Datasets:

ericbotti
/

connections-gameplay-sft

Modalities:

Text

Formats:

Size:

Libraries:

Dataset card Data Studio Files Files and versions

xet

Community

Dataset Viewer

Auto-converted to Parquet Duplicate

Split (1)

train · 714 rows

puzzle_id stringlengths 3 6	prompt listlengths 2 2	completion listlengths 5 15	accuracy float64 0.57 1	guess_history listlengths 4 9	categories listlengths 4 6	complete_reason stringclasses 1 value
112939	[ { "role": "system", "content": "You are playing Connections. Items are grouped by theme. Find the group members.\n\nSelect items that belong together:\n- All correct = group completed\n- Incorrect guess = 1 mistake. 4 mistakes = game over.\n- All but one member of a group = \"One Away\" hint, still costs ...	[ { "role": "assistant", "content": "", "reasoning_content": "Looking at this puzzle, I need to find patterns in these text fragments. The hint \"When the grid resolves say what you see\" and tags \"saywhatyousee\" and \"football\" suggest these fragments combine to form something visual or phonetic.\n\nL...	1	[ { "category_idx": 0, "items": [ "BO", "BB", "YC", "HA", "RLT", "ON" ], "result_message": "Correct! Group members: [`BO`, `BB`, `YC`, `HA`, `RLT`, `ON`]. Theme: Bobby Charlton.", "status": "correct" }, { "category_idx": 4, "items": [ "PA", ...	[ { "group": "Bobby Charlton", "linking_terms": [ "Bobby", "charlton" ], "members": [ "BO", "BB", "YC", "HA", "RLT", "ON" ] }, { "group": "Martin Peters", "linking_terms": [ "Martin", "Peters" ], "members": [ "MA...	all_categories_found
112938	[ { "role": "system", "content": "You are playing Connections. Items are grouped by theme. Find the group members.\n\nSelect items that belong together:\n- All correct = group completed\n- Incorrect guess = 1 mistake. 4 mistakes = game over.\n- All but one member of a group = \"One Away\" hint, still costs ...	[ { "role": "assistant", "content": "", "reasoning_content": "I need to analyze these fragments carefully. The hint \"When the grid resolves say what you see\" suggests these pieces will form words or phrases that I need to \"say what I see\" - possibly phoenetically or as complete words.\n\nLet me try to...	1	[ { "category_idx": 4, "items": [ "HA", "RR", "YK", "AN", "E" ], "result_message": "Correct! Group members: [`HA`, `RR`, `YK`, `AN`, `E`]. Theme: Harry Kane.", "status": "correct" }, { "category_idx": 0, "items": [ "BO", "BB", "YM", ...	[ { "group": "Bobby Moore", "linking_terms": [ "Bobby", "Moore" ], "members": [ "BO", "BB", "YM", "OO", "RE" ] }, { "group": "Gordon Banks", "linking_terms": [ "Gordon", "banks" ], "members": [ "GOR", "DO", ...	all_categories_found
112904	[ { "role": "system", "content": "You are playing Connections. Items are grouped by theme. Find the group members.\n\nSelect items that belong together:\n- All correct = group completed\n- Incorrect guess = 1 mistake. 4 mistakes = game over.\n- All but one member of a group = \"One Away\" hint, still costs ...	[ { "role": "assistant", "content": "", "reasoning_content": "Okay, this looks like a puzzle based on significant events in specific years. I've identified four potential groups, each corresponding to a distinct year.\n\nMy first guess will be the events from 1986:\n- `Chernobyl explodes` (April 1986)\n- ...	1	[ { "category_idx": 0, "items": [ "Chernobyl explodes", "Challenger shuttle disaster", "Halley's comet last visit", "Out of Africa wins Oscar" ], "result_message": "Correct! Group members: [`Challenger shuttle disaster`, `Halley's comet last visit`, `Out of Africa wins Oscar`, ...	[ { "group": "1986 eighties 1980s", "linking_terms": [ "1986", "eighties", "1980s" ], "members": [ "Challenger shuttle disaster", "Halley's comet last visit", "Out of Africa wins Oscar", "Chernobyl explodes" ] }, { "group": "1994 nineties 1990s", ...	all_categories_found
