Skip to content

BDSP AutoStory: pose certainty per step; 00.05 starts on walls - #400

Draft
dayfine wants to merge 78 commits into
bdsp/center-run-settlefrom
bdsp/start-certainty
Draft

dayfine wants to merge 78 commits into
bdsp/center-run-settlefrom
bdsp/start-certainty

Conversation

@dayfine

@dayfine dayfine commented Nov 11, 2026 •

Copy link
Copy Markdown
Owner

Depends on #375; merges in #413 (bdsp/connected-maps)

00.05 walked its route from the save's declared start (48, 62), a tile the save was never seen on; the console's start frames show (46, 60), on the plaza above the steps. Steps now declare in their action spec the pose certainty they need at the start and what they leave (shared, Pokemon/AutoStory/Pokemon_StoryCertainty.h); a lint lists steps that need more than the steps before guarantee, and BDSP has a test for it with today's 46 gaps listed as known. 00.05 and 00.04d's end now say (46, 60), and 00.05 walks to the Center north, east to the buildings' wall, west 1, north to the Center's wall, east 1. A new fault sweep starts every row 1 to 3 tiles off its declared start: 00.05 passes 11 of 12 (from a tile north the lamp at (49, 53) stops the east run); 19 rows that fail are listed as known. FRLG is unchanged: it would need its specs' certainty and the same lint. Later commits: a person who talked during a walk is marked as no wall for the replan, and ROUTE steps take "avoid" rows ([x, y] tiles, [x0, y0, x1, y1] boxes on the target map) planned as off the map; 00.05c avoids Jubilife's plaza (44..52, 20..24), where BDSP's Global Wonder Station stands and pret's map has open ground, so from the street (57..58, 22) it goes round by row 26 (red test first, lab-B 2026-11-12). With #413 merged in, BDSP's robust routes plan on the shared Pokemon::plan_robust_legs, which takes BDSP's held open legs as RobustLegCosts::hold_open_legs (off by default); every planned route is the same before and after. Tested with SerialProgramsTests --filter PokemonBDSP::AutoStory on lab/dev-bdsp.

🤖 Generated with Claude Code

https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss

…rts on walls

Each action's spec says how sure of the tile it must be at its start and
what it leaves (Pokemon_StoryCertainty.h, shared); certainty_gaps() lists
steps that need more than the steps before guarantee, a save's declared
start counting as unseen. 00.05 starts where the console shows the save
(46, 60) and walks to the Center on the buildings' wall and the Center's
front wall. New sweep: every row 1 to 3 tiles off its declared start.

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 11, 2026
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine and others added 26 commits November 11, 2026 16:21
On lab-B the 00.05 save's start is 4 taps below the row under the lamps
(row 54), not 6: the planned 6 taps passed the small brick building and
the walk north missed the Center. 00.04d now ends on the same tile.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
00.05c on console B (2026-11-11): during company_clown, a one-tile step along
the row ended in a fade (walk "transition") that step_along() ignored; the
search went on in the Poketch Company's lobby, 00.05c saved there, and
along_main_boulevard replanned from boulevard tiles until "too many plans".
Steps now count only tiles they moved, fail the step on a fade or dialogue,
and the walk back must reach home.

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

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Tap lengths, tile times per gait and axis, start delay and how open legs
are pressed (BDSP taps, FRLG timed holds on foot), with an entry for BDSP
and one for FRLG and a check of each.

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

The console save believed at (15, 19) was in the Poketch lobby. 00.05b now
ends with a step west into the signboard (13, 19), which pins (14, 19) facing
west; 00.05c starts on that tile. Console B is saved there (2026-11-11).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Maps joined by their warps and edges, with the player's place as a map
and a tile of it. Where a warp or an edge leads is computed once here;
whether a step is allowed stays per game. connect_by_world() joins maps
placed on one world grid, for BDSP's maps, which have no connections.

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

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
Each StoryWorld now keeps its maps as Pokemon::ConnectedMaps too, joined
where they meet on its grid, so the planner can move to a map and a tile
(seam 2's one world type). Nothing reads it yet; a test checks that on the
story's worlds every step off a map's edge lands where the grid puts it.

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

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
BDSP's search over the sets of tiles a player may stand on after an
uncertain open leg, lifted as a template over a step model so it can plan
on any state (a world tile, or a map and a tile).

