490. Orbis, not neo
2026-09-10 - directed, continuing 489
Four inbox items serviced, all from the umbrella thread or oops-libs.
The manifest said
neo and meant orbis
The target axis has four values and they are not interchangeable:
orbis and prospero are the base machines,
neo and trinity the mid-generation refreshes.
obSCEne's probe is not Pro-specific, so its previous-generation
artifacts are orbis.
D663 got this right the same day the manifest got it
wrong. The Platform enum derives exactly those
four from Generation × Revision; the manifest copied
neo from the sketch it was written to. A vocabulary settled
in code does not settle itself in data (D666).
Renamed before obSCEne publishes, which is free because of the origin list: every entry names a sibling checkout first and the release second, so a URL that does not exist yet costs nothing while the local path answers. Sequencing recorded: orbistoun renamed first.
Two more from the same audit
target= is a third sense of the word.
Prosperous has a registered target, SELFish a build
target, this manifest an install target. A manifest entry
holds two at once - the file it fetches is named for a build target
while the field says where it lands - so the field keeps its name and
the file says which sense it is.
Generation-4/5 wording swept from the three tracked
files still carrying it. Only the wording:
data/hardware/ps5-full.txt is a real obSCEne path and
shadPS4 is somebody's project, and renaming either would
have been a different kind of wrong.
The delegation, finally
oops-libs shipped the two-argument
enable_portable_sentinel, so orbistoun's copy is gone. The
note stays here and is passed in - its words name this tool, so a shared
function hardcoding them would be wrong for every other caller. Their
crate now owns the heal orbistoun had and they lacked.
Surprises
- A filename cannot carry a compatibility
environment. obSCEne measures it as
OBS|context, and the same physical machine answers differently depending on where a run lands. That lands squarely on the conformance differential: pairing a run against a sweep by filename is the mistake I already made once, and pairing by measured context is the fix.
Next
- Pair the differential by measured
OBS|contextrather than by leg filename. - Audit findings C through E, still open.
- The corpus fetch path, still not walking the origin list.