COST: one word. Re-priced 2026-08-30 by frankD during the Track U triage. This is a ratification, not an open design fork. The measurement that killed the decided criterion is sound and reproducible, the substitute has shipped, is pinned in v391, and is documented on the public website; option 2 would mean changing the evidence to fit the predicate, and nothing measured suggests option 3. "Ratified" closes it and the code does not move.

It stays in Track U rather than being re-filed into a lane because no agent may ratify a deviation from an owner decision on its own — but it should be read as a paperwork item that has been mis-shelved among real decisions since it was filed, which is why it has sat at p30.

Is "binds the cpyext runtime" the ratified extension-module check, or still awaiting overrule?

Track U — a decision, not work. No files.
Raised by frankD (Track D) while writing [[docs-d-name-resolution-pages-state-the-import-rule-with-no-cpyext-carve-out]].

The fork

[[decide-nilpy-import-rule-vs-a-cpyext-extension-module]] decided the criterion in its part 4: PyInit_<name>, in preferred shape (b) — the unit DECLARES itself and PyInit_<name> VERIFIES the claim.

[[feature-n-a-cpyext-extension-module-is-bare-importable-not-a-pascal-unit]] kept shape (b) and kept the declaration, but substituted the verifier, having measured that the decided one is false:

unit	`PyInit_` in its `uses`-clause C	is it `PyInit_<name>`?
`hello_ext`, `argerr_ext`, `container_ext`	matching	yes
`cyadd_ext`	`PyInit_cyadd`	no — the vendored module's name
`markupsafe_ext`	`PyInit__speedups`	no — same
`fmt_ext`	none	no — a C-API consumer, not a module

3 of 6. The two vendored cases are the normal case for a genuinely vendored extension, not a corpus quirk. What shipped instead (PyUnitDeclaresExtensionModule, compiler/pasparser_proc.inc:3035) is a lone {$PYEXTENSION} line plus pyruntime.c present in the unit source.

The resolving agent wrote: "This deviates from the decided criterion and is flagged for the owner to overrule." That flag sits in a done/ ticket, which is a session record nobody re-reads, so it has had no reader in nine days.

Why it is worth one line of your attention now

It is no longer only an internal detail. It is pinned in v391, and this ticket's sibling has just published the substituted criterion to docs/** — i.e. to the website, as the documented contract for how a unit declares itself an extension module. If the criterion is going to move, moving it before it is public costs nothing and moving it after costs a docs correction plus anyone who wrote to it.

Options

Ratify the substitution (recommended). The measurement that killed PyInit_<name> is sound and reproducible, the substitute holds for all six real units and for none of the 20 ordinary .pas files in test/nilpy_units/, and the negative test (test_nilpy_pyextension_declaration_required) pins the carve-out to the declaration rather than to what the unit links. Cost: amend the decided ticket so the record says what the code does. Docs already match.
Order PyInit_<name> restored. Requires fabricating a stub PyInit_fmt_ext and re-pointing the two vendored units' shims — i.e. changing the evidence to fit the predicate, which is the failure _ext was rejected for. Costs a docs correction.
A third verifier. Nothing measured suggests one, but the owner may want the declaration to stand alone with no check at all — simpler, at the cost of the anti-widening guard the negative test exists to hold.

Recommendation: option 1, and treat this ticket as the ratification record rather than a re-litigation. The substance was decided; only the paperwork is open.