Part III — Keeping the book
Chapter 14
Correct the book
On July 4, 2026, four entries in my team's book were marked confirmed. The confirming came from our own side of the work. We were saying the value had landed, about value our own work had created.
On July 12 the system caught it, and all four went back to waiting.
It's the clearest picture I have of why a book of value needs correcting. Whether the value had landed wasn't the problem. Who said so was. A confirmation from the wrong hand is a claim with a nicer label on it.
What this is
Correcting the book means finding entries that the rules or the evidence say are wrong, and fixing them where everyone can see the fix. You move a confirmed entry back to waiting. You move an entry into the right lane: the outside lane, for value that reached people outside your own side, or the inside lane, for value your own side received. You repair the link to the evidence, or the date. You separate two things one column was holding. And you write down why.
Our vocabulary's definition of the Value Ledger ends on the sentence that makes this necessary: "A filled-in number that does not correspond to value a real person received is worse than an empty one."
If you lead by KPIs today. The number you'd recognize here is the one that's never empty, the count that fills in every month whether or not anyone received anything. What this chapter gives you to read in its place is a book that shows where it was wrong: each correction, with what changed, why, and on whose ruling. Chapter 13 is how you take that count across to a question about value.
These are corrections from our own book:
- July 12, 2026. The four entries above went back to waiting. That correction became a rule I ratified on August 24: I never confirm value somebody outside the team received. Only they can.
- August 10. One column held the name of whoever drafted the entry. Our rulebook said the confirmer's name went in that same column, which would have written over the drafter's name every time someone confirmed. The confirmer got a place of its own.
- September 4. Ninety-one links to evidence opened onto nothing. They were repaired.
- September 5. The same drafting column was filled on a batch of 64 entries, and reading it as the person who did the work would have been a guess. We said so in our rulebook, and gave "who did the work" a place of its own.
- September 25. Five older entries had put people I lead in the inside lane. Under a ruling earlier that month, that everyone except me and our AI agents counts as outside, all five moved to the outside lane.
In my team, Pax, the AI agent who owns this practice, brings a correction forward, and the rulings about whose word counts are mine. In your organization, decide before you need it who brings a correction and who rules on it, and make sure neither of them is the person whose work the entry is about.
How to tell where you are
- An entry is confirmed by the same hand that did the work.
- A receiver is written as a sentence describing who benefited, not as a person. On September 5, 2026, 118 of the 163 receiver names in our book were written that way.
- A link to the evidence opens onto nothing.
- A person sits in the wrong lane.
- One column is doing two jobs, and people read it whichever way suits the moment.
- An entry has no real date, or has the date somebody tidied the records.
- A placeholder that says "something happened here, details to follow" has sat untouched for months.
- A count has been written up as if it were a moment.
- A rule changed, and older entries were written under the old one.
What to do
1. Read the book against the rules, one entry at a time. Not by feel, and not only the entries that look odd. For each confirmed entry, ask:
- Whose hand confirmed it, and is that a hand your rules allow?
- Is that hand different from whoever did the work?
- Does the evidence open, and does it show the moment?
- Is the receiver a person, or a real record of the team or the thing that received it?
- Is it in the right lane?
- Is its date the day the value happened?
2. Make the smallest true correction. Move it back to waiting. Move it to the right lane. Repair the link or the date. Separate the two jobs. Or leave it alone, if it passes.
3. Write down what changed, when, why, and on whose ruling. Beside the entry, and in a record anyone can find later. A corrected book that can't tell you what was corrected reads exactly like one that was never wrong.
4. Let a confirmation given under the rules of its day stand. When my team ruled in September that every person I lead counts as outside, two older entries had already been confirmed by me, on August 9, under the rules of that day. I was asked whether to undo them. I picked "Keep them confirmed." A new rule reaches forward. It doesn't reach back and take away a confirmation that was right when it was given.
5. When a correction shows a rule that nothing enforces, add the check. We found that of 39 entries that had passed our own audit, 37 named the AI agent that did the work as the one that received the value. After that, our system started refusing, at the moment an entry is written, any entry that names the same person as both the receiver and the doer. I'll be plain about its limit: it runs only when the entry is created, and only when both names arrive with it. A partial check is still worth having, as long as nobody mistakes it for a closed door.
6. Correct only against a rule or the evidence. An entry you'd rather not have, that passes the rules, stays in the book. So does confirmed value that no CARE record (Chapter 2) intended.
How you know it worked
Every confirmed entry rests again on a word your rules allow. For an entry about someone outside your team, that's their own words, or something they did under their own name.
The real test goes one step further. Put a corrected entry in front of the person it names. They should recognize the moment, and the word that confirmed it should be theirs, not yours.
Where it goes wrong
Fixing quietly. The entry changes and nobody records why. Six months later nobody can tell the corrected rows from the original ones, and the book's honesty depends on everyone's memory.
Correcting toward a better-looking book. If every correction makes the numbers go up, look at who is doing the correcting.
The opposite mistake: a rule so strict nothing gets confirmed. Back in June, my team allowed only a person above the AI agents to confirm anything. When we measured it, the book held 53 entries, not one of them confirmed, and nothing new had been written in 13 days. Our rulebook says that rule "protected the integrity of the number by ensuring there was no number." We replaced it in August with a simpler test: the receiver's own words, cited to a real moment, count as their confirmation, whoever finds them.
Reaching backward with a new rule. Undoing every older confirmation because a rule changed punishes people for following the rules they were given, and teaches everyone that a confirmation in your book is provisional forever.
The doer correcting their own entry. Moving a row up, or into a better lane, about your own work is the same mistake as confirming it.
Fixing the instance and not the rule. Four entries moved back to waiting would have been a tidy-up. The rule that came from them is what stopped it happening again.
Findings left waiting. A correction pass will turn up entries it can't settle by itself: a receiver written in the wrong lane, a person left in the lane a ruling moved them out of, an entry about someone outside the team whose confirming hand is yours and not theirs. Flag each one in the open, and bring it to whoever rules on it, the person you named before you needed one.
Read further
- The Value Era, Chapter 14, "Only the person who received it", on why the confirmation has to come from the receiver's side. Its section on leaving people out makes a point every correction pass should keep in mind: whoever keeps the book still decides whose moments get written into it.
- The Value Era, Chapter 17, "Who keeps the definition", on keeping words honest over time, which is the same job as keeping a book honest.
- In this book: Chapter 4, on the two hands, the one who received and the one who did the work; Chapter 8, on whose word counts; and Chapter 9, on the two lanes.