Skip to content

PokemonBDSP AutoStory: route around a still walk from several possible tiles - #379

Merged
dayfine merged 19 commits into
bdsp/autostory-v4from
bdsp/blocked-leg-replan
Nov 12, 2026
Merged

dayfine merged 19 commits into
bdsp/autostory-v4from
bdsp/blocked-leg-replan

Conversation

@dayfine

@dayfine dayfine commented Nov 11, 2026

Copy link
Copy Markdown
Owner

Depends on #374

On lab-B (2026-11-11, 00.05 to_the_center) a walk east that never moved from 15 possible tiles below Jubilife's Center marked nothing, since mark_blocker() only marks when one tile is possible, so every replan chose the same "east 7 (to wall)" until "too many plans". A walk that never moved from several possible tiles now marks the tiles just past them all (not the goal, not the tiles we may stand on), and the next plan routes around them. Note this alone would not have saved that run: the decomp map has kid_with_nds_29 at (52, 46), and the walks fit row 46, not 41..43, so "north 20 (open)" went about 4 tiles short; that open-leg uncertainty is a separate fix.

Tested with a new Planned route case that drives walk_route() against a small simulated console on Jubilife (red before the fix, green after), and the PokemonBDSP AutoStory suites on a lab/dev-bdsp scratch tree via SerialProgramsTests: no new failures.

🤖 Generated with Claude Code

https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss

dayfine and others added 4 commits November 11, 2026 11:22
…tiles

A walk that never moved from several possible tiles marked nothing, so
every replan took the same leg again until "too many plans" (lab-B
2026-11-11, 00.05 to_the_center). Mark the tiles just past them all,
but the goal and the tiles we may stand on.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Open legs run together in one direction were tapped as one walk:
"north 20 (open)" to Jubilife's Center ended about 4 tiles short on
lab-B (2026-11-11). Town costs now cap a straight run of open legs at
max_open_tiles, so a long stretch meets a wall on the way.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine and others added 15 commits November 11, 2026 11:51
Every story ROUTE step whose start the table knows is walked on the
simulated console from each start tile, once as planned and once with
every open leg of n tiles stopping min(4, n / 4) tiles early, unseen
(lab-B 2026-11-11: "north 20 (open)" ended about 4 short). The sim no
longer counts arrive()'s banner pauses as walker calls.

Red: 31 of 62 routes end off their target with the fault (33 on #374).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Console measurement (lab-B 2026-11-11, Jubilife): D-pad taps are
tile-exact (5, 10, 15 and 20 taps, each moved a tile) and stick runs
timed for 10 tiles overshot by about 3, three of three; none came up
short. The sim's open legs are taps and so exact; a timed stick hold
now goes 0..ceil(0.35 n) tiles further until a wall stops it. The
walker taps every open leg, so no story route sees the fault.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
The cap answered "north 20 (open)" ending 4 tiles short on lab-B. The
console measurement since (2026-11-11, Jubilife) shows D-pad taps are
tile-exact; that walk was a timed hold, which overshoots. This walker
taps every open leg, and the story-route fault sweep passes 62 of 62
with and without the cap, which only made routes longer (00.05b
company clown 27 -> 35, 00.05c president 48 -> 54, 01.06 102 -> 124).

This reverts commit 18304df.

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

lab-B 2026-11-11 13:38: "north 20 (open)" walked as a held run
overshot and the walk ended far from the Center. The simulated console
now times a hold to a limit and, with the fault on, goes up to
ceil(0.35 n) tiles past it until a wall stops it. From (48, 62) the
walk with every timed hold overshooting must still end on (52, 41);
this walker taps open legs, so it does. It replaces the open-leg cap
test.

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
@dayfine
dayfine marked this pull request as ready for review November 12, 2026 00:38
@dayfine
dayfine changed the base branch from bdsp/planned-routes to bdsp/autostory-v4 November 12, 2026 00:46
@dayfine
dayfine merged commit 9482e97 into bdsp/autostory-v4 Nov 12, 2026
20 of 23 checks passed
@dayfine
dayfine deleted the bdsp/blocked-leg-replan 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