A review from someone with no reason to give it
I built a careful bundle for a role I wanted. The first person to read it had known me for a single phone call, and he sent it back. He was right, and every gate I owned had passed it.
良薬は口に苦し
Good medicine is bitter to the mouth.
The call was with a recruiter I had never spoken to before. He was placing a product role at a school network, and he knew the hiring manager personally: classmates once, colleagues after. The relationship was the channel. He told me what the manager screens for, which was people who actually ship AI, and what he is allergic to, which was ceremony and buzzwords.
So I did what I do. I built him a bundle. A one-page brief on why the role fit, a tailored resume in a designed layout, a short note he could pass along. Every piece went through the gates I keep for anything customer-facing: no em-dashes, no cliches from the banned list, every number traced to a source. It was, by my own instruments, clean.
He read it and came back with the review that mattered. The resume read as AI-generated. The manager would see it that way inside a few seconds, and he was not going to forward it. Not a style note. A refusal to spend his relationship on it.
My first reflex was the one I have written about before: the gates passed, so the problem must be his taste. That reflex lasted about as long as it took to look at the page with his eyes instead of mine. The layout was polished in exactly the way generated things are polished. The phrasing was tidy in exactly the way generated things are tidy. Nothing on it was false. All of it was suspect, because correct and trustworthy are different properties and I had only built gates for the first one.
A gate can only catch the failures its author already imagined. The failure you cannot imagine arrives in the voice of someone who owes you nothing.
Bar raisers are usually appointed. A company picks someone from outside the hiring team, gives them a veto, and asks them to hold the line on quality with no stake in filling the seat. The design works because the outsider does not share the team's momentum.
This recruiter had no appointment. He had something better: a reputation he would spend by forwarding me, and a precise model of one reader. He was not grading the resume against a rubric. He was simulating a specific person opening it. My gates checked the artifact against rules I wrote. He checked it against the reader it was actually for. That is the whole difference, and it is not one my own review could close from the inside, because my rules and my blind spots came from the same place.
His note sent me back through the whole bundle with his reader in mind, and that second pass found something worse than a tone problem. A metric had drifted into the draft from a scratch file during tailoring, a clean-looking percentage that no vetted master could back up. No gate caught it. A blunt question did: did this come from anywhere? It had not. It came out.
The fix was not to argue. It was to ship the plainest possible version and then make the lesson mechanical so it could not depend on meeting the right stranger again.
The plain resume went out. The interview got booked. Whether it lands is not known at the time of writing, and this note does not pretend otherwise.
In an earlier note I wrote that a strand cannot audit itself, that it takes a second strand with no stake to give the first an edge to stand off of. I built that second strand as a reviewer agent and felt good about it. This was the version I could not build. The reviewer agent shares my rubric. The recruiter shared nothing with me but one phone call and an interest in not being embarrassed.
So the practice I am keeping is small. When a stranger gives you a review for free, the default is to assume it is the bar raiser you did not appoint. Treat the objection as signal before you treat it as taste. Then turn the part that generalizes into a gate, so the next draft does not need the stranger to be there.
Anyone who reads your work for someone else is a bar raiser. They are holding a reader you cannot simulate, and they are telling you what that reader will see.
灯台下暗し
It is darkest at the foot of the lighthouse.