Skip to content

BDSP AutoStory: wait for wandering clowns inside their zone; 00.05b checks its coupons, 00.05c re-talks none - #432

Draft
dayfine wants to merge 50 commits into
bdsp/0005c-plazafrom
bdsp/wanderer-zone-stand
Draft

dayfine wants to merge 50 commits into
bdsp/0005c-plazafrom
bdsp/wanderer-zone-stand

Conversation

@dayfine

@dayfine dayfine commented Nov 12, 2026 •

Copy link
Copy Markdown
Owner

Base: bdsp/0005c-plaza (#427). The diff shows only this change.

On lab-B (2026-11-12), 00.05b failed twice at a wandering clown, and 00.05c failed once re-talking one:

  • At 11:03, mart_clown asked twelve times from (56, 32) while he stood diagonal on (55, 33). He can also stay on column 54, which column 56 can't reach.
  • At 11:47, company_clown swept row 19 below him, asking north, while he walked off.
  • At 12:48, 00.05c's re-talk to the company clown, who had already given his coupon, waited in vain.

wait_for_wanderer now stands inside both ends of a wanderer's zone: the step's stand, then a new optional far_stand. A side may also hold tiles a step in from its end, toward him. Between stands on one row or column it walks straight and asks when a step is blocked; between columns otherwise, it walks along the row above or below the zone. It asks twice per tile, moves one D-pad tap at a time, and walks round him on the way out if he blocks the column.

In 00.05b:

  • The Mart clown is waited for from (56, 32..34) facing west, then (54, 32..34) facing east.
  • The company clown from (14, 18) and (15, 18) facing east, then (16, 18) and (15, 18) facing west. The step ends against the company's sign on (14, 19).
  • A final step, check_coupons, reads Coupons 1, 2 and 3 in the Bag's Key Items. If one is missing, the row fails before its save. Only English names are known; in other languages the check is skipped and logged.

00.05c no longer talks to any clown. As on lab 4587cd7, it walks from its start to the president. Its later routes keep step 2's plaza avoid boxes, and the president's step note says he stands on (46, 35), per the decomp.

00.05c now ends on (59, 21) facing north, below the east-exit sign and west of the guard's trigger column 60; 01.01 starts there. On lab-B at 13:32 the old walk's open leg north ended a row off, the leg east to the wall stopped at the Mart's hedges, and the walk "arrived" beside them. Now the walk only goes up the main road to (53, 27), and a measured legs step (newly registered in the story table) taps south onto row 28's wall, east to the hedges, north 1, east along row 26 to its wall and north to the sign. Each leg is ended by a wall, so every tile the road's walk may end on (rows 25..27, columns 49..54) ends on (59, 21); a count that fits no tile fails the step instead of arriving.

Testing: these sim cases failed before the fix and pass now:

  • Wander wait: Mart clown diagonal to (56, 32); Mart clown staying on any tile of his box; company clown wandering or staying in his row; company clown idling at his row's other end.
  • Sim failure cases: 00.05b with a coupon missing stops at check_coupons; 00.05c's walk up the main road a row off still ends on (59, 21) and goes on into 01.01 (red on the lab's plan).

The plaza and greeter tests that walked 00.05c to the company clown now walk it to the president. The open-leg greeter test keeps its old target. The route plan and route bench goldens were regenerated. In the BDSP AutoStory suites, 359 cases pass, including the story sims, Monte Carlo and fault sweeps. Two fail for reasons outside this PR: the sim plan golden differs at 00.01, and the Resources check fails because local Resources are missing.

🤖 Generated with Claude Code

https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss

dayfine and others added 3 commits November 12, 2026 11:06
…6's reach (red)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…le and two asks at a time

On lab-B (2026-11-12 11:03) 00.05b's mart_clown asked twelve times from
(56, 32) with the clown diagonal on (55, 33), and he can stay on column 54,
out of reach of column 56. wait_for_wanderer now takes a far_stand (his
box's west column, 54, facing east) and walks between the ends along the
row above or below the box; it asks twice a tile and steps on a tile at a
time (D-pad taps), so it reaches his row. If he stays in the way after the
talk, the way out steps into the box and goes round him.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…own's moves as an enum in the tests

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine added a commit that referenced this pull request Nov 12, 2026
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine and others added 4 commits November 12, 2026 12:11
…f the swept row (red)

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

On lab-B (2026-11-12 11:47) 00.05b's company_clown swept row 19 below him,
asking north, while he walked off: no coupon question from any tile. His
zone is the one row (14..16, 18): the step is now a wait_for_wanderer on
its west end (14, 18) facing east, then its east end (16, 18) facing west,
so he has to come next to us on our row. It ends against the company's
sign on (14, 19) facing west, as the save tile was before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
… plaza too; the president stands on (46, 35)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…mpany's sign and keeping off the plaza

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
@dayfine dayfine changed the title BDSP AutoStory: wait for the Mart clown at both ends of his box, a tile at a time BDSP AutoStory: wait for wandering clowns inside both ends of their zone, a tile at a time Nov 12, 2026
dayfine added a commit that referenced this pull request Nov 12, 2026
dayfine and others added 9 commits November 12, 2026 13:15
…ns, a missing one at 00.05b's end and the company clown idling two tiles away (red)

read_coupons sets a flag for each coupon in the Key Items; check_coupons
fails when one is missing. The wander wait walks a shared row straight,
asking a blocked step.

