Long-horizon RPG & adventure
Progress through RPGs and point-and-click adventures.
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- Pokémon (emulated) · Ace Attorney
- Class GUI-need label
G3· GUI-only
Task proposals · 20
Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.
R13.01.02.02-T01 Earn the Boulder Badge in Pokémon Red from a new game
Original instruction
An emulator (mGBA, harness-supplied ROM) shows Pokémon Red at the title screen with no save. Start a new game and progress until you defeat Brock in Pewter City and receive the Boulder Badge. Savestates, fast-forward and cheats are disabled; in-game saving is allowed.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Pokémon (emulated)
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
key· Shortcutsvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Harness reads WRAM: badge bitfield bit 0 set. Milestone flags (starter chosen, Oak's parcel delivered, Viridian Forest exited, Pewter reached) from event flags. mGBA scripting is disabled for the agent; input log replay verifies.
Proposed partial credit
30% starter chosen and parcel delivered; 20% Viridian Forest cleared; 10% Pewter reached; 40% Boulder Badge
R13.01.02.02-T02 Catch a Pikachu in Viridian Forest
Original instruction
mGBA loads a Pokémon Red in-game save at the Viridian Forest entrance: party of a level-12 Charmander, 10 Poké Balls, 5 Potions. Catch a Pikachu (about 5% encounter rate) and keep it in the party or PC.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Pokémon (emulated)
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
key· Shortcutsvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
WRAM party and PC box species lists must contain species 0x54 (Pikachu) with original trainer ID = player. Savestates/cheats disabled.
Proposed partial credit
30% Pikachu encountered (battle log flag); 70% caught
R13.01.02.02-T03 Reorder the party so Pikachu leads
Original instruction
mGBA loads a Pokémon Red save in the Pewter Pokémon Center with a party of Charmander, Pidgey and Pikachu (in that order). Use the start-menu party screen to move Pikachu to the first slot and save the game in-game.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Pokémon (emulated)
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
key· Shortcutsvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 1 / 5
Proposed verifier
In-game save file (.sav) parsed: party slot 0 species = Pikachu; other members still present. mGBA scripting, savestates, cheats and memory tools are disabled for the agent and no shell is available, so the .sav can only be written by the in-game save.
Proposed partial credit
70% Pikachu in slot 1; 30% saved in-game
R13.01.02.02-T04 Buy five Potions at the Viridian Mart
Original instruction
mGBA loads a Pokémon Red save outside the Viridian City Mart after Oak's parcel has been delivered, with ₽3,000. Enter the Mart, buy exactly 5 Potions and save the game.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Pokémon (emulated)
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
key· Shortcutsvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Save parsed: bag Potion count = initial + 5 and money = 3,000 − 5 × 300. Buying more or other items fails the exactness milestone. mGBA scripting, savestates, cheats and memory tools are disabled for the agent and no shell is available, so the .sav can only be written by the in-game save.
Proposed partial credit
40% Mart entered and purchase made; 60% exactly 5 Potions with correct money
R13.01.02.02-T05 Defeat Roxanne for the first gym badge in Pokémon Emerald
Original instruction
mGBA loads Pokémon Emerald from a save just after choosing Treecko in Littleroot. Progress to Rustboro City and defeat gym leader Roxanne. Use in-game saves only.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Pokémon (emulated)
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
key· Shortcutsvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Save/WRAM flags: Stone Badge flag set. Milestones from event flags: rival battle on Route 103, Petalburg visit with Norman, Petalburg Woods cleared, Rustboro reached. mGBA scripting, savestates, cheats and memory tools are disabled for the agent, and the input log is replayed to confirm the run.
Proposed partial credit
30% rival battle and Petalburg visit; 20% Woods cleared; 10% Rustboro reached; 40% Stone Badge
R13.01.02.02-T06 Present the decisive evidence in Ace Attorney case 1
Original instruction
Phoenix Wright: Ace Attorney Trilogy (PC) is loaded at the start of 'The First Turnabout' cross-examination of Frank Sahwit. Press and present evidence until the case is won. You start with 5 penalty marks.
- Platforms
win· Windows- Applications
- Ace Attorney
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Save slot written at the chapter end is parsed: episode 1 completion flag set. Penalties taken are logged from save snapshots; the case must be won without a game over. The agent has no shell or file-manager access, and the harness logs each write to the save folder by process, so a save not written by the game is detected.
