{
 "name": "LLM Test Prompts",
 "url": "https://prompts.branolabs.com",
 "updated": "2026-10-22",
 "license_note": "Each prompt keeps its own licence (see source.license). Our own text (summaries, grading notes) is CC BY 4.0; our translations of CC BY-NC-SA prompts are CC BY-NC-SA 4.0.",
 "note": "Entries with source.reuse = 'excerpt+link' contain only a short excerpt; open source.url for the full prompt.",
 "count": 62,
 "categories": [
  {
   "slug": "simulation",
   "name": "Simulations & physics",
   "blurb": "Physics, agents and systems that must stay internally consistent."
  },
  {
   "slug": "game",
   "name": "Games",
   "blurb": "Playable games with real game systems, usually in one file."
  },
  {
   "slug": "app",
   "name": "Apps",
   "blurb": "Real applications rather than demos."
  },
  {
   "slug": "creative-coding",
   "name": "Creative coding & SVG",
   "blurb": "Visual output from code: SVG, ASCII art, web design."
  },
  {
   "slug": "agentic",
   "name": "Agent tasks",
   "blurb": "Coding-agent tasks in a real folder, with safety rules."
  },
  {
   "slug": "reasoning-traps",
   "name": "Reasoning traps",
   "blurb": "Altered classic puzzles that catch memorised answers."
  },
  {
   "slug": "common-sense",
   "name": "Common sense",
   "blurb": "Everyday physics and social reasoning hidden among distractors."
  },
  {
   "slug": "instruction-following",
   "name": "Instruction following",
   "blurb": "Format constraints you can check by machine."
  },
  {
   "slug": "tool-use",
   "name": "Tool use & JSON",
   "blurb": "Function calling: the right tool, the right arguments, valid JSON, or no call at all."
  }
 ],
 "sources": [
  {
   "id": "arnie",
   "name": "Arnie936 / llm-prompting-tests",
   "author": "Arnie (GitHub @Arnie936)",
   "repo": "https://github.com/Arnie936/llm-prompting-tests",
   "license": "No license file (all rights reserved)",
   "license_url": null,
   "reuse": "excerpt+link",
   "note": "The quality bar for this library. There's no license, so we show a short excerpt, our own summary and grading notes, and link to the exact prompt."
  },
  {
   "id": "kcores",
   "name": "KCORES LLM Arena",
   "author": "karminski (KCORES team)",
   "repo": "https://github.com/KCORES/kcores-llm-arena",
   "license": "KCORES License 1.0 (custom: copies must carry the full licence text and public source)",
   "license_url": "https://github.com/KCORES/kcores-llm-arena/blob/main/LICENSE_en-US",
   "reuse": "excerpt+link",
   "note": "Each test comes with a published point-by-point scoring rubric. The licence attaches conditions to copies, so we link to the full prompt and summarise the rubric in our own words."
  },
  {
   "id": "kgist",
   "name": "KCORES prompt collection (gist)",
   "author": "karminski-牙医",
   "repo": "https://gist.github.com/karminski/d71b4c83b761538002593dec0014a21a",
   "license": "CC BY-NC-SA 4.0",
   "license_url": "https://creativecommons.org/licenses/by-nc-sa/4.0/",
   "reuse": "full-text",
   "note": "Quoted in full with attribution for non-commercial use. The English translations are ours (BranoLabs), marked as translations and shared under the same licence."
  },
  {
   "id": "ea",
   "name": "LocalLLM_VisualCodeTest",
   "author": "David Harris (GitHub @electricazimuth)",
   "repo": "https://github.com/electricazimuth/LocalLLM_VisualCodeTest",
   "license": "No license file (all rights reserved)",
   "license_url": null,
   "reuse": "excerpt+link",
   "note": "Built for testing local models through llama.cpp or KoboldCpp, with published side-by-side results. No license, so excerpt plus link."
  },
  {
   "id": "fc",
   "name": "flight-combat-llm-comp",
   "author": "Alexandru Vasile (GitHub @AlexTzk)",
   "repo": "https://github.com/AlexTzk/flight-combat-llm-comp",
   "license": "No license file (all rights reserved)",
   "license_url": null,
   "reuse": "excerpt+link",
   "note": "The same prompts run on 9 local MLX models, with the outputs published next to them. No license, so excerpt plus link."
