the carried entry points at its home instead of restating it

[enesismail]

UploadWedgedStops carries its measurement from UploadSlowIsNotStalled and restated about forty-six lines of that entry's reasoning, byte for byte.

Why one home rather than a row comparing two copies

The registry guard holds the two entries to agreeing on their fields — source, reason, side, instrument, window, and the measurements including the pins and their read-backs. It compares no comments at all. So the data could not drift and the prose could, and an edit to either copy reddened nothing. The duplication was a liability wearing the appearance of diligence.

A row comparing two copies is the fallback for when one home is impossible, and here it never was. Carried{From, Reason} already names the source, and carryProblem in registry_test.go already resolves that name against the registry — so the pointer is data the mechanism checks, not a cross-reference that can rot. What stays in the carrying entry is the one fact local to it, that it carries the same twelve passes, plus a sentence saying where the argument lives.

This file's own first rule, applied to itself: a constant carried between two places carries its number, not the reason the number was chosen.

The mutation did not red, and that is the reported result

The card asked whether a pointer that misdescribes its home can be made to red, and pre-authorised saying so if it cannot.

step	result
control — the package as it stands	ok, exit 0
pointer edited to say SEND where the home says RECEIVE	ok, exit 0 — no red
the guard package, in case anything there reads prose	ok, exit 0
revert	byte-identical to sha256 52370b02…, ok

A flat contradiction a reader would catch instantly passes the suite. Nothing in this estate compares comments, so the pointer's only guarantee is that it is short.

That is still the better failure mode, and it is the argument for this change rather than against it: one sentence beside a resolvable Carried.From is read in full by any reviewer, and a forty-six-line duplicate is one nobody diffs. Recorded as measured rather than dressed up as a row that does not exist.

Scope

One commit, not the two the card describes. The card's second deliverable says the shared positive control pinsWereApplied must stop holding a floor and take it from its caller. It holds no floor — it asserts both ends were ever pinned and then states the stop rule as prose. The floor is pacedFloor = 6 << 20 (upload_test.go:1067), applied inside pacingFor (:1102), which four rows and probes call. The ruling's reasoning transfers, but the function it names is not the function that does it, so that half is held for a card correction rather than retargeted here.

make ci green: exit 0, zero --- FAIL, internal/timing ok on both the race and plain legs, three declared skips and no undeclared one. surfacecheck over 0d253d0..0645117: examined 1 published surface(s), clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XfBYBGuUkF7zUnZfAFD3dW