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

00.05c reads the Key Items after its start probe and skips the talk to
each clown whose coupon is there (lab-B 2026-11-12 12:48: the re-talk to
the Poketch Company's clown, who had given his, failed the row). Its walk
to him now ends against the company's sign, where the talk ends, so a
skipped talk leaves the pose right. 00.05b ends with the same check and
fails before its save when a coupon is missing. The company clown's wait
also stands a tile in from each end of his row, toward him.

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

00.05b now ends with all three coupons checked in the Bag, so 00.05c's
re-talks (insurance for a coupon lost on 2026-11-11) go, as on lab
4587cd7: it walks from its start to the president. The plaza and
greeter tests that walked 00.05c to the company's clown walk it to the
president instead.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…ny's clown; 00.05c's certainty gap is step 6

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

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
@dayfine dayfine changed the title BDSP AutoStory: wait for wandering clowns inside both ends of their zone, a tile at a time BDSP AutoStory: wait for wandering clowns inside their zone; 00.05b checks its coupons, 00.05c re-talks none Nov 12, 2026
dayfine and others added 10 commits November 12, 2026 13:45
… each ended by a wall

The walk up the main road ends on row 27 west of the Mart's hedges; a
measured LEGS step then taps south onto row 28's wall, east to the
hedges, north 1, east along row 26 to its wall and north up column 59
to the sign. Every tile the road's walk may end on (a row or a few
columns off) ends on (59, 21) facing north, west of the guard's
trigger column 60. 01.01 starts there.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
… 00.05c's end below the east-exit sign

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…oute, not a measured walk

The measured walk inferred its start from tap counts and failed on lab-B
(2026-11-12 13:55: east moved 0). The planned route's legs are taps
(a turn tap first, a whole tile each) or run to a wall; it ends on (59, 21)
facing north from (45, 35), off the guard's column 60. Sim case for the
turn tap after the east leg, plan test, goldens.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…back from a known tile

lab-B 2026-11-12 14:24, 00.05c: "east 14 (open)" was a run into the wall
past (59, 26), and "south 15 (to wall), north 1" ran past a known tile
and back. The east-exit avoid zone uncovered its walls (so no wall held
the player there); a wall in a zone stays a wall. The robust planner
skips an open leg that ends where the leg to the wall does, and no leg
follows a wall leg from one known place straight back. plan_fastest_route()
no longer adds run-to-the-wall-and-tap-back variants (run_variants).
Route plan, bench and leg-trace goldens regenerated.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
"Simulation: plans match the golden" no longer matched the table (rows
and steps it stored are gone); regenerated from the test's own log.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
… there, not in failure (lab-B 2026-11-12 14:50)

The sim case reproduces 01.02 south 3 (to wall) from (5,17..20): the turn after
the battle counted as a moved chunk, no tile of the leg fit, and the leg failed.

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

A held run on to a wall and tapped back is fast and exact (00.04
into_twinleaf: south 37 to the wall, north 4: 10.2 s, not 26.5 s tapped), so
plan_fastest_route()'s run_variants() and the robust planner's wall-and-back
are back. Not after a tapped leg: the player is exactly where the taps left
them, and a run past that tile and back only adds a turn (00.05c, lab-B
2026-11-12 14:24). The leg that ends where its wall does stays a wall leg.
Route plan, bench, leg-trace and simulation plan goldens regenerated; the
sidestep plans in 00.03/00.04/00.04d are gone.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…its first moved chunk may be the turn

Facing another way than the leg goes, the walker turns with a tap that only
turns and the held walk counts from its first step. With the facing unknown
(after a battle) a walk that stopped with a moved chunk does not count the
first one: it may be the turn (lab-B 2026-11-12 14:55, 01.02
south_to_the_cliff after Barry: three chunks over-counted).