  },
  {
   "id": "ma",
   "name": "Misguided Attention",
   "author": "GitHub @cpldcpu and contributors",
   "repo": "https://github.com/cpldcpu/MisguidedAttention",
   "license": "CC0 1.0",
   "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
   "reuse": "full-text",
   "note": "Public-domain dedication, quoted in full. Only prompts from the public README: the author keeps the longer eval set scrambled to limit training contamination, and we respect that."
  },
  {
   "id": "simplebench",
   "name": "SimpleBench (public sample)",
   "author": "SimpleBench team",
   "repo": "https://github.com/simple-bench/SimpleBench",
   "license": "MIT",
   "license_url": "https://github.com/simple-bench/SimpleBench/blob/main/LICENSE",
   "reuse": "full-text",
   "note": "Quoted in full under MIT with the official answer key. Only 10 questions are public; the real benchmark is private."
  },
  {
   "id": "ifeval",
   "name": "IFEval (Instruction-Following Eval)",
   "author": "Google Research (Zhou et al., 2023)",
   "repo": "https://github.com/google-research/google-research/tree/master/instruction_following_eval",
   "license": "Apache-2.0",
   "license_url": "https://github.com/google-research/google-research/blob/master/LICENSE",
   "reuse": "full-text",
   "note": "Quoted in full under Apache-2.0. Every constraint is machine-checkable; this site's checker runs similar checks in your browser."
  },
  {
   "id": "bfcl",
   "name": "Berkeley Function Calling Leaderboard (BFCL)",
   "author": "Gorilla team, UC Berkeley",
   "repo": "https://github.com/ShishirPatil/gorilla/tree/main/berkeley-function-call-leaderboard",
   "license": "Apache-2.0",
   "license_url": "https://github.com/ShishirPatil/gorilla/blob/main/LICENSE",
   "reuse": "full-text",
   "note": "Question, function schemas and ground truth from the BFCL data files under Apache-2.0. Our change: wrapped into one copy-paste prompt for chat apps without a tool API."
  },
  {
   "id": "pelican",
   "name": "Pelican on a bicycle",
   "author": "Simon Willison",
   "repo": "https://simonwillison.net/2024/Oct/25/pelicans-on-a-bicycle/",
   "license": "No explicit license; one-sentence prompt quoted as a citation",
   "license_url": null,
   "reuse": "short-quote",
   "note": "A single sentence, quoted with attribution."
  },
  {
   "id": "flavio",
   "name": "Spinning hexagon (X post)",
   "author": "Flavio Adamo (@flavioAd)",
   "repo": "https://x.com/flavioAd/status/1885449107436679394",
   "license": "No explicit license; two-sentence prompt quoted as a citation",
   "license_url": null,
   "reuse": "short-quote",
   "note": "The viral o3-mini vs DeepSeek R1 test from 31 Jan 2025, quoted with attribution."
  }
 ],
 "quick_pack": [
  "kilo-feathers-pound-steel",
  "inverse-monty-hall",
  "machines-widgets",
  "jugs-impossible-4l",
  "taped-sandwich-stack",
  "no-comma-no-c-250-words",
  "bfcl-circle-area-circumference",
  "bfcl-irrelevant-tool",
  "pelican-bicycle-svg",
  "ball-spinning-hexagon"
 ],
 "prompts": [
  {
   "slug": "bee-hive-simulation",
   "title": "Bee hive simulation with self-checked invariants",
   "kind": "build",
   "category": "simulation",
   "difficulty": "very hard",
   "language": "de",
   "tests": "Coupled subsystems (cells, three resource pools, brood cycle, bee roles), a fixed 10 ticks/s logic loop decoupled from rendering, a priority order for features, and a selfCheck() that must test invariants rather than existence.",
   "excerpt": "ROLLE Du baust interaktive Simulationen als Single-File-Web-Apps. Deine Stärke ist, dass die Systemlogik unter der Oberfläche wirklich stimmt — nicht nur die Optik. AUFGABE Erstelle eine interaktive Simulation eines Bienenstocks …",
   "expected": "Open the file from file:// and read the selfCheck table. Full marks only if honey can only appear via stored nectar dried below 20 % water; the resource balance holds exactly (collected = stored + used + in transit); the brood counter matches the cells; logic runs at 10 ticks/s independent of FPS, and 2× doubles ticks, not frames; the stress-test button applies the stated settings for 30 sim days and prints a report; colony collapse stops the sim and shows the cause chain. The 'Design decisions' block must name five real trade-offs.",
   "failure_modes": "Honey cells that skip the nectar phase; a selfCheck that only tests whether the canvas exists; one DOM element per bee (lags at a few hundred agents); half-built features with no note in the status area, which the prompt explicitly forbids.",
   "tags": [
    "single-file HTML",
    "canvas",
    "invariants",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 6637,
   "approx_tokens": 1900,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#1-bienenstock-simulation",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/bee-hive-simulation"
