adding an alias to a top-level field seems to change its scope

[mvdan]

cd ${WORK}/without_alias
exec cue export .

cd ${WORK}/with_alias
exec cue export .

cd ${WORK}/with_aliasv2
exec cue export .

-- without_alias/one.cue --
package p

foo: "foo value"
one_field: foo
-- without_alias/two.cue --
package p

two_field: foo
-- with_alias/one.cue --
package p

A=foo: "foo value"

one_alias: A
one_field: foo
-- with_alias/two.cue --
package p

two_field: foo
-- with_aliasv2/one.cue --
@experiment(aliasv2)

package p

foo~A: "foo value"

one_alias: A
one_field: foo
-- with_aliasv2/two.cue --
@experiment(aliasv2)

package p

two_field: foo

As of 256bd97:

> cd ${WORK}/without_alias
$WORK/without_alias
> exec cue export .
[stdout]
{
    "foo": "foo value",
    "one_field": "foo value",
    "two_field": "foo value"
}
> cd ${WORK}/with_alias
$WORK/with_alias
> exec cue export .
[stderr]
two_field: reference "foo" not found:
    ./two.cue:3:12
[exit status 1]
FAIL: f.txtar:5: unexpected command failure
> cd ${WORK}/with_aliasv2
$WORK/with_aliasv2
> exec cue export .
[stdout]
{
    "foo": "foo value",
    "one_alias": "foo value",
    "one_field": "foo value",
    "two_field": "foo value"
}

This is very odd. Surely, adding an alias to a field declaration should not suddenly change the scope of a field.

Confusingly, this is broken with the current aliases, but works fine with aliasv2. I wonder if this is by accident.