Skip to content

PLZA AutoStory: no retry from a save that moved past the checkpoint's start - #619

Draft
dayfine wants to merge 5 commits into
plza/newgamefrom
plza/0304-save-moved
Draft

dayfine wants to merge 5 commits into
plza/newgamefrom
plza/0304-save-moved

Conversation

@dayfine

@dayfine dayfine commented Nov 16, 2026 •

Copy link
Copy Markdown
Owner

Base: plza/newgame (#610); the diff shows only this change.

03.04's into_pokemon_center failed twice this morning (walk_no_end) because the console's save had moved: the 03:51 run autosaved in pokedex_app and then failed in after_pokedex, so the 06:36/06:38 runs loaded a save outside the Pokemon Center and walked the legs from there. The 06:38 frames match the 03:53 after_pokedex frame exactly. The plan's legs, end event and start did not change between bb1e275 and 1cc2391. This change keeps the step an autosave showed in; while a checkpoint has one, its retry reload fails as save_moved instead of starting from past its start, and a failed row logs [Save] <id>: the game autosaved in <step> before the run failed. The load line no longer claims "as at its start" for a start nothing reads. New case in the PokemonLZA::AutoStoryGame chain tests; files syntax-checked locally, CI runs the suite.

Second change (03.07 threw outside Wild Zone 1 because 03.06 autosaved in wild_zones_tutorial and walked on to the gate): an autosave now counts as a checkpoint's save only when the badge last showed in its last step; otherwise the run logs [Save] <id>: the game autosaved in <step>, before <next> and saves nothing. Plans name that step in a new execution field autosave_step (required with "save": "autosave", added to all ten such rows), and a plan whose autosave_step is not its last step fails to load, so the built-in story test catches it before a console run. New PokemonLZA::AutoStoryGame autosave step tests and one chain case; the 03.06/03.07 split itself is on #610.

dayfine and others added 5 commits November 16, 2026 06:44
… start

Console B 2026-11-16: 03.04 autosaved in pokedex_app at 03:52 and failed in
the step after it (after_pokedex, since folded). The game's latest save was
then outside the Pokemon Center, so the 06:36 and 06:38 runs of 03.04 loaded
it, logged "off the station platform, as at its start", and walked
into_pokemon_center's legs from the wrong place: walk_no_end.

AutosaveSightings now keeps the step an autosave showed in. While it has
one, run_plza_segments' reload fails as "save_moved" instead of retrying
from past the start, and a failed row logs "[Save] <id>: the game
autosaved in <step> before the run failed: the latest save is past <id>'s
start". The load line now says the row's own start is unchecked.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AKYzhrMnPK61XHWEch4a6E
A checkpoint's autosave is the next checkpoint's start, so an autosave
seen before the last step no longer counts as the checkpoint's save: the
run logs "[Save] <id>: the game autosaved in <step>, before <next>" and
the next row has no start save. Plans name the step the badge shows in
(execution "autosave_step", required with "save": "autosave"), and
the plan fails to load when that step is not the checkpoint's last, so a
mid-checkpoint autosave fails in the unit tests, not on the console
(console B, 2026-11-16: 03.07 reloaded 03.06's autosave from before the
run to Wild Zone 1's gate).
@dayfine

dayfine commented Nov 16, 2026

Copy link
Copy Markdown
Owner Author

Follow-up (from #620's 03.08 diagnosis, not changed there): the autosave badge detector false-fires on Wild Zone 1's green zone walls. Run 1 on console B, 2026-11-16, logged an autosave badge 20.2 s into catch_more_species, but runs 2 and 3 reloaded the latest save at Taunie's rocks (marker 135-146), the 03.07 end, so no save happened there. Relevant to #604 too.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant