The Dog May As Well Have Eaten It
Eighty pages that could not have come out wrong are not a report. They are the excuse, typeset.
By KW Norton. A short companion on length as a substitute for specification.
1. The offer
The standing response to an incident is a document. It arrives long. It has a timeline, a description of the environment, a section on lessons, an appendix of terminology, and a closing paragraph about continued commitment. It is written carefully by people who are not lying. And when you finish it you know exactly what you knew before you started, phrased better.
That is the schoolroom joke turned inside out. The child says the dog ate the homework because the homework does not exist. Here the homework exists, at length, and the absence is inside it. Nobody has to eat anything. Eighty pages arrive on time and the graded content is missing from the middle of them.
2. What page count cannot do
Length is not evidence. A document is informative in proportion to the number of ways it could have come out differently. A report that could only have been written one way — because every sentence in it describes rather than commits — carries no information about the world, however many sentences it has.
This is the same substitution the archive has been tracking all week, one register up. A metaphor substitutes for a mechanism. A diagnosis substitutes for a repair. A forecast percentage substitutes for a denominator. Volume substitutes for all of them at once, and does it more effectively, because volume looks like effort. Nobody accuses a long document of being empty. That is precisely what makes it the preferred container for emptiness.
3. The five lines
The test does not require reading eighty pages. It requires searching them for five things, each of which fits on one line:
One. The objective as written — the sentence the system was actually asked to satisfy, quoted, not paraphrased.
Two. The reward as implemented — the scoring function in the code, which is a different object from the objective and is where the divergence lives.
Three. The dated escalation record — who raised the concern, on what date, to whom, and what happened next.
Four. The named overruled objection — at least one internal disagreement, stated as it was stated, with the reason it was set aside.
Five. One pre-committed failure measurement — a number, fixed before the run, that would have declared the run a failure had it landed on the other side.
A four-page document containing those five is a report. An eighty-page document missing them is a press release with load-bearing footnotes. The difference is not tone, thoroughness, or good faith. It is whether the paper is attached to anything that could have said no.
4. Why the long version is worse than nothing
A missing report leaves a visible hole, and visible holes get asked about. A long report closes the hole with material. It gives the question an answer-shaped object to land on, and answer-shaped objects end inquiries. This is the mechanism that matters: the document does not merely fail to inform, it consumes the demand that would have produced information.
The same effect appears in the child’s version. The invented dog is not a failure of homework; it is a device for ending the conversation about homework. Length is the adult, credentialed, typeset form of the same device, and it works considerably better, because no one has to believe in the dog.
5. The reader’s instrument
What a reader can do with any incident document, in about four minutes: search it for a quoted objective, a scoring function, a date attached to a named person, a recorded disagreement, and a number fixed in advance. Count how many of the five are present. Publish the count. Not the critique — the count. A number between zero and five, reported publicly against a specific document, is itself a measurement that could come back the other way, which is more than most of the documents contain.
This piece is subject to the same rule. It names a trap, which is the cheap half. The step it carries is the count.
6. Status and falsifiers
Established. That informativeness scales with the number of ways a statement could have come out otherwise is a standard result, not a claim originating here.
Interpretive. That document length functions as a substitute for specification, and that a long document more effectively suppresses further inquiry than a missing one, are readings advanced here. No survey supports them; they are patterns claimed from public incident reports, and they are checkable.
Unsupported. Any implication that the writers of such documents intend the suppression. Nothing here establishes intent, and the mechanism does not require it — a length norm and a legal review are sufficient to produce the same output.
Falsifiers. (1) If a sample of long incident reports is found to contain the five artifacts at the same rate as short ones, length is not doing the work claimed here and the essay is wrong. (2) If published incidents whose reports omitted all five were followed by instrumented, externally checkable changes at the same rate as those that included them, the five-artifact test is not the discriminator and should be replaced by whatever is. (3) If a reader applies the count to this archive’s own pieces and finds the archive scoring no better than the documents it criticizes, the correct verdict is that this essay is another eighty pages, shorter.
Companion pieces: Pretty to Think So, The Elephant Is a Checklist.