canon_lane's exit 66 is flattened to 2 at 41 call sites, and the restart skill's lread reads every exit 2 as "no answer" #324

What happens

canon_lane returns 66, not 2, for an alias table it cannot read (lanes-edit.sh:5636-5646, :5670), because 2 already means Amendment 15(d)'s two-rows refusal or "a helper that predates the subcommand" to lane-start, lane-end, lane and lane-handoff, and a third meaning would land in the fail-open branch. Only the canon-lane subcommand carries it (:18016-18023, exit $?). Every other subcommand arm spells lane="$(canon_lane ...)" followed by exit 2 on failure: 41 sites in lanes-edit.sh (grep finds 43 matches at 33ce26d, two of them comments at :5319 and :5603). The comment at :5636-5646 assumes every caller sees 66.

The #260 review measured it: with aliases.tsv replaced by a directory, register-row exits 2 and canon-lane exits 66 (the measurement is the reviewer's, not re-run here). The skills and lane-handoff read a 2 that is not 15(d) as an old helper and fall through.

The skills have the matching half:

• skills/restart/SKILL.md:69: lread reads exit 8 and exit 2 alike as empty. The handoff skill's sread has fenced 15(d) since The handoff skill's step-1 fence refuses Amendment 15(d)'s exit 2 instead of reading it as no answer, and its window ref is checked as a whole string (#72 item 8, #80 item 4) #260 (skills/handoff/SKILL.md:97-115). lread is used at :117, :130, :140-141, :158, :183, :247 and :267; those that take a lane name can meet a case-twin register and carry on as "no answer".
• skills/handoff/SKILL.md:579: the step-4 lane-dir read has the same arm. Harmless there, because step 1 refuses a case-twin lane first.
• The REFUSED wording (skills/handoff/SKILL.md:106, lane-handoff:1458) names only case-twin rows, though log_path_ci (lanes-edit.sh:2991) and others exit 2 for 15(d) too. A nit.

Why it matters

An unreadable alias table is the case 66 was made for. Flattened, a lane whose rename alias is invisible is bound, or its row is added a second time, from a read that never happened. #310 (open) will make canon_lane refuse on an unreadable register as well, which adds 66 exits the 41 sites then flatten.

Suggested fix

Spell the 41 sites canon_lane ... || exit $? so a 66 is carried (a hygiene check can hold the line), teach lread the 15(d) fence sread has, and widen the REFUSED wording. Fold into #310's design or land first.

Found by: the #260 writer's list (lane 5, 2026-10-18T21:2xZ) items 1 to 3, and the review of #260 (2026-10-19). Filed on Brett Heap's word of 2026-10-23T18:52:34Z, by multiple choice.

Lane: openRepoTools-5