The corpus fabricates in one column and tells the truth in another, and the difference is not the author's honesty. It is whether the column can be filled by somebody who does not know the answer.
Potential Drawbacks can. How might this fail? is answerable from imagination and a bit of self-knowledge, and the nine documents carrying that field produce the most candid material in the folder.
Measurable Outcome cannot. What happened when people did this? has one honest answer when nobody has done it, and nothing yet is not a permitted value in a grid whose purpose is to demonstrate a system's worth. So the cell gets a number, and the number is invented, and the author does not experience himself as inventing anything.
Now put that beside this vault's own template.
Fourteen mandatory headers on every page. A required Cross-Domain Handshakes section. A required The Live Edge. And a rule, in force for months, that every page carry at least two handshakes.
⚑ That rule produced exactly what the corpus's outcome column produces. CLAUDE.md says so in its own words: pages are checked for section count and never for whether the second handshake was real, so the floor won every time, and across one build it mandated four hundred and forty-two connections that had to exist whether or not they did.
Two independent systems. One diagnosis. Neither noticed the other.
The vault fixed its version and did not generalise the fix.
The 2026-08-10 amendment did the right thing in the narrowest possible way: it permitted zero handshakes where a page has no honest cross-domain purchase. That is precisely the repair — it made an unfillable cell legitimately empty.
⭐ But it was written as a fact about handshakes, not as a fact about mandatory fields. So the other required sections were never put through the same test, and the test takes about a minute each:
The Live Edge — can a page honestly have none? A page recording a settled, closed, fully-attested finding has no live boundary, and the header still demands three 🧭 items. Nobody has checked whether that section is being manufactured on pages where the material is simply finished.
Analytical Case Study — the same question, and the same absence of any permission to omit it.
Author Tensions & Convergences — a page built from a single unremarkable document may have neither.
⚠ This is not an argument for fewer sections. The corpus's Failures column proves the opposite case: a mandatory field is the most reliable machine for extracting something an author would otherwise skip. The distinction is the whole finding, and it is one sentence long.
A required field is a rigour instrument when it can be answered without evidence, and a fabrication engine when it cannot.
One audit, cheap, and it implicates this build. Take the fourteen headers. For each, ask: could an honest writer, working from a thin source, fill this truthfully? Handshakes — now yes, since zero is permitted. Live Edge — unknown. Case Study — unknown. Any header that fails gets the same treatment the handshake floor got: explicit permission to say none, and here is why, in one line.
And one prediction that can be checked against pages already on disk. If a mandatory section is being manufactured, the manufacture will look the same everywhere: the section will be shortest, most abstract, and most likely to restate the page's own body rather than reach outside it. That is measurable on the corpus of pages this build has produced, without reading a single source.
⭐ The uncomfortable version: this collision was found by applying a rule discovered in somebody else's corpus to the template the finding itself was written in. The page you are reading has a required section list too.
[SPECULATIVE] — the field-design rule is stated from two corpora and one template amendment. The audit has not been run and the prediction has not been tested. ⚠ Nothing here should be treated as a licence to drop sections; the proposal is explicit permission to record an honest absence, which is what the handshake amendment already established as the pattern.