Skip to content

FRLG AutoStory: number checkpoints by walkthrough part (01.01 to 09.02) - #363

Merged
dayfine merged 93 commits into
frlg/stage8-rocktunnelfrom
frlg/part-ids
Nov 11, 2026
Merged

dayfine merged 93 commits into
frlg/stage8-rocktunnelfrom
frlg/part-ids

Conversation

@dayfine

@dayfine dayfine commented Nov 11, 2026 •

Copy link
Copy Markdown
Owner

Depends on #293 (base frlg/stage8-rocktunnel; the diff shows only this change).

FRLG AutoStory checkpoint ids were all 00.NN, which said nothing about where in the story a checkpoint is. They are now part.NN after Bulbapedia's FireRed/LeafGreen walkthrough parts: 00.01–00.32 become 01.01–09.02 (00.05 becomes 02.01), including #345's 00.30–00.32 JSON. The story files, the ids inside them, tests, docs and replay fixtures are renamed; the fixtures are re-hashed and PokemonSimulation is pinned at d3eea1b, which has the Parcel as 02.01. This was redone on stage8 after #345 superseded the story-0030/31/32 line; the table header says why the order can differ from the walkthrough, and START/END_CHECKPOINT are saved as indexes, so settings are unchanged. Code health is clean and the app builds on macOS; the unit tests are left to fork CI.

🤖 Generated with Claude Code

https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss

dayfine and others added 30 commits November 9, 2026 10:20
Unverified authored-checkpoint drafts, outside the embedded StoryData set:
the Lavender Mart, Route 8, the Underground Path, a Pidgey for Fly, and the
Rocket Hideout to Giovanni. The README lists the whole 00.29-00.51 draft,
its conventions, the user's open decisions and the leg check.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
… aren't rows

#293's 00.29 (Route 10 to Lavender's nurse) passed on the console, so the
draft that reused the id is dropped. The README says what blocks 00.30 and
00.31 from becoming table rows: no Celadon or Rocket Hideout maps in the
locked runtime pack, no row to Celadon's Center, no spinner or lift step.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
… on the sim

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…t's trainers

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…case table

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…0.32; the steps and maps they need

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…the poster

Centers share one layout and the lift is 5x7 tiles, so the story walks
them by fixed taps (center_enter, center_exit, lift_floor from below the
door to the destination's door) and the sim builds their rooms from code
read off pokefirered. B2F/B3F spinner tiles and the Game Corner poster and
stairs are coded too, so the candidate maps need no extra regions or
objects. Bike pace and bag waits from console B (2026-11-09).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
The authored checkpoint format gains what the FRLG table uses: sections
(each event logged with its section), notes on checkpoints, sections and
events, once_per_save, and FRLG execution fields for supplies, live
start and handoff. Variants are checked against what FRLG's executor does.

