Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
28 changes: 28 additions & 0 deletions docs/drift.md
Original file line number Diff line number Diff line change
Expand Up @@ -360,3 +360,31 @@ the near neighbour and not the same: row 27's remedy produces nothing at all and
is falsified inside the same rev by the same repo, where this one half works,
passes the obvious test, and nothing in the repo had ever contradicted it. The
tell is a remedy whose verification and whose purpose do not take the same input.

**A DRIFT REMEDY that lands in the suite and not in the document, so the copy a
reader reaches first is the one nothing holds.** haus's hand census of which
checks CI covers was deleted for the right reason — its own comment says it had
been found wrong at least seven separate times — and replaced by a check that
derives the answer, subtracting one platform's `attrNames` from the other's and
diffing that against a fixture beside itself. The check is sound and cannot go
stale. But on haus `893bb427` the same exemption was still stated in words in
five other places, four of them in the workflow file a person opens to find out
what CI runs, and the suite had been told about none of them; a comment beside
an unrelated check in the same flake then rotted pointing at "the census", which
by then named neither the list nor the check. Green said nothing about any of
it, correctly — nothing had asked it to. **Caught by** asking, of every written
copy of a fact a new check now derives, which copy a READER reaches first, and
whether the check was pointed at that one or only at the fixture it shipped
beside — then by counting the copies before calling any one of them the record,
because the count is the finding: a remedy that holds one prose copy leaves the
rest exactly as unheld as they were, and reads in its own commit like the whole
job. Row 20 is the nearest and not the same, and the difference is worth
stating: row 20's check compares a snapshot of the very data the prose
describes, so re-blessing it puts green back over a live falsehood, where this
check derives its fact from the population and is right every time it runs —
what fails here is jurisdiction, not freshness. Row 33 is the other neighbour:
there the census IS a hand list and the remedy is the check that walks the
population; here that remedy had already landed, and what drifted is the prose
it was supposed to have retired. The tell is a remedy commit that deletes a
record and adds an assertion without either one passing through the file the
record used to be for.
Loading