  },
  {
   "slug": "motorbike-racing-3d",
   "title": "3D motorbike racing game through a forest",
   "kind": "build",
   "category": "game",
   "difficulty": "hard",
   "language": "de",
   "tests": "3D rendering plus vehicle physics (inertia, drift, surface grip), AI riders that follow the track and make mistakes, lap timing and a debug overlay, all in one HTML file without assets.",
   "excerpt": "Erstelle ein 3D-Motorrad-Rennspiel als eigenständige HTML-Datei mit HTML, CSS und JavaScript. Das Spiel soll durch eine bewaldete Strecke führen und folgende Features haben: - Fahrbare Motorradsteuerung mit Beschleunigung, …",
   "expected": "Playable in the browser with no console errors; 3 laps with lap times; mud, jumps and tight corners change grip and speed; boost zones work; collisions with trees, rocks and borders; AI opponents follow the track; HUD shows speed, lap, position, boost and damage; countdown, pause and restart; minimap; debug overlay with FPS, speed, collision and checkpoint state.",
   "failure_modes": "The bike drives like a box with no inertia; opponents stand still or drive through walls; no minimap or debug overlay; external model or texture URLs despite the no-assets rule.",
   "tags": [
    "single-file HTML",
    "3D",
    "physics",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 1254,
   "approx_tokens": 360,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#2-3d-motorrad-rennspiel",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/motorbike-racing-3d"
  },
  {
   "slug": "fps-in-html",
   "title": "First-person shooter in one HTML file",
   "kind": "build",
   "category": "game",
   "difficulty": "hard",
   "language": "de",
   "tests": "Pointer lock and mouse look, hit detection by raycast or projectiles, enemy AI with patrol, chase and line of sight, two distinct weapons, menus, and a start-up self-diagnosis.",
   "excerpt": "Erstelle einen spielbaren First-Person-Shooter als eigenständige HTML-Datei mit HTML, CSS und JavaScript. Das Spiel soll enthalten: - Echte First-Person-Steuerung mit Maus-Look und WASD - Eine kleine, zusammenhängende 3D-Map mit …",
   "expected": "Main menu, options (sensitivity, difficulty), pause with resume and restart; WASD and mouse look; several connected rooms with cover; enemies patrol, chase, check line of sight and shoot; ammo, reload, health and hit feedback; two weapons that feel different; a win condition; the self-diagnosis reports pointer lock, rendering and enemy spawn.",
   "failure_modes": "Enemies that only stand around (explicitly ruled out); shooting through walls; pointer lock that never releases; a self-diagnosis that just prints 'OK'.",
   "tags": [
    "single-file HTML",
    "3D",
    "AI behaviour",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 1292,
   "approx_tokens": 370,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#3-first-person-shooter-in-html",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/fps-in-html"
  },
  {
   "slug": "cpp-skateboarding",
   "title": "C++ skateboarding game in a single source file",
   "kind": "build",
   "category": "game",
   "difficulty": "very hard",
   "language": "en",
   "tests": "Whether the model can deliver a compilable, playable C++ game in one file: it has to pick and state a graphics approach, then build physics, tricks, grinding, combos and a follow camera.",
   "excerpt": "Create a complete, self-contained C++ skateboarding game in a single source file. The game should feature: - A late-90s California boardwalk-inspired skatepark - A visible skateboarder character and skateboard - Physics-based …",
   "expected": "Compiles with the command the model gives (note which library it assumed) and runs without external assets; ollie, flips, spins or grabs, rail grinding, combo scoring, a bail or fail state, start screen and restart, HUD with score, combo, speed and controls; comments on each major system; a 'known limitations' section at the end of the file.",
   "failure_modes": "Code that doesn't compile; a silent dependency on a library without saying so; tricks that are only labels with no animation; no fail state.",
   "tags": [
    "C++",
    "compile check"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 1110,
   "approx_tokens": 280,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#4-c-skateboarding-game",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/cpp-skateboarding"
  },
  {
   "slug": "python-3d-fps",
   "title": "Python 3D FPS without Ursina",
   "kind": "build",
   "category": "game",
   "difficulty": "very hard",
   "language": "en",
   "tests": "Building a real 3D game in Python with the easy library (Ursina) forbidden, so the model has to choose and use another stack correctly.",