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

plan_robust_route() and robust_route_cost() keep their signatures and plans;
the set search moves to Pokemon::plan_robust_legs(), with the story world's
step rules as its model.

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

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…h; 01.01 checks it before the gate

The clown talk confirms the coupon by its Yes/No box and skips tiles that
face a sign or someone standing still. 00.05c starts from the sign or the
street east (a step west tells them apart), re-talks the three clowns,
and fails when no Poketch shows after the president. 01.01 fails at its
start without the Poketch instead of fighting the east-exit guard.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Conflict in SourceFiles.cmake: kept both file lists.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
Conflict in the registration lists (SerialPrograms/cmake/SourceFiles.cmake ): kept both sides.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…all leg is not at its wall

Console 2026-11-11 21:00, 00.05b below_the_mart_clown: "east 8 (to wall)"
along row 34 crossed the column the boy below the Mart wanders (51,
33..35), stopped against him after 2800 ms about 5 tiles in, and the time
count (17 tiles) passed it as the wall at (56, 34).

- town_plan_costs() charges 15 a wander tile (was 3): the walk now goes
  north to row 36, east to the wall (56, 36) and up to (55, 34).
  plan_fastest_route() keeps a plan with those costs that enters no zone.
- A wall leg whose first walk took far longer than its tiles is held up,
  not at its wall: it plans again from every tile of the leg.
- Sim: SimWorldKnobs::standing and pushing_someone; tests for the boy on
  row 34 and someone on row 36 (red before the fix), a scripted held-up
  leg, and a story-table check that no 00.05b/00.05c leg enters a zone.
- talk_to_wanderer leaves its declared tile (counted taps), the 00.05c
  start probe pins its tile; known certainty gaps updated (renumbered
  00.05c step 12, 01.01 step 2). The chain test keeps the pose through
  check_poketch and the probe. 01.08's youngster case passes now.

Goldens regenerated on lab/dev-bdsp. Route bench updated
(PA_UPDATE_ROUTE_BENCH=1): 00.05b below_the_tv_clown 21463 -> 23700 ms
and 01.08 below_the_ravaged_path worst 25150 -> 26950 ms, because both
now go round a wander zone (the boy below the Mart; Route 204's
youngster) instead of through it; 00.05c beside_the_president
10500 -> 6800 ms.

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
StorySim.cpp stood at 999 of its 1000 lines, so neither #262 nor #400
could add a line. The story variables and the coord-event triggers they
arm (story_vars(), story_triggers()) move unchanged to SimTriggers.cpp;
the declarations stay in StorySim.h.
Review with: git diff --color-moved HEAD~1

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…g before the door ends off it

The Mart clown wanders (54..56, 32..34) round (55, 33) (PokemonSimulation's
jubilife_city.json; the story map leaves him out). Instead of chasing him
along the row below his home, 00.05b and 00.05c now wait for him:

- New TALK action wait_for_wanderer (WanderWaitPlan, wait_beside_wanderer()):
  from the tile the route left us on, walk column 56 tile by tile on the
  D-pad to the step's "stand" tiles, face west and press A about every 2 s
  (12 asks), then try the next stand tile; the talk counts only when his
  Yes/No box opens ("again": any talk). A step along the column that
  something stops asks that way too. After the talk it walks to the
  declared end (56, 31) and taps north and east: both must not move.
- The stand tiles and facing are in the plan step's "stand" arg: (56, 33),
  (56, 34), (56, 32), facing west. The sim talks to him most from (56, 33):
  41 of 64 starts and seeds, against 37 from (56, 34) and 28 from (56, 32).
- 00.05b comes along row 36 to the wall (56, 36); 00.05c along row 31 to the
  wall (57, 31). Both leave from (56, 31) west along row 31, above the boy
  below the Mart (51, 33..35).
- 01.08 below_the_ravaged_path ends on (11, 3), two tiles below the door
  (11, 1): its last leg was "north 1 (open)" to (11, 2), the tile before the
  door, since the higher wander cost keeps it off the youngster's row 2.
  to_banner goes north through the door from there.

Tests: Wander wait (the clown is reached from column 56 for each start
tile and seeds 1..8, from both rows' starts; the first stand tile is the
one the sim meets him most from). The Jubilife zone check also counts the
clown's box. The scene-end-pose sweep skips wait_for_wanderer, which
proves its end with the two wall taps.