112883	[ { "role": "system", "content": "You are playing Connections. Items are grouped by theme. Find the group members.\n\nSelect items that belong together:\n- All correct = group completed\n- Incorrect guess = 1 mistake. 4 mistakes = game over.\n- All but one member of a group = \"One Away\" hint, still costs ...	[ { "role": "assistant", "content": "", "reasoning_content": "Let’s break this down. \n\nThe hint says: \n> Should you still be seeking a link after the grid has been resolved, the initial letters, when read from left to right, will spell out an acceptable answer. It may help to start with the 3x3 puzzl...	1	[ { "category_idx": 3, "items": [ "Ornate titi", "Emperor tamarin", "Macaque", "King colobus", "Yellow baboon", "Nilgiri langur" ], "result_message": "Correct! Group members: [`Macaque`, `Ornate titi`, `Nilgiri langur`, `King colobus`, `Emperor tamarin`, `Yellow bab...	[ { "group": "Types of cattle.", "linking_terms": [ "Cow", "cattle", "bovine", "bull", "calf" ], "members": [ "Camargue", "Ayrshire", "Texas longhorn", "Telemark", "Lincoln red", "English longhorn" ] }, { "group": "Types of ...	all_categories_found
112841	[ { "role": "system", "content": "You are playing Connections. Items are grouped by theme. Find the group members.\n\nSelect items that belong together:\n- All correct = group completed\n- Incorrect guess = 1 mistake. 4 mistakes = game over.\n- All but one member of a group = \"One Away\" hint, still costs ...	[ { "role": "assistant", "content": "", "reasoning_content": "I need to find groups of 4 items that share a common theme.\n\nLet me analyze the items:\n\n- `Jack White's band` - The White Stripes (red and white)\n- `1981 film` - Could be many films\n- `Chambered nautilus` - Has a spiral shell\n- `Staircas...	0.8	[ { "category_idx": 0, "items": [ "Dalmatians", "King of the Mountains", "Ladybirds", "Measles" ], "result_message": "Correct! Group members: [`King of the Mountains`, `Measles`, `Dalmatians`, `Ladybirds`]. Theme: All associated with spots.\\n\\nDalmatians and ladybirds have sp...	[ { "group": "All associated with spots.\\n\\nDalmatians and ladybirds have spots, as does someone with the measles.\\nIn the Tour de France the best climber is known as the King of the Mountains and wears a polka dot jersey.", "linking_terms": [ "spot", "dot", "pol", "pri", "pat...	all_categories_found
112489	[ { "role": "system", "content": "You are playing Connections. Items are grouped by theme. Find the group members.\n\nSelect items that belong together:\n- All correct = group completed\n- Incorrect guess = 1 mistake. 4 mistakes = game over.\n- All but one member of a group = \"One Away\" hint, still costs ...	[ { "role": "assistant", "content": "", "reasoning_content": "Let’s look for possible connections. \n\nSome items look like they could be parts of compound words or common phrases. \n\n- Albert, Green, White, Bath — could be place names (Albert Hall, Green Park, White House, Bath Spa) bu...	0.571429	[ { "category_idx": 1, "items": [ "Blood", "Brain", "Dura", "Soft" ], "result_message": "Correct! Group members: [`Blood`, `Dura`, `Soft`, `Brain`]. Theme: ______ Cell.", "status": "correct" }, { "category_idx": 3, "items": [ "Sports", "Dance", ...	[ { "group": "________ Room", "linking_terms": [ "Room" ], "members": [ "Wiggle", "Bath", "Staff", "Green" ] }, { "group": "______ Cell", "linking_terms": [ "Cell" ], "members": [ "Blood", "Dura", "Soft", "Brain" ]...	all_categories_found
112117	[ { "role": "system", "content": "You are playing Connections. Items are grouped by theme. Find the group members.\n\nSelect items that belong together:\n- All correct = group completed\n- Incorrect guess = 1 mistake. 4 mistakes = game over.\n- All but one member of a group = \"One Away\" hint, still costs ...	[ { "role": "assistant", "content": "", "reasoning_content": "Let's start by scanning the items for possible connections.\n\nSome items look like names: `Andrew`, `Connie`, `Prunella`, `John`, `Ballard` — possibly actors, characters, or surnames. \nOthers look like movie titles or genres: `Psycho`, `Gang...	0.571429	[ { "category_idx": 0, "items": [ "Graffiti", "Pie", "Beauty", "Psycho" ], "result_message": "One away! Incorrect guess, try again.", "status": "one_away" }, { "category_idx": 0, "items": [ "Graffiti", "Pie", "Beauty", "Gangster" ], ...	[ { "group": "American _________ movies.", "linking_terms": [ "American", "movie", "film", "picture" ], "members": [ "Graffiti", "Gangster", "Beauty", "Psycho" ] }, { "group": "_______ Chart", "linking_terms": [ "Chart", "man"...	all_categories_found