Proposed partial credit
30% first contradiction found; 30% clock contradiction; 40% verdict 'Not guilty'
R13.01.02.02-T07 Find all clips mentioning a keyword in Her Story
Original instruction
Her Story (PC) is at the police-database desktop. Using the in-game database search, find and view every clip in which the interviewee says 'mirror' (there are 5), and add each to the session tray.
- Platforms
win· Windowsmac· macOS- Applications
- Her Story
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
type· Text & formsclick· Point & clickav· Audio/video over time- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Game save (viewed-clip list and session tray) parsed: the 5 ground-truth clip IDs viewed and in the tray. Search returns only 5 clips at a time, so related keywords are needed to reach all of them. The agent has no shell or file-manager access, and writes to the save folder are logged by process, so a save not written by the game is detected.
Proposed partial credit
50% scaled by target clips viewed; 50% scaled by target clips in session tray
R13.01.02.02-T08 Reach Fort Kearney in Oregon Trail with three party members alive
Original instruction
The Oregon Trail Deluxe (DOSBox) is at the start screen. Choose occupation, buy supplies at Matt's General Store and travel to Fort Kearney with at least 3 of 5 party members alive and at least 1 ox.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Oregon Trail
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clicktype· Text & formsvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Harness reads game memory: landmark index ≥ Fort Kearney, alive count ≥ 3, oxen ≥ 1. Fallback: checkpoint screen hash plus OCR of the party status. The DOSBox debugger and config file are not accessible to the agent, so game memory cannot be patched; input is OS key events only.
Proposed partial credit
20% supplies bought; 30% scaled by distance; 30% Fort Kearney reached; 20% party alive
R13.01.02.02-T09 Retrieve the shield and the sword in Link's Awakening
Original instruction
mGBA loads The Legend of Zelda: Link's Awakening DX (harness-supplied ROM) from a new game. Leave Marin's house with the shield and retrieve the sword on Toronbo Shores.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Zelda (emulated)
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
key· Shortcutsvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
WRAM inventory flags: shield and sword obtained. Savestates and cheats disabled; input replay confirms.
Proposed partial credit
40% shield; 60% sword
R13.01.02.02-T10 Win the insult sword fight against the Sword Master in Monkey Island
Original instruction
ScummVM runs The Secret of Monkey Island (harness-supplied data) from a save at the start of Part I's three trials. Learn insults and comebacks by duelling pirates, find the Sword Master's location and defeat her in insult sword fighting.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- ScummVM
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
ScummVM save (variables dump via the harness) shows the Sword Master defeated flag and the T-shirt item in inventory. ScummVM debugger is disabled.
Proposed partial credit
30% scaled by insult/comeback pairs learned; 20% Sword Master location found; 50% Sword Master defeated
R13.01.02.02-T11 Finish the Ruins in Undertale without killing anything
Original instruction
Undertale (PC) new game. Complete the Ruins and leave through the door after Toriel's fight, sparing every monster and Toriel.
- Platforms
win· Windowsmac· macOS- Applications
- Undertale
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
key· Shortcutsvis· Visual judgmentrt· Real-time control- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Save file (file0) parsed: area past the Ruins door reached; kill counter = 0; Toriel spared flag set. Debug mode not enabled. The agent has no shell or text-editor access, and writes to the save folder are logged by process, so a file0 not written by the game is detected.
Proposed partial credit
30% Ruins puzzles completed; 30% Toriel spared; 40% exit with zero kills
R13.01.02.02-T12 Defeat Garland and rescue the princess in Final Fantasy (NES)
Original instruction
An NES emulator (harness-supplied ROM) starts Final Fantasy with a preset party (Fighter, Thief, White Mage, Black Mage). Level up as needed, go to the Temple of Fiends, defeat Garland and return Princess Sara to Corneria Castle so the bridge is built.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Final Fantasy (emulated)
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
key· Shortcutsvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 4 / 5
Proposed verifier
WRAM event flags: Garland defeated and bridge built. Savestates disabled; in-game saving allowed.
Proposed partial credit
20% party reaches level 4; 40% Garland defeated; 40% bridge built
R13.01.02.02-T13 Win an Ace Attorney trial day keeping three health segments
Original instruction
Ace Attorney Trilogy (PC) save at the start of trial day 1 of 'Turnabout Sisters'. Complete the day's trial with at least 30% of the health bar remaining, pressing statements to reveal hidden testimony where needed.
- Platforms
win· Windows- Applications
- Ace Attorney
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Save at day end parsed: chapter progress flag for trial day 1 complete; health value ≥ 30% (read from the save). The agent has no shell or file-manager access, and the harness logs each write to the save folder by process, so a save not written by the game is detected.