Goldens regenerated (PA_UPDATE_ROUTE_PLANS=1, PA_UPDATE_ROUTE_BENCH=1):
00.05b below_the_mart_clown 8050 -> 5450 ms and below_the_tv_clown
23700 -> 27350 ms (row 31 out of (56, 31) instead of row 36 out of
(55, 34)); 00.05c below_the_mart_clown 15250 -> 20350 ms and
beside_the_president 6800 -> 12200 ms (round the clown's box along row
31); 01.08 below_the_ravaged_path worst 26950 -> 26050 ms (one leg fewer).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
dayfine and others added 6 commits November 12, 2026 02:03
Conflict in ComputerPrograms_UnitTestRunner.cpp: kept both
add_tests_ConnectedMaps and add_tests_MoveMechanics.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
lab-B 2026-11-12 01:55, 00.05c below_the_company_clown again: with the
woman who talked marked, the replan from her tile's neighbour and the
tile behind it began "west 1 (to wall)" against her. Walking into her
opened her talk again, fought as a trainer, and the walk on ran out of
time short of the "wall". This scripted walk fails now: the first call
after the replan is a walk to a wall west, into her.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
A tile marked for whoever talked was blocked like a wall, so the replan
from the tiles beside her began "west 1 (to wall)" against her, walked
into her and opened her talk again (lab-B 2026-11-12 01:55, 00.05c). A
person may turn, talk or walk off: their tile now comes off our map
(StoryWorld::uncover()), which no leg to a wall ends against, on every
path that found them by a talk: the open leg's talk after the wait, and
the A pressed at a blocker on open legs and legs to a wall.

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
dayfine and others added 4 commits November 12, 2026 02:18
…Station's plaza (red)

From (58, 22) the walk to the Poketch Company's clown plans "west 43" along
row 22, through the plaza where BDSP's Global Wonder Station stands (lab-B
2026-11-12: stopped at (50, 22), (51, 21), (49, 23), the woman at (45, 24)).

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

ROUTE steps take "avoid" rows on their target's map, [x, y] tiles or
[x0, y0, x1, y1] boxes, planned as off the map by the walk (RouteWalkPlan's
avoid) and by the table's route tests alike. 00.05c's walk to the Poketch
Company's clown avoids Jubilife's plaza (44..52, 20..24), where BDSP's
Global Wonder Station stands and pret's Diamond map has open ground: from
the street (57..58, 22) it now goes south to row 26 and round.

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
… keeps off the Wonder Station's plaza) into lab/dev-bdsp
…er) into bdsp/start-certainty

BDSP's robust routes now plan on Pokemon::plan_robust_legs. The held open
legs this line added to BDSP's own planner (#375: hold_open_legs,
per_open_grass_tile, open_hold_error(), TileLeg::held) move into the shared
planner as RobustLegCosts::hold_open_legs / per_open_wild_tile and
RobustLeg::held, off by default, so BDSP plans the same legs as before.
SourceFiles.cmake keeps both sides' new files.

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 12, 2026 03:54
…ault)

RobustLegCosts::hold_open_legs plans open legs that may end off as holds
(RobustLeg::held) that pay no per_open_tile; a held leg whose possible ends
past the shortest enter tall grass, a wander place or hop a ledge is not
used, and held legs are never joined. per_open_wild_tile adds to each wild
tile an open leg enters. open_leg() now steps each place once for all its
lengths (step_prefixes). Same code as on bdsp/start-certainty.

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

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

# Conflicts:
#	SerialPrograms/Source/Pokemon/Navigation/Pokemon_RobustLegPlanner_Tests.cpp
dayfine and others added 6 commits November 12, 2026 18:51
… at (54,16)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
… 01.06 up_jubilifes_road in the route bench

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
… strands 01.06 up_jubilifes_road in the route bench"

This reverts commit b32b47a.
…overy

A person or item with no wander box now counts as a wall in at_wall() and
in safe_wall(); only wander boxes and warps still spoil a recovery ray. This
lets 01.06 up_jubilifes_road recover at PokemonSimulation 4b886bb, where
Route 203's youngster stays at (54,16). Plan, leg-trace and bench goldens
follow; 00.04 into_moms_call now holds against Mom (3100 ms, was 2550).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VUubhHeYSMGSSDXZkKZss
…r); plan, leg-trace and bench goldens

Co-Authored-By: Claude Sonnet 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