Write prescription links, and confirm a scanner accepts them

[bbfrederick]

The library could read a copy reference and not create one: duplicate_step
wrote empty RelationsFrom and RelationsTo entries, which is right for a
fresh scan and leaves an imported copy of a linked one silently unslaved.

link_steps

It writes the relation on the source and mirrors it onto the target as a
$ref back-pointer to the same object rather than a second copy, which is how
the console spells it. The group is checked against the menu the corpus
established, because a name outside it would put a payload in the archive that
no dialog can have produced.

The payload is an XML element whose attribute order was a guess, so the test
compares the stored string rather than decoded fields. copyparametertest
slaves eleven scans to a twelfth across all ten groups and both flags, so
stripping its links and rewriting them is an answer key rather than a sample:
all twelve come back byte for byte, Data included.

A scanner has now loaded them

An archive exercising the whole write surface at once — 54 scans: 25 sequences
that print a Special card drawn from five different exports, each carrying
one edit; and ten copy references from one source, one per menu group, on
eleven deliberately identical scans.

54 in, 54 out. Every edit survived, the six Special-card ones included.
All ten links returned with the same source, target, group and constraint, and
the eleven copies were still byte-identical. That is the one thing about
link_steps nothing could establish offline.

The console repairs a coupled set, as well as refusing one

Fifteen of the 25 came back byte-identical in ASCCONV. Nine differed only in
lScanTimeSec/lTotalScanTimeSec, the derived times Manifest.stale already
names — that is what the eja_* spectroscopy scans showing greyed during
import turned out to be, and nothing in them was damaged.

rslh_ep3d_vaso differed in three more, and they are exactly linear in the TR
that moved:

alTI[0] = 24 * TR + 10320
alTI[1] = 72 * TR + 10320
alTR[1] = 96 * TR + 20640

exact at both the old and the new value. The console recomputed a coupled set
left stale rather than refusing the protocol.

That is a second outcome beside the greyed-out scan, and it is easy to mistake
for plain success: where the sequence can derive the rest it does, and the
protocol returns consistent and changed in more places than were written.
The rule is unchanged — which outcome a given incomplete write gets is still
only knowable on a scanner, and Include Nav. = Off is the same shape and was
refused — but "rejected" and "silently repaired" are both live, and only a
diff of what came back separates them.

Testing

2349 passed, 17 skipped. Two new tests: the byte-for-byte relink against
copyparametertest, and a refusal for a group outside the console's menu.

🤖 Generated with Claude Code