. It is a process nobody owns, and each new person inherits the same manual seam.
That reading is good. It is also, most of the time, wrong — and the way it was found to be wrong is the useful part.
What counting did to it
The inference was written down as though it were the meaning of a repeat. Then the repeats were separated by shape: the same seat advertised again after a gap, versus several seats open at the same time.
Most were several seats open at once. Almost none were the same seat again.
Several roles open simultaneously is not a failed hire. It is growth — the volume is winning faster than hiring can keep up with it, which is a real problem and an entirely different one. The rare shape, the same seat re-advertised after a gap, is the one where the failure reading is even available.
So the inference had been attached to the wrong thing. Worse, it had been attached to the common thing, which is exactly how a wrong idea gets a lot of use before anyone notices.
Then the count would not hold still
The obvious next move is to give you the number. It cannot be given, and the reason is the better half of this piece.
Counted one way the rare shape appeared twice. A normalizer that treated Senior Planner and Planner as one seat had merged two different jobs, so the count went to zero. Widening it to catch several different roles opening together took it to six. Then each of those six turned out to be the same role advertised in a different city — Cork, Galway, Limerick on one day — and it went back to zero.
Four counts, one corpus, and the corpus never changed once.
Every move was a different answer to a question nobody had written down: what makes two postings the same seat? Title, seniority, location, and the qualifier in the parentheses are four axes, and the count is a function of all four. Until that rule exists, any number here is a statement about the normalizer rather than about the organizations.
What it would have cost
There is a real organization in that list with seven coordination roles open at once, every posting opening with the words due to continued expansion.
Under the original reading, they would have received a message saying they had advertised this role before and that their last hire had not fixed the problem. Both false, and falsifiable by them in the time it takes to read one sentence. Not wrong about the work — the work is genuinely rule-shaped and genuinely growing — but wrong about their organization, told to them by a stranger, in an opening line.
That is not a message that gets a reply. It is a message that teaches the reader the sender does not know what they are talking about, and everything after it is discounted.
The rule that came out of it
The repeat is a reason to open a conversation. The cause is a question to ask, never a fact to tell them.
A re-advertised seat can be a process nobody owns — or a backfill after somebody resigned. Several at once can be growth. The person reading the message knows which; the person sending it does not, and pretending otherwise trades a reply for the appearance of insight.
None of the three causes is disqualifying, which is the part worth sitting with. A backfill is still a person about to absorb rule-shaped volume by hand. Growth still means the volume is outrunning the hiring. The work is the same in every case. Only the story about it differs, and the story was the part that was invented.
And the count that was supposed to settle which story applied turned out to depend on a definition nobody had written down. The honest position is not a smaller number. It is that the question underneath it is still open.
The transferable version
This is not really about job ads.
An organization with a signal it trusts is one step away from an organization with an inference it has stopped examining. The signal was correct throughout — the postings were real, dated, and accurately flagged. The failure was entirely in the sentence attached to them, and no amount of collecting better signals would have found it.
What found it was counting the thing that had been assumed. The split into shapes took an afternoon. The inference had been stated as fact in three separate places before anybody tried to count it at all.
And counting it was not the end. The count moved four times, because what makes two postings the same seat had never been written down either. The first assumption was found by counting; the second was found by the count refusing to hold still.
If a system is producing a claim about the outside world, the question worth asking is not whether the data is good. It is: what would this look like if the claim were false, and has anyone checked?
Totomoko re-architects how organizations work so that AI does the job — inside the systems they already run, on an architecture built for portability: model routing and a data layer the organization owns. Built that way, swapping a vendor is a configuration change rather than a migration.