The released app showed every key's word, so a person who saw a word twice may
have hidden one of its two keys. When this version first read that hide, it
judged the board as it shows it now, one key per word, and recorded the hide
as taking the word away. The other key was then left out for the hidden one.
For example, in Core and topics in Spanish, a hidden "no" left no "no"
anywhere.
- record_kept_hides is the settle step's rule (aac-web shell.rs): a hide
found without a record on a board kept by an earlier version takes the word
only where the word showed at no other key then (only_place and no shown
mate). A key still waiting for a later stage counts as before, so in Finnish
with both "a" and "one" hidden, yksi stays hidden.
- The cell editor's hides keep today's rule (record_hides).
- Tests (aac-core tests/released_hides.rs) use the released app's own data
(tests/fixtures/released_r8_hides.json, from review-probes/vsd/r8). All
seven cases, from the stored board (update and backup restore) and from the
OBZ export, show the word once where the released app showed it, and
nowhere where it showed nowhere. With today's rule in the settle step, the
four failing cases fail.
- SPEC §4.1 item 5 is updated.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
e7ccf3c636