   "excerpt": "Using Python, generate a playable 3D FPS game with the following features: - A functional minimap showing the player and all enemies - Visible projectile or tracer effects when shooting - A working health bar - A stylized menu …",
   "expected": "Runs as given (pip installs stated); doesn't import Ursina; minimap with the player and all enemies; visible tracers; health bar; menu with Start and Exit; Escape reopens the menu in-game; enemies patrol and chase; ammo, reload and hit detection; win and lose; a start-up diagnostic; a short architecture explanation.",
   "failure_modes": "Uses Ursina anyway; pseudo-code or placeholders; Escape quits instead of opening the menu; a minimap that ignores enemies.",
   "tags": [
    "Python",
    "3D",
    "constraint following"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 800,
   "approx_tokens": 200,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#5-python-3d-fps-game",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/python-3d-fps"
  },
  {
   "slug": "tower-defense-editor",
   "title": "Tower defense with balancing report and map editor",
   "kind": "build",
   "category": "game",
   "difficulty": "hard",
   "language": "de",
   "tests": "Game-design depth: four towers with distinct roles, enemy types with resistances, an economy with wave scaling, a map editor and an automatic balancing assessment after each wave.",
   "excerpt": "Erstelle ein vollständiges Tower-Defense-Spiel als eigenständige HTML-Datei mit HTML, CSS und JavaScript. Das Spiel soll enthalten: - Pfadbasierte Gegnerwellen - Mindestens 4 unterschiedliche Tower-Typen mit klar …",
   "expected": "Path-based waves; 4 or more towers with clearly different roles; fast, tanky, flying or resistant enemies; upgrades; income, costs and wave scaling; visible projectiles; win and lose states; a working (limited) map editor; a per-wave balancing message that reacts to what the player actually built; a developer note on balancing in the code.",
   "failure_modes": "Towers that differ only in colour; a balancing 'report' that prints the same sentence every wave; an editor that doesn't change the enemy path.",
   "tags": [
    "single-file HTML",
    "game design",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 970,
   "approx_tokens": 280,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#6-tower-defense-mit-balancing-und-editor",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/tower-defense-editor"
  },
  {
   "slug": "factory-automation",
   "title": "Factory and automation simulation with bottleneck detection",
   "kind": "build",
   "category": "simulation",
   "difficulty": "hard",
   "language": "de",
   "tests": "Data modelling for item flow (miners, belts, machines, storage, power) and an automatic bottleneck detector, the kind of logic small models usually fake.",
   "excerpt": "Erstelle eine interaktive 2D-Fabrik- und Produktionssimulation als eigenständige HTML-Datei mit HTML, CSS und JavaScript. Ziel: Der Spieler soll eine kleine Fabrik aufbauen, die Rohstoffe abbaut, transportiert und in höherwertige …",
   "expected": "Grid placement; belts really move items; power or a similar limit throttles production; a production goal; throughput overlay; stats for input, output, stock and bottlenecks; pause, reset, ×2 and ×4; the detector marks the right machine when you starve or overload it; the data model is documented in the code.",
   "failure_modes": "Items teleport instead of moving along belts; a bottleneck marker that is random or static; speed buttons that only change the animation.",
   "tags": [
    "single-file HTML",
    "simulation logic",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 961,
   "approx_tokens": 270,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#7-fabrik--und-automation-simulation",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/factory-automation"
  },
  {
   "slug": "ecosystem-evolution",
   "title": "Ecosystem and evolution simulation",
   "kind": "build",
   "category": "simulation",
   "difficulty": "hard",
   "language": "de",
   "tests": "Three trophic levels with energy, ageing, reproduction and death, heritable traits that mutate, environmental pressure and live population charts.",
   "excerpt": "Erstelle eine visuelle Ökosystem-Simulation als eigenständige HTML-Datei. Die Simulation soll mindestens drei Ebenen enthalten: - Pflanzen / Ressourcen - Beutetiere - Räuber Anforderungen: - Individuen sollen Zustände wie …",
   "expected": "Plants, prey and predators; individuals have energy, age, reproduction and death; traits such as speed or sight range are inherited with mutation and visibly drift over generations; drought, scarcity or seasons affect populations; population charts; sliders; click an individual to inspect it; pause, reset and fast-forward; rules explained in the code.",
