What the turning gives back
The turning gives back the whole number at or below what it was handed. A version written as a fraction just above the first therefore reads as the first, and the fraction is reported nowhere.
So a payload whose version was damaged into a fraction is read by a reader rather than refused, and read by whichever reader the whole part names.
The complaint, and what it misses
A version that will not convert at all gets a complaint of its own, naming two grounds: a value of the wrong kind, and one whose text is no number.
A figure too large to be made whole fails on a third ground, which that complaint does not name. It escapes as it was raised, so a caller sees a failure from the arithmetic rather than anything about a version.
Why it matters here and nowhere else
A snapshot's version is never converted. It is compared as it arrived, so a snapshot's fraction is refused by name where an event's is cut down and read.
Related
Source evidence
---
row: N63-08
baseline: 49aa659
created: 2026-11-08 19:26 UTC
modified: 2026-11-08 19:26 UTC
evidence:
- claim: "An event's version is turned into a whole number before any reader is chosen, the one place in the codec where a version is converted"
source: "`nexus/infrastructure/wal_codec.py` 616, against the snapshot's at 114"
- claim: "The turning is a cutting down rather than a test, so a version written as a fraction is read as the whole number below it"
source: "`nexus/infrastructure/wal_codec.py` 616 into 621-622"
- claim: "A version that will not convert gets a complaint naming two grounds"
source: "`nexus/infrastructure/wal_codec.py` 617-619"
- claim: "A figure too large to be made whole fails on a third ground that complaint does not answer, so it escapes undressed"
source: "`nexus/infrastructure/wal_codec.py` 617, catching two kinds"
---