StoryData/frlg-<id>.plan.json is each C++ row exported once (the export
test), split into sections, with the table's design comments as notes.
Tests compare every JSON row with its C++ row and the whole story's sim
trace with the C++ table's and a golden copy of it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
story_checkpoints() is now load_authored_story(): every StoryData plan,
in order of id. The C++ row definitions are gone (the table's rows and
the VermilionGym, Celadon and Lavender row files); the step helpers and
map names stay for tests. The migration's C++ comparison and export
tests go with them; the golden sim trace stays, with tests that a plan
with unknown fields, variants or sections is rejected and that Rock
Tunnel's events are logged with their sections.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
From Lavender's Center: the Mart's Escape Ropes and Super Repels, Route 8
on the Bicycle with only Gamer Rich fought, the Underground Path, a
Pidgey from Route 7's grass, and Celadon's Pokemon Center. The plan
loader accepts the bike, Center, item and hideout variants from the
merged drafts; the sim stubs a Center's way in when the Center has no
map, and the fault sweep ends a row there. Route 7 and Route 8 get their
land rates in the sweep; the story sim trace golden gains 00.30.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
The run plans, the checkpoint count, Rich among the trainers fought, a
bike leg after the BICYCLE, and the Pidgey's route ending on the first
grass tile; the sim trace golden regenerated with the stub report whole.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Takes the core's step types in Types/ and its story log; FRLG's section,
notes and once_per_save stay on the step, and event_log_line takes the
section as an optional argument, so lines without one are unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
On the bike a tap never only turns the player: it takes a step. The
walker's turn tap before each leg made every open leg one tile long on
the bike, and 00.30's walk to Route 8 ended on Route 12 (console B,
2026-11-09). turns_first() now skips the tap on the bike, for the
walker and the sim walker alike. The sim models the bike: a tap steps
facing either way, a hold through a turn loses no tile, and no step tap
is ignored. Bike holds measured on Route 12's pier go in the movement
mechanics table.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
A GAME step's variant is now only its action's name (mart, teach, pc,
grind, grind_check, capture, fossil, level); what the action reads moves
to StoryStep::args, and a Mart's purchases to the step's items. Each
action declares its args in the action table, so lint_story_checkpoint()
checks every step against them, and each step reads its plan with
parse_<action>_step(). The story's plans are converted; the sim trace is
unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
#314 moved the GAME steps' arguments into args; the next commit moves
00.30's steps and the use_item and lift_floor actions over.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
The use_item and lift_floor actions read their item and floors from
StoryStep::args and declare them in the action table, as #314's actions
do. 00.30's Mart and capture steps and the draft plans are converted;
the sim trace is unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Holds counted back with one-tile taps on console B (Lavender Town,
2026-11-09) moved a tile per 100ms from rest, the same each time.
At 112ms a tile, 00.30's "west 16 (open)" on Route 8 went 17 or 18
tiles and the walk missed the Underground Path's door.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Console B, 2026-11-09: three 00.30 runs stood 3 tiles east of the
walker's belief after Route 8's open legs, the hold north to the
Underground Path's door ran up to the cliff (16, 2), and the step onto
the door bumped the house. The test starts the steps after Rich's battle
from tiles up to 3 off the table's and expects Route 7 with no trainer
met. Red on the current plan: from (45, 7) it fails as the console did.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine and others added 13 commits November 10, 2026 04:14
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
00.31's B3F maze legs and 00.32's B2F legs to the MOON STONE and the lift
room are now "blind_taps": fixed taps, each waited out, never checked,
and the end trusted. Checked taps read slides as no move, tapped again
and re-planned off the route (console B, 2026-11-10 22:35).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
The lift's exit fade only dims the screen (console B, 2026-11-10, 00.32
lift_to_b4f), so the black-fade check missed it and the retries walked
into the wall outside. The lift-out leg is now south 3 marked
RoomLeg::dims: confirmed by the scene from before the leg gone and the
screen settled (RouteStep::fade_dims). Room warps name their place in
the log. The sim no longer shows a fade for a lift exit, and its room
legs fail on a warp with no fade, as the console does.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
FRLG overworld: a spinner ahead of a blocked tap is no wall
Checkpoint 00.NN ids become part.NN (01.01 to 09.02), story files,
tests, sim trace and docs renamed to match. Start/end settings are
indexes, so saved settings are unchanged. The PokemonSimulation-pinned
replay fixtures (pin, binding, route-candidate) keep their paths.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Simulation #87 renames the Parcel capture and plan to 02.01. Point the
replay pin, the copied binding, the route candidate's binding hash and
both CI checkouts at that revision.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine and others added 2 commits November 11, 2026 01:19
Redo of #363's renumbering on stage8: 00.01-00.32 become 01.01-09.02
(00.05 becomes 02.01). Story files, ids, tests, docs and replay fixtures
renamed; fixtures re-hashed and PokemonSimulation pinned at d3eea1b.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…ktunnel

Stage8 rebuilt 00.30-00.32 as JSON (#345) and supersedes the
story-0030/31/32 line this branch was on. The tree is taken whole from
frlg/part-ids-stage8, the same renumbering redone on stage8.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine and others added 3 commits November 11, 2026 02:54
…g/part-ids

# Conflicts:
#	.github/workflows/cpp-ci-serial-programs-base.yml
…ref, 02.01 id)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
@dayfine
dayfine marked this pull request as ready for review November 11, 2026 19:00
@dayfine
dayfine merged commit cb72c76 into frlg/stage8-rocktunnel Nov 11, 2026
15 checks passed
dayfine added a commit that referenced this pull request Nov 11, 2026
The #363 renumber merged after #373 left the new B4F lift-exit tests on
the old row id, so two failed and lift_to_b4f's pose check passed vacuously.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
@dayfine
dayfine deleted the frlg/part-ids branch November 13, 2026 03:26
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