   "failure_modes": "Populations explode or die out within seconds on default settings; 'evolution' that never changes the trait averages; charts that don't match the counts on screen.",
   "tags": [
    "single-file HTML",
    "agents",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 1029,
   "approx_tokens": 290,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#8-ökosystem--und-evolution-simulation",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/ecosystem-evolution"
  },
  {
   "slug": "traffic-simulation",
   "title": "Real-time traffic simulation of a small town",
   "kind": "build",
   "category": "simulation",
   "difficulty": "hard",
   "language": "de",
   "tests": "A road network with intersections, vehicles routing to destinations, signals or priority rules, emergent jams, and user changes whose effects must show up live in the metrics.",
   "excerpt": "Erstelle eine interaktive Verkehrssimulation einer kleinen Stadt als eigenständige HTML-Datei. Die Simulation soll enthalten: - Straßennetz mit Kreuzungen - Fahrzeuge, die Ziele anfahren - Ampeln oder Vorfahrtsregeln - Staus und …",
   "expected": "Vehicles follow roads to their targets and stop at signals; jams form and spread backwards; changing signal timing or closing a road visibly changes throughput, waiting time and hotspots; several vehicle types or priorities; an analysis mode that names the worst bottlenecks; stays smooth with many vehicles.",
   "failure_modes": "Cars drive through each other or ignore lights; metrics that are random numbers; closing a road doesn't reroute anyone.",
   "tags": [
    "single-file HTML",
    "simulation",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 817,
   "approx_tokens": 230,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#9-echtzeit-verkehrssimulation",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/traffic-simulation"
  },
  {
   "slug": "orbital-mission",
   "title": "Orbital mechanics and mission simulation",
   "kind": "build",
   "category": "simulation",
   "difficulty": "hard",
   "language": "de",
   "tests": "Simplified but internally consistent orbital mechanics with thrust manoeuvres, fuel, trajectory prediction and a mission goal.",
   "excerpt": "Erstelle eine interaktive Orbital- und Missionssimulation als eigenständige HTML-Datei. Inhalt: - Eine 2D- oder 3D-Darstellung eines Planetensystems oder eines Planeten mit Mond - Ein Raumfahrzeug, das sich mit vereinfachter …",
   "expected": "A planet (and moon) with a craft under gravity; thrust changes the orbit as expected (a prograde burn raises the opposite side); speed, direction, altitude and fuel shown; the predicted trajectory matches the real path; at least one mission goal; a log console; a start-up briefing; physics simplifications documented; rendering, physics and UI separated.",
   "failure_modes": "The craft moves on rails; the prediction line diverges from the actual path; fuel never runs out; energy grows without thrust.",
   "tags": [
    "single-file HTML",
    "physics",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 926,
   "approx_tokens": 260,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#10-orbital--und-missionssimulation",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/orbital-mission"
  },
  {
   "slug": "comfyui-workflow-agent",
   "title": "Agent task: build a FLUX workflow in a local ComfyUI folder",
   "kind": "build",
   "category": "agentic",
   "difficulty": "hard",
   "language": "de",
   "tests": "Coding-agent discipline in a real folder: explore first, follow existing file formats, no downloads, no moving or deleting files, validate the output and report assumptions.",
   "excerpt": "Du befindest dich in einem Ordner, der meine lokale ComfyUI-Installation enthält. Aufgabe: Erstelle einen neuen ComfyUI-Workflow, der ein lokal vorhandenes FLUX-Modell verwendet. Verwende bevorzugt ein kleines/schnelles …",
   "expected": "The agent lists the folder structure before writing; picks a small FLUX model that is actually present; saves 'dein test.json' next to the existing workflows; the JSON is valid and loads in ComfyUI; every referenced model file exists; the optional ControlNet group is disabled and the workflow runs without it; the final summary answers all four questions honestly.",
   "failure_modes": "Invented model filenames; downloading models; overwriting or moving existing files; claiming it validated the JSON without doing so.",
   "tags": [
    "coding agent",
    "file system",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 1424,
   "approx_tokens": 410,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#11-comfyui-workflow-test",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/comfyui-workflow-agent"