112023	[ { "role": "system", "content": "You are playing Connections. Items are grouped by theme. Find the group members.\n\nSelect items that belong together:\n- All correct = group completed\n- Incorrect guess = 1 mistake. 4 mistakes = game over.\n- All but one member of a group = \"One Away\" hint, still costs ...	[ { "role": "assistant", "content": "", "reasoning_content": "Let’s look for possible connections.\n\nSome items look like they could be parts of common phrases or compound terms:\n\n- `Bike` → could be “bike lane,” “bike rack,” “bike path,” or “bike rental” (connects to `Rent`?).\n- `Rent` → “for...	0.8	[ { "category_idx": null, "items": [ "Track", "Bike", "Court", "Floor" ], "result_message": "Incorrect guess, try again.", "status": "incorrect" }, { "category_idx": 2, "items": [ "Coffee", "Floor", "Zero", "Rent" ], "result_messa...	[ { "group": "Dirt _______", "linking_terms": [ "Dirt" ], "members": [ "Track", "Cheap", "Bag", "Bike" ] }, { "group": "________Earth", "linking_terms": [ "Earth" ], "members": [ "Planet", "Google", "Flat", "Down to" ...	all_categories_found
111834	[ { "role": "system", "content": "You are playing Connections. Items are grouped by theme. Find the group members.\n\nSelect items that belong together:\n- All correct = group completed\n- Incorrect guess = 1 mistake. 4 mistakes = game over.\n- All but one member of a group = \"One Away\" hint, still costs ...	[ { "role": "assistant", "content": "", "reasoning_content": "Let's look for patterns. \n\nSome items look like they could be verbs meaning \"to throw\": \n- `Launch` \n- `Hurl` \n- `Toss` \n- `Pitch` \n- `Lob` \n- `Chuck` \n\nThat’s 6 items, so maybe there are two groups related to throwing —...	0.571429	[ { "category_idx": 0, "items": [ "Logan", "Black", "Blue", "Mary" ], "result_message": "One away! Incorrect guess, try again.", "status": "one_away" }, { "category_idx": 3, "items": [ "Rabbit", "Hammer", "Straw", "Black" ], "resu...	[ { "group": "______ Berry", "linking_terms": [ "Berry" ], "members": [ "Logan", "Chuck", "Mary", "Blue" ] }, { "group": "Throw, chuck", "linking_terms": [ "Throw", "chuck", "synonym", "lob", "bowl" ], "members": [ ...	all_categories_found
111616	[ { "role": "system", "content": "You are playing Connections. Items are grouped by theme. Find the group members.\n\nSelect items that belong together:\n- All correct = group completed\n- Incorrect guess = 1 mistake. 4 mistakes = game over.\n- All but one member of a group = \"One Away\" hint, still costs ...	[ { "role": "assistant", "content": "", "reasoning_content": "Let’s start by scanning for possible patterns. \n\nSome items look like they could be compound phrases or idioms, maybe with a missing word in the middle or a pun. \n\nFor example: \n- `Mortar up` → maybe “mortar” + something? \n- `Flat win...	0.666667	[ { "category_idx": 2, "items": [ "Jack sheep", "Jet magic", "Penny coffee", "Egg pepper" ], "result_message": "One away! Incorrect guess, try again.", "status": "one_away" }, { "category_idx": 2, "items": [ "Jack sheep", "Jet magic", "Penny co...	[ { "group": "Man. Follows the first word, precedes the second.", "linking_terms": [ "Man" ], "members": [ "Ladies hunt", "Every Friday", "Target date", "Iron cave" ] }, { "group": "Board. Follows the first word, precedes the second.", "linking_terms": [ ...	all_categories_found
111536	[ { "role": "system", "content": "You are playing Connections. Items are grouped by theme. Find the group members.\n\nSelect items that belong together:\n- All correct = group completed\n- Incorrect guess = 1 mistake. 4 mistakes = game over.\n- All but one member of a group = \"One Away\" hint, still costs ...	[ { "role": "assistant", "content": "", "reasoning_content": "Let's look for patterns. \n\nSome items look like they could be grouped as synonyms: \n- `Rent`, `Let`, `Lease`, `Hire` — all can mean \"to pay for temporary use\" or \"to allow use for payment\". \nThat feels like a solid group. \n\nOt...	0.571429	[ { "category_idx": 3, "items": [ "Rent", "Godspell", "Hairspray", "Chess" ], "result_message": "Correct! Group members: [`Rent`, `Chess`, `Hairspray`, `Godspell`]. Theme: Musicals.", "status": "correct" }, { "category_idx": 2, "items": [ "Let", "Lea...	[ { "group": "Last word in James Bond film titles:\\nYou only live TWICE\\nDie another DAY\\nFrom Russia with LOVE\\nOn her majesty’s secret SERVICE", "linking_terms": [ "Bond", "007", "James" ], "members": [ "Twice", "Day", "Love", "Service" ] }, { ...	all_categories_found

