14cf2f47

test: e2e — the healing leg pins the cut it restores, and names what it cannot

a73x   2026-08-28 19:53

test: e2e — the healing leg pins the cut it restores, and names what it cannot

I5. The leg passed with the heal reverted, because everything it asserted
a default tree also satisfies. It now resizes in run 1 (so the saved tree
is not the default one) and pins the thing only a restore produces: at
80x24 the aspect rule cuts BESIDE — a vertical rail near column 40, which
run 1's capture has — and a restore that reported failure falls through to
`setRootOrient` and puts that rail back. Run 2 must have no vertical rail
at all. Run 1's own rail is asserted at >= 43 first, so run 2's silence is
a difference and not an empty capture.

Graded against the heal reverted (`restoreLayout` bailing whenever a saved
leaf is unmatched, which leaves the round-trip leg's exact match alone):
"run 2 has a vertical rail at column 42, so the aspect rule cut this wall
— the sidecar was not restored at all".

The bar grep moved BELOW the rail assertions, because the cut is what
makes a bar readable: a stacked stripe is the full 80 columns and a beside
one is 39, half a socket path.

...and the leg now NAMES what it does not claim, instead of framing a
withheld assertion as a design choice. Measured, sidecar read either side
of run 2:

    saved by run 1:   beside 0 / leaf 43 #0 / leaf 36 #b / focus 0
    after the heal:   stacked 0 / leaf 1  #0 / leaf 1  #c / focus 0

The survivor's weight and the container's orientation are both lost:
dropping the unmatched leaf collapses the two-child container to a bare
leaf, and the newcomer's `insert` re-wraps that leaf in a fresh STACKED
container with default weights. That and the focus landing on the newcomer
are defects, reported rather than fixed here.