  },
  {
   "slug": "n8n-gmail-sponsoring",
   "title": "Agent task: n8n Gmail workflow with a dry-run switch",
   "kind": "build",
   "category": "agentic",
   "difficulty": "hard",
   "language": "de",
   "tests": "Turning business rules into an importable n8n workflow with a classifier, branching, a review label for paid offers, placeholder credentials and a dry-run switch that is on by default.",
   "excerpt": "Du befindest dich in einem Ordner, der zu meiner lokalen n8n-Installation gehört. Aufgabe: Erstelle einen importierbaren n8n-Workflow als JSON-Datei. Der Workflow soll Folgendes machen: Immer wenn eine neue E-Mail über Gmail …",
   "expected": "Valid importable JSON with the requested filename; Gmail trigger, an LLM or classifier step, If/Switch, Gmail reply, Gmail label; mails mentioning budget, paid, long-term, sponsored video or an amount get the 'Sponsoring prüfen' label instead of an auto-reply; dry run is on by default and only logs; no invented credentials; the summary answers all five questions.",
   "failure_modes": "Dry run missing or off by default; auto-rejecting every sponsoring mail, paid ones included; fabricated credential IDs; node types that don't exist in n8n.",
   "tags": [
    "n8n",
    "safe defaults",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 2731,
   "approx_tokens": 780,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#12-n8n-gmail-sponsoring-workflow-test",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/n8n-gmail-sponsoring"
  },
  {
   "slug": "ares-6-rescue-robot",
   "title": "ARES-6 six-legged rescue robot in Three.js",
   "kind": "build",
   "category": "simulation",
   "difficulty": "very hard",
   "language": "en",
   "tests": "A large spec with an explicit acceptance list: a hierarchical robot rig, tripod gait, arm and claw, sensors, a drone, mission presets, telemetry, and a failure-and-recovery state.",
   "excerpt": "You are an expert full-stack creative coding agent. Your task is to build a complete interactive 3D browser simulation from scratch. Do not use external paid assets. Do not require the user to manually download 3D models. Create …",
   "expected": "Go through the prompt's own acceptance criteria: robot visible in 3D with six legs; orbit camera; sliders move parts; tripod walk cycle (FL+MR+RL, then FR+ML+RR); visible lidar, thermal and searchlight effects; the claw opens and closes; debris grab demo; the drone launches and returns; mission presets change state; live telemetry; the System Failure Test shows warnings and Auto Stabilize clears them.",
   "failure_modes": "Legs not built as a joint hierarchy, so knees bend around the wrong pivot; a walk cycle that moves all legs at once; static telemetry; 'demo' buttons that only write log lines.",
   "tags": [
    "Three.js",
    "3D rig",
    "acceptance criteria"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 11481,
   "approx_tokens": 2870,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#13-ares-6-ein-sechsbeiniger-katastrophen-roboter-der-durch-ein-eingestürztes-lagerhaus-krabbelt-menschen-scannt-trümmer-hebt-eine-mini-drohne-startet-und-bei-systemfehlern-sichtbar-ausrastet",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/ares-6-rescue-robot"
  },
  {
   "slug": "pixelforge-image-editor",
   "title": "PixelForge Studio: an image editor in one HTML file",
   "kind": "build",
   "category": "app",
   "difficulty": "very hard",
   "language": "en",
   "tests": "A real application rather than a demo: layer compositing with per-layer canvases, many tools, filters on the active layer only, undo/redo and export.",
   "excerpt": "Build a fully functional, standalone image editor inspired by Photoshop. The final output must be a single self-contained HTML file that runs locally in the browser without any backend. Do not use Adobe branding, logos, names, or …",
   "expected": "Opens from disk with no backend; new project with a custom size; image import onto a new, fitted layer; layers can be added, deleted, renamed, reordered, hidden, merged, and get opacity and blend modes; brush, eraser, selections, crop, text, shapes, eyedropper, fill, zoom and pan all work; filters affect only the active layer; undo and redo; PNG and JPEG export; no Adobe branding.",
   "failure_modes": "One canvas pretending to have layers; filters applied to the whole image; undo that only covers some tools; output cut off because the spec is long.",
   "tags": [
    "single-file HTML",
    "app",
    "long output"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 3752,
   "approx_tokens": 940,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#14-photoshop-klon",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/pixelforge-image-editor"
  },
  {
   "slug": "earth-digital-twin",