End of preview. Expand in Data Studio

Connections Gameplay SFT Dataset

A dataset of 714 complete Connections puzzle playthroughs generated by a language model, intended for supervised fine-tuning. Each example is a full game trajectory — from the initial puzzle prompt to the final correct guess — with only winning, valid-guess-only rollouts included.

Puzzles are drawn from the train_sft split of ericbotti/connections-puzzles, intended as a warm-up for an RL training environment.

Overview

Total Examples: 714
Quality Filter: All examples won the game with no invalid guesses
Average Accuracy: 79.93% (correct guesses / total guesses)
Perfect Accuracy: 37.3% of examples (no incorrect guesses at all)

Structure

Each example is a full game trajectory in chat format using tool-calling.

Fields

puzzle_id: Matches the puzzle ID in ericbotti/connections-puzzles
prompt: The conversation prefix — a system message followed by the user's puzzle presentation. This is the model input.
completion: The model's full response sequence. Two message types:
- assistant messages: guesses submitted via tool call
  - content: empty string
  - reasoning_content: the model's chain-of-thought reasoning
  - tool_calls: list containing a single guess function call with arguments as JSON {"items": [...]}
- tool messages: environment feedback
  - tool_call_id: matches the preceding assistant tool call id
  - content: result text (e.g. "Correct! Group members: ..." or "Incorrect guess.")
accuracy: Fraction of guesses that were correct (correct_guesses / total_guesses)
guess_history: Structured record of each guess with fields:
- category_idx: Index of the matched category
- items: The guessed items
- result_message: Feedback message returned by the environment
- status: One of correct, incorrect, one_away, auto
categories: The puzzle's category definitions, each with:
- group: Category theme name
- linking_terms: Keywords used to evaluate theme guesses
- members: The words belonging to this category
complete_reason: Why the game ended (e.g. all_categories_found)

Generation

Examples were originally generated using DeepSeek V3 (deepseek-chat) via an iterative pipeline, then post-hoc migrated from XML-guess format to tool-calling format to match the current connections environment.

See the generation scripts [on Github]](https://github.com/ercbot/connections-env).

Citation

@dataset{connections_gameplay_sft_2025,
  title={Connections Gameplay SFT Dataset},
  author={Eric Botti},
  year={2025},
  url={https://huggingface.co/datasets/ericbotti/connections-gameplay-sft}
}

Downloads last month: 57

Collection including ericbotti/connections-gameplay-sft

Connections

Collection

2 items • Updated Mar 8