Repository navigation
Conversation
dayfine
added a commit
that referenced
this pull request
Nov 16, 2026
…from the marker walk_toward_marker() and walk_back_from_marker() walk until the objective marker's distance reads a given value; moved here from #604 so this PR owns MarkerWalk's code.
Name the drawn plate boxes in the caught-mark tests, comment the filter range arguments, and keep party_bar_filled_slots in the detector's .cpp, since only its tests call it.
StoryBackend::log_detail() writes the file log only (log() by default); PLZA's console backend skips the overlay there and FakeStoryBackend records it in lines() but not overlay_lines(). The scene machine sends its per-screen lines and press summaries through it; black fades (with their count, "fade 2/3"), the banner and the controls hint showing, and the step's own lines stay on the overlay.
01.01's arrival on the Italian game waited 110 s on "MISSIONE PRINCIPALE 01" with no press; one A cleared it and the third black fade followed (console B, 2026-11-15). With end_after_black_fades, a banner up 2 s now gets A (banner_press), up to 3 times 1 s apart, and BANNER_SHOWING may come up 120 times before the run fails as scene_stuck: its frames change too much for max_still to end the wait.
A building banner's mint (about 96, 203, 147) in the strips above and below the icon counted as cyan and hid the English badge on 01.01 (en_t166). Both cyan tests now need blue at least green less 20, which also keeps a green sign's glowing text from matching the icon. Over 6096 console captures and 1173 PokemonLZA test frames: en_t166 found, 3 green-sign false matches dropped, nothing else changed.
The panel grows left to fit its longest row. In Italian (01.02, Villy) the bottom row's text ran into the right strip the reader sampled, so the row read as "other" and no rows were found. The right strip now sits past any row's text (x 1698-1712), the highlight's left end is found by scanning, and a plain row is told from a blue wall by bright text anywhere between the highlight's left end and the text's right limit.
…nu) — drop before upstream
…ind-pressed 01.02's battle area hands control back with only the "A Talk" prompt to Ryon at the gate: no controls hint, and no party bar (no Pokemon yet). The scene machine took the prompt for nothing, blind-pressed A 5 s later and restarted his talk, for 3 min (console B 2026-11-15, Italian). The prompt-as-control fix of 2026-11-10 (13a6c3c) was lost when the scene became a ScreenMachine. In machines that end on control (not stop_at_choice / stop_at_battle, whose talk shows the prompt between lines and whose blind A may start it), the prompt is now a sign of control: no blind A while it shows, and held a second past min_length it ends the step (talk_prompt). 01.02's plan says why. Frame test: the Italian "A Parla" prompt (CommandLineTests plza/new-game-intro 848e0b0).
…ting a talk 01.02's battle area blind-pressed Ryon's talk open again and again for 3 min (console B 2026-11-15); reproducing it needs a wiped save. Now when blind presses (after a quiet spell, blind_press_after above zero) bring back text 4 times in a row with only text and cutscenes between (no fade, banner, choice or sign of control), the machine stops with "scene_loop" at that frame instead of pressing on. story_battle_start and talk_to_pokedex_screen, whose blind press every 2 s is their pace through a talk, are not watched. Sim tests: the loop from the log, a fade resetting the count, and a long battle talk.
…pt) — drop before upstream
…s restarting a talk" This reverts commit 4afd5d6.
…enes
The scene_loop guard (reverted in the commit before) stopped a passing
01.01 run at Jacinthe's subtitles: the blind A every 2 s it watched was
the problem, not the text. Now every screen the machine presses on is one
a detector classifies: text box or advance chevron (subtitles too) A,
mission complete card A, tutorial card A (new state, TUTORIAL_CARD_READING),
banner up 2 s with end_after_black_fades A. Nothing classified waits and
is logged. mash_to_move_hints and story_battle_start drop their blind
presses; only talk_to_pokedex_screen keeps blind B, for Pokedex screens
no detector reads. A still screen's wait ends on any other state (a still
subtitle with its chevron). Each run logs the states its step ends on
("start, expecting ...") and its failures name them.
Replay tests (PokemonLZA_AutoStorySceneReplay_Tests): the machine's logs
of two passing Italian 01.01 runs, the run the guard stopped, 01.02's A
Parla loop and an English 01.02 pass, replayed look by look, must reach
the plan's end state.
…er_started 02.02's street scene moved 41 s with no text and no HUD after its black fade; the scene step failed as scene_never_started at 45 s (console B, 2026-11-15). With no HUD (party bar, controls hint, A prompt), a fade or a moving screen now counts as the scene under way. A step with a start window waits on the HUD before its scene (awaiting_scene) and fails as scene_never_started only when the HUD has shown for the window with no scene. The 02.02 log is a replay case (the controls hint after the cutscene is synthetic).
Brings in the autosave sightings (slice 2b and the plan flags in slices 3, 4 and 5). Conflict in PokemonLZA_AutoStoryGame_Tests.cpp: this slice's cases kept, with 01.00's "autosaved (seen)" expectation.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AKYzhrMnPK61XHWEch4a6E
Owner
Author
|
Superseded by #644 (one lean branch; split for review later) |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Depends on #598 (base
plza/slice5); the diff shows only this change.Slice 5b of the lean PLZA AutoStory chain: checkpoint 03.08, the last piece of plza/autostory's story the chain did not cover. 03.08 patrols Wild Zone 1's road between the catch spots at 100 and 60 (by the objective marker), throws only at lock-on targets whose name plate shows no caught mark, and is done when the party bar shows 5 filled slots. Each catch autosaves, and the run logs showed that autosave mid-checkpoint, so the checkpoint is split there: 03.08 (the catches, autosave) and 03.08b (back to the road, face heading 265 on the minimap and run to the far gate with the existing
run_to_gate, then out with upstream'sleave_zone_gate; no save). Ported from plza/autostory: the caught-mark reader and the party bar detector. Left out: the lock-on bracket watch and sweep, nearest-target choice and walk-up, the throw-again-after-a-miss logic, the Registered! banner tally and species check, and the far gate's pillar reader (the run is steered by heading only). The stop-at-distance marker walk (toward and back) it uses lives in #598, which owns the marker walk files.Tests: synthetic-frame tests for the caught-mark reader, the party bar's frame tests (frames are in the fork's CommandLineTests only), and the story data and console-only run cases for 03.08/03.08b. No console run yet.