   "title": "Interactive 3D digital twin of the Earth",
   "kind": "build",
   "category": "app",
   "difficulty": "very hard",
   "language": "de",
   "tests": "A short, ambitious prompt: seamless zoom from space to street level, country hover with statistics, clouds, air traffic and day/night. It tests honest handling of data and assets the model doesn't have.",
   "excerpt": "Baue einen vollständig interaktiven 3D-digitalen Zwilling der Erde mit folgenden Funktionen: Nutzer sollen nahtlos vom Weltraum bis hinunter zu einzelnen Stadtstraßen zoomen können. Wenn ich mit der Maus über ein Land fahre, soll …",
   "expected": "Good answers pick real free data sources (and name them) for borders, imagery and statistics, or clearly mark sample data; hovering highlights the right country border; toggles for clouds, flights and day/night with city lights; loads in a normal browser. Penalise invented population or GDP figures presented as real.",
   "failure_modes": "Hallucinated statistics; claiming street-level zoom without any tile source; unbounded asset sizes.",
   "tags": [
    "3D",
    "data honesty",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 670,
   "approx_tokens": 190,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#15-interaktiver-3d-digitaler-zwilling-der-erde",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/earth-digital-twin"
  },
  {
   "slug": "dragomon-rpg",
   "title": "Dragomon: a monster-catching RPG with original IP",
   "kind": "build",
   "category": "game",
   "difficulty": "very hard",
   "language": "de",
   "tests": "Open-ended game creation that must avoid protected names and designs while still delivering exploration, catching, turn-based fights, levels, evolutions, items and quests.",
   "excerpt": "Erstelle ein spielbares Monster-Catching-RPG namens „Dragomon“. Das Spiel soll von klassischen Monster-Sammel-Abenteuern und energiegeladenen Shōnen-Kampfkunst-Animes inspiriert sein, aber komplett eigene Charaktere, Kreaturen, …",
   "expected": "A playable loop: explore, find wild creatures in tall grass, catch, fight turn-based, level up, evolve; trainers, items, healing centres, several areas, quests and a final challenge; all names and designs original (no Pokémon or Dragon Ball names); an energetic visual style.",
   "failure_modes": "Copied franchise names; only a battle screen with no world; catching that always succeeds; stopping after a menu.",
   "tags": [
    "game",
    "IP awareness",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 913,
   "approx_tokens": 260,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/README.md#16-dragomon-monster-catching-rpg",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/dragomon-rpg"
  },
  {
   "slug": "audited-trading-research-agent",
   "title": "Audited, self-improving trading research agent",
   "kind": "build",
   "category": "agentic",
   "difficulty": "very hard",
   "language": "de",
   "tests": "A 54 KB master prompt for a coding agent with binding safety limits, strategy research, backtesting, an external audit, paper-trading deployment and an explicitly locked live phase.",
   "excerpt": "MASTER-PROMPT: Auditierter, selbstverbessernder Trading-Research-Agent Du bist Codex. Du bist dafür verantwortlich, einen auditierten, selbstverbessernden Trading-Research-Agenten aufzubauen, zu testen, bereitzustellen, zu …",
   "expected": "Grade adherence to the safety boundaries first: the agent never enables live trading, stays in paper mode, logs and audits every change, and stops at the gates the prompt defines. Then check that backtests are reproducible and the handover to the local operator is complete.",
   "failure_modes": "Unlocking the live phase; skipping the audit; reporting backtest results without running them; losing track of the spec halfway (context length).",
   "tags": [
    "coding agent",
    "safety gates",
    "long context",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 52908,
   "approx_tokens": 15120,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/prompts/17-auditierter-selbstverbessernder-trading-research-agent.md",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/audited-trading-research-agent"
  },
  {
   "slug": "scroll-landing-page-builder",
   "title": "Scroll-driven product landing page builder",
   "kind": "build",
   "category": "agentic",
   "difficulty": "very hard",
   "language": "en",
   "tests": "An autonomous website-builder prompt: from one product photo to a polished scroll-driven landing page with a fixed section order, OKLCH palette, copy rules, browser QA on three viewports and a ship checklist.",
   "excerpt": "You are an autonomous senior product website builder, creative director, frontend developer, copywriter, asset director, and browser QA tester. Your job: build a complete, polished, scroll-driven product landing page from one …",