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

The wild escape policy ended a Run on 'no running from a Trainer battle'
only after five refused Runs. A NO_RUNNING outcome (read from the box
after Run) now stops the attempts at once and fights it like the story's
trainer battles, with the potion check after.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…its uncounted chunk out of RouteWalker

Keeps PlannedRoute.cpp within its code-health size limits.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine and others added 9 commits November 12, 2026 19:05
…one call; the wall leg's tile counts leave PlannedRoute.cpp

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

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
# Conflicts:
#	SerialPrograms/Source/PokemonBDSP/Programs/AutoStory/Data/route_bench.golden.txt
#	SerialPrograms/Source/PokemonBDSP/Programs/AutoStory/Data/route_plans.golden.txt
…ond greeter tile follow

safe_tap uses recovery_hazard instead of the removed OccupancyUse. The plan
and bench goldens are regenerated after the merge. With Mom and the still
greeter as walls the walk past Jubilife's greeter replans west, so the second
talk in the lab-B 07:44 test moves from (47, 29) to (45, 30) on the new path.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
# Conflicts:
#	SerialPrograms/Source/PokemonBDSP/Programs/AutoStory/Data/route_bench.golden.txt
…ulation's plan golden follows the anchors

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…keeps moving past the wall leaves the position unknown

lab-B 2026-11-12 20:20, 01.02 past_curtis: the 'east 3 (to wall) to find us again' hold walked out of the Gate's east door
(screen change 267, fade between frames) and was taken for found at (16, 22); every later leg ran in Oreburgh City.
The walk now ends in TRANSITION on a scene change no battle follows, reads banners through that wait, find_again and the
at-the-wall probe fail on another map's banner or on movement past the farthest wall, and a wall leg still moving at its limit fails.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
… after any interruption

lab-B 2026-11-12 21:01: camper_1 stopped the taps on (17, 22), the count read
"blocked after 9 of 13", and the walk placed us a tile or two back, marked the
beaten camper's tile and planned a six-leg loop round it; the hold to "find us
again" then ran on to picnicker_0's line. In the Gate the only way on is the east
door along row 22, so past_curtis and by_the_east_mouth (action "exit", leave
east) end any interruption in a hold east, fighting what stops us, stepping round
a wall, to the door's fade and banner; the step after takes the warp as done.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine and others added 8 commits November 12, 2026 21:56
… fade or a banner; a swirl waits for the battle
…lided across maps)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…h scene steps

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…the whole-screen change

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine added a commit that referenced this pull request Nov 13, 2026
Add the scene_ends_further fault (scene-step slide knob) and make a
hold that crosses an unseen warp stop with TRANSITION as the walker
does; person_stands now also covers non-wanderer people. Both pending
fault JSONs are active; they fail on 73f04b6 / 8a570a6 and pass on
the #432 tip. The data-driven replay test keeps discovering JSONs, so it
documents its exception to the test-case table rule.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine and others added 3 commits November 13, 2026 03:59
… the wanderer; the cave exit's row is not assumed

On lab-B (2026-11-13 02:35 and 02:37) 01.10 walked Route 204's row 28, a
row below the plan's, and "west 10 (to wall)" stopped against the youngster
who wanders x 14..16 there, 5 tiles in. The push read as stopped late, the
time counted 13 tiles, the stop passed as the wall, and every later leg ran
5 columns off until "west 1 (to wall)" ran on to "position unknown".

A wall leg whose way passes someone's wander zone (on it or a row beside)
and stops where its time counts the wall now presses A: someone who answers
is waited for and the walk goes on to the wall; someone still there after
two waits stands on a zone tile ahead, and the walk plans again from the
tiles before the zone, marking nothing.

The cave exit may leave us on row 26 or 27, so 01.09 ends and 01.10 starts
on both, and line 1 runs south to the wall below the cave instead of an
open step (a leg north could walk back into the cave). Each row's start
screen is kept in AutoStory/starts/ to settle the start pose next run.
Sim cases for both readings of the console run (red before, green now).

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

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

A start two rows below the cave (20, 29) stops line 1's south leg to the
wall early, and the walk gives up on too many plans; the start set covers
rows 26 and 27, where the cave leaves us.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine added a commit that referenced this pull request Nov 13, 2026
…eck, breakpoints, 01.10 wall legs by a wander zone

Folds what lab/dev-bdsp gained after acb34a2: #449 (Roark told by the
talk after the battle, TalkOver box tally, Trainer Card badge check), the
#432 walker fix 697e769 (a wall leg by a wander zone tells the wall from
the wanderer; 01.09/01.10 start rows; start-pose screenshots; regenerated
route goldens; 01.10 start sweep stays known failing), and the AutoStory
breakpoints with their lab pause wait.

Conflicts: PokemonBDSP_AutoStory.cpp keeps rebuild/src's run plan and
Starter RNG hook and adds the check level and the breakpoints; StoryEnv
holds both checks and run_starter_rng; the two test files take both sides.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine added a commit that referenced this pull request Nov 13, 2026
01.09/01.10 start rows and 01.10's wall leg by the wander zone (lab #432
697e769), the matching sim failure cases, the full route plans golden,
and the sim goldens rewritten for the new 01.10 plan.

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