The short version
- A repeated correction is where the question changes from this piece of work to this arrangement.
- Check the standard before you judge the person. Twenty AI-written posts for this site missed our own content bar because nothing in the brief made anyone open it.
- How fast you can know is set by how often work ships. Weekly delivery gives around 50 correction opportunities a year against 12 on a monthly rhythm.
- Judge what shipped, not the update about it. We verify by measuring the built page rather than accepting a claim that it was done.
The reliable signal is a correction you have now given twice. A single miss tells you about one piece of work. When the same note comes back around, the question stops being about the work: either the standard was never written down anywhere the person could find it, or the role is wrong for them.
How do you know when it is not working?
By counting repeats rather than weighing severity. Nine standing editorial rules exist on one of our accounts because the same notes kept being given twice, and they were written at issue 20 of the newsletter rather than issue 1.
How quickly a repeat surfaces is a property of cadence rather than of the person. We deliver weekly on every account, which produces around 50 correction opportunities a year against the 12 a monthly rhythm allows. A quarterly cycle produces four, and a second look lands six months in.
The same evidence, arriving at three delivery cadences:
| Work ships | Reviews in a year | Time until the same note can return | Reviewed work by month three |
|---|---|---|---|
| Weekly | About 50 | Two weeks | About twelve pieces |
| Monthly | About 12 | Two months | Three pieces |
| Quarterly | 4 | Six months | One piece |
Is the problem the person or the standard?
Ask it in that order, because the answer is frequently the standard. A batch of twenty AI-written posts for this site came back structurally clean and missed our own content bar completely: no first-party data, no tables, no FAQs and no schema, at around 595 words against a 1,000 to 1,400 word requirement. The standard existed. Nothing in the brief made anyone open it.
The same failure reproduces cheaply with people. A bar held in a reviewer’s head is one somebody has to infer, and the inference will be wrong in ways nobody predicted. Write it into the brief, run one more cycle, and if the note returns with the standard visible on the page you have your answer.
What changes when nobody shares an office?
The ambient signal disappears. A person stuck on something is visible across a room and invisible across four time zones, so what an office did for free has to be scheduled explicitly. The check-in becomes the only place difficulty surfaces.
The evidence moves to output. We verify our own work by measuring the built page rather than accepting a claim that it was done, and the same discipline belongs here. An update describing progress and a shipped page are different objects.
What do you owe someone before you make the call?
The standard in writing, at the point the work starts rather than at review, and then a correction attached to something a person can act on. We turn call recordings into action items rather than notes, because the transcript is raw material and the action item is the deliverable.
Disappointment expressed in a review gives somebody nothing to do on Monday. Name what changes and the date it changes by, and the conversation becomes work they can complete or fail. Either result tells you something.
If the bar was never written down anywhere the person could find it, the gap is yours.
Why does the decision get delayed for months?
Because nobody’s name is on it. Whether a role is working sits between whoever hired the person and whoever manages them, and both are waiting for the next quarter to read differently.
The identical mechanism took a client blog offline for weeks. The technical fix was not difficult. It stayed down because it sat between two vendors and two internal functions with no single name against the restore. Temporarily becomes permanent at the speed of whoever keeps raising it.
The two habits that make this decision earlier are covered in how to handle the conversation itself and why we ship weekly.
Frequently asked questions
How do you know when to let an employee go?
The signal is a correction you have already given once, returning unchanged, with the standard written down and visible at the time. Repetition is the trigger rather than severity, because a repeat means the bar was never legible or the role is wrong.
How long should you wait before acting on a performance problem?
Long enough for the work to ship several times against a written standard, which is a property of cadence rather than a number of weeks. Weekly delivery gives around 50 correction opportunities a year against the 12 a monthly rhythm allows.
How do you tell whether the problem is the person or the process?
Check whether the standard was written down where the work starts. Twenty AI-written posts for this site missed our own content bar because nothing in the brief made anyone open it. If the bar lives in a reviewer's head, fix the brief first.
How do you spot a struggling team member on a remote team?
From shipped output, on a short cycle. Across four time zones nobody sees the stuck afternoon an office would have made obvious, so the check-in has to be scheduled and the evidence has to be the work itself rather than a progress report.
What should you do before deciding to let someone go?
Put the standard in writing where the work starts, convert the correction into a dated action item rather than a note about disappointment, and run one more delivery cycle. A person can complete or fail a specific change. Nobody can act on general dissatisfaction.
Why do managers delay a decision they have already made?
Because no single name is against it. The question sits between whoever hired the person and whoever manages them, and each waits for the next quarter to read differently. The same ownership gap kept a client blog offline for weeks.
Sources
- Chua Network delivery data across 8 client accounts (internal fact bank)
- Chua Network engagement records, anonymized (internal experience bank)
