Cited package and vendor-land-pattern authorities for four mechanical candidates, plus DDR4 and M.2 pattern obligations

[gunbai-bot]

Cited package and vendor-land-pattern authorities for four mechanical candidates, plus DDR4 and M.2 family-pattern obligations and four unread-document part rows.

This PR does not select any part for any board role. An earlier revision of this branch claimed it did, and that claim is withdrawn. Selection is a functional-compatibility join over requirement, electrical interface, reference precedent, candidate part and package authority; it happens at the product layer (#8233). This PR supplies exactly one of those five inputs — the package authority — and nothing else. No row here asserts a frequency, a reset threshold, a regulator protocol or a host interface, and none of them says a part suits a role.

Reviewers of the previous head approved on the strength of MountablePart making bad states unwritable. It did not: the witness hand-built two malformed MountablePart values, which is the evidence the type was never admission-closed. That is fixed here.

What changed under the integration ruling

1. MountablePart split into a candidate and an admission. PartMechanicalCandidate is freely constructible and carries no claim. admit_part_candidate returns AdmittedMountablePart or PartRefused with a nonempty cause set, and the admitted type is reachable only through it — so the malformed values the REDs build cannot inhabit it.
2. PackageOutline → PackageBodyDimensions. The old name and its note called the body "how much volume the part occupies", which is false for any leaded package: the terminals project past the plastic, so a clearance check on the body admits boards that cannot be assembled. Body dimensions, overall terminal envelope, vendor land pattern and courtyard are now four named concepts, of which this module carries two and deliberately produces neither of the others.
3. Terminal indexing is now a required obligation on every candidate. Matching designator, count and pitch do not establish a pin map, and a mirrored pattern passes all three. There is no established arm, because modelling one means modelling the pin map. Consequence, asserted by execution: no candidate in this repository is admissible for footprint realization today, and terminal indexing is the sole reason.
4. Through-hole and surface mount no longer share an impossible success arm. The DDR4 solder-tail socket previously carried a standing whose only established arm held surface-mount land geometry — a success it could never reach whatever document arrived. Surface-mount parts keep SurfaceMountPatternStanding; parts needing a drill schedule carry MountingPatternObligation, a record with no success arm, until a plated-hole geometry carrier exists.
5. Tool failures are no longer stored as vendor facts. The DDR4 and M.2 rows now state the vendor fact — the specification publishes the board layout as a figure, and the per-part customer drawing governs on conflict — so the pattern exists and is published and is simply not transcribed here. That a fetch on a given day did not yield text is this repository's observation and belongs downstream.
6. 2308107 is a base product number, not orderable. The row is named ..._family and says so: orderable variants differ in tail length by board thickness and live on the customer drawing.

Reference-design precedent (Mt. Jade naming three of the unread parts) is a relation between a part and a particular board, so it is named in the notes as belonging at the product layer and is not carried as a property of any part here.

Evidence

dag/test/claim/board_part_land_pattern_witness_test.dag, green by execution, with discriminating REDs: a land pattern bound to the wrong package refuses on all three mechanical axes plus indexing (4 causes); an undercounted pattern refuses on the count axis alone (1 cause); and a positive control proves the surface-mount established arm is reachable, so the outstanding arms are not vacuous.