Proposed partial credit
60% scaled by trial segments cleared; 40% health threshold met at end
R13.01.02.02-T14 Obtain and hatch the Togepi egg in Pokémon Crystal
Original instruction
mGBA loads Pokémon Crystal from a save just after receiving the Zephyr Badge. Return to Violet City's Pokémon Center to receive the Mystery Egg (Togepi) from Elm's aide and walk until it hatches.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Pokémon (emulated)
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
key· Shortcutsnav· Pan/zoom/scroll- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Save/WRAM: party contains Togepi (species 175) with is-egg flag cleared. Savestates and speed-up disabled.
Proposed partial credit
40% egg received; 60% egg hatched
R13.01.02.02-T15 Collect all investigation evidence in an Ace Attorney chapter
Original instruction
Ace Attorney Trilogy (PC) save at the start of the investigation phase of 'Turnabout Sisters'. Examine locations and talk to characters to collect all evidence items required before the trial begins.
- Platforms
win· Windows- Applications
- Ace Attorney
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Save parsed: court-record evidence IDs equal the full investigation set and trial-start flag set. The agent has no shell or file-manager access, and the harness logs each write to the save folder by process, so a save not written by the game is detected.
Proposed partial credit
70% scaled by evidence items collected; 30% investigation completed
R13.01.02.02-T16 Escape the factory level in Beneath a Steel Sky
Original instruction
ScummVM runs Beneath a Steel Sky (freeware data) from the beginning. Solve the opening puzzles with Joey, leave the factory area and reach the level-2 elevator to the city.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- ScummVM
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
ScummVM save parsed: room ID of the level-2 area reached and Joey in the party. Debugger disabled.
Proposed partial credit
30% Joey activated; 30% factory exit puzzles; 40% level-2 reached
R13.01.02.02-T17 Spare Papyrus in Undertale
Original instruction
Undertale save at the Snowdin Forest save point before the Papyrus encounter, with no kills. Reach the Papyrus boss fight and end it by sparing him (survive his special attack).
- Platforms
win· Windowsmac· macOS- Applications
- Undertale
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
key· Shortcutsrt· Real-time controlvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Save parsed: Papyrus spared flag set, kills still 0. Debug mode is not enabled, the agent has no shell or text-editor access, and writes to the save folder are logged by process, so a save not written by the game is detected.
Proposed partial credit
30% fight reached; 70% Papyrus spared
R13.01.02.02-T18 Teach Cut and clear the tree to the Vermilion gym
Original instruction
mGBA loads Pokémon Red after the S.S. Anne with HM01 in the bag. Teach Cut to a compatible party Pokémon (replacing a move of your choice), cut the tree blocking the Vermilion Gym and enter the gym.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Pokémon (emulated)
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
key· Shortcutsvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
WRAM: a party member knows Cut; player map ID = Vermilion Gym. mGBA scripting, savestates, cheats and memory tools are disabled for the agent, and the input log is replayed to confirm the run.
Proposed partial credit
50% Cut taught; 50% gym entered
R13.01.02.02-T19 Reach Oregon City with most of the party and supplies
Original instruction
The Oregon Trail Deluxe (DOSBox), new game as a banker. Reach the Willamette Valley at the end of the trail with at least 3 party members alive, choosing river crossings and rafting the Columbia River yourself.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Oregon Trail
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickrt· Real-time controlvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Final score screen OCR plus memory: trail completed, alive count ≥ 3. Rafting minigame requires real-time steering. The DOSBox debugger and config file are not accessible to the agent, so game memory cannot be patched; input is OS key events only.
Proposed partial credit
50% scaled by landmarks passed; 20% Columbia rafting completed; 30% arrival with ≥ 3 alive
R13.01.02.02-T20 Clear the first dungeon in Link's Awakening
Original instruction
mGBA loads Link's Awakening DX with the sword and shield. Obtain the Tail Key from the Mysterious Woods, open Tail Cave and defeat its boss to get the Full Moon Cello.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Zelda (emulated)
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
key· Shortcutsrt· Real-time controlnav· Pan/zoom/scroll- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 4 / 5
Proposed verifier
WRAM: Tail Key flag, dungeon 1 completion and instrument 1 obtained. mGBA scripting, savestates, cheats and memory tools are disabled for the agent, and the input log is replayed to confirm the run.
Proposed partial credit
25% Tail Key; 25% Tail Cave entered; 20% Roc's Feather obtained; 30% Full Moon Cello
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.