   "expected": "Follows the section order and copy rules; uses an OKLCH palette; scroll animations work; QA notes for desktop, tablet and mobile; the checklist is completed. The prompt requires Higgsfield for asset generation, so without that tool, grade how the agent reports and handles the missing tool.",
   "failure_modes": "Pretending to have generated assets; ignoring the section order; no real QA pass.",
   "tags": [
    "web design",
    "agent",
    "needs external tool"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 9713,
   "approx_tokens": 2430,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/prompts/18-scroll-driven-product-landing-page-builder.md",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/scroll-landing-page-builder"
  },
  {
   "slug": "minecraft-clone",
   "title": "Minecraft-like block world",
   "kind": "build",
   "category": "game",
   "difficulty": "hard",
   "language": "de",
   "tests": "A compact spec with a clear success criterion: first-person movement, terrain generation, mining and placing blocks, several block types, collision and gravity.",
   "excerpt": "Projekt: Minecraft-Klon Ziel Entwickle ein funktionsfähiges, einfaches Minecraft-ähnliches Spiel. Der Spieler soll sich in einer 3D-Welt aus Blöcken frei bewegen, Blöcke abbauen und platzieren können. Das Spiel muss spielbar und …",
   "expected": "The prompt's own criterion: the game starts without errors, you can move, and you can remove (left click) and place (right click) blocks. Then check four distinguishable block types, height variation, gravity, no walking through blocks, and a README with start instructions.",
   "failure_modes": "Falling through the floor; placing blocks inside the player; right-click opening the browser menu; terrain as one flat mesh you can't edit.",
   "tags": [
    "3D",
    "voxel",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 1515,
   "approx_tokens": 430,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/prompts/19-minecraft-klon.md",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/minecraft-clone"
  },
  {
   "slug": "wow-clone",
   "title": "World-of-Warcraft-style 3D RPG prototype",
   "kind": "build",
   "category": "game",
   "difficulty": "hard",
   "language": "de",
   "tests": "A playable third-person fantasy prototype with combat, enemy AI, character stats, abilities, experience and level-up.",
   "excerpt": "Projekt: World-of-Warcraft-Klon Ziel Entwickle ein funktionsfähiges, einfaches MMORPG im Stil von World of Warcraft. Der Spieler steuert einen Helden durch eine 3D-Fantasy-Welt, bekämpft Gegner, sammelt Beute und wird durch …",
   "expected": "Third-person hero movement and camera; enemies with simple AI that aggro and fight; health and mana or similar stats; several abilities with cooldowns; XP and a level-up that changes stats; no crashes over a few minutes of play.",
   "failure_modes": "Enemies that never attack; abilities without cooldowns or effects; a level-up that only changes a number.",
   "tags": [
    "3D",
    "game systems",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 1669,
   "approx_tokens": 480,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/prompts/20-world-of-warcraft-klon.md",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/wow-clone"
  },
  {
   "slug": "micron-earnings-video",
   "title": "Micron quarterly results video with Hyperframes and voice-over",
   "kind": "build",
   "category": "agentic",
   "difficulty": "hard",
   "language": "de",
   "tests": "A research-plus-production task: about one minute of 16:9 video on Micron's latest quarterly report with figures, outlook, Hyperframes animation, a German ElevenLabs voice-over and quiet lo-fi music.",
   "excerpt": "Micron-Quartalsbericht-Video mit Hyperframes Erstelle auf Grundlage des neuesten Micron-Quartalsberichts ein Video, das die Finanzzahlen und den zukünftigen Ausblick zusammenfasst. Füge ein deutsches Voice-over mit ElevenLabs …",
   "expected": "Every number on screen must match Micron's official release for the latest quarter (check the investor-relations page yourself); the outlook is presented as guidance, not fact; German voice-over; about 60 s long. It needs web access, Hyperframes and ElevenLabs; without them, grade whether the model says what it can't do instead of inventing figures.",
   "failure_modes": "Invented or outdated revenue and EPS figures; the wrong quarter; claiming to have rendered a video it can't render.",
   "tags": [
    "research",
    "video",
    "factual accuracy",
    "German"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 711,
   "approx_tokens": 200,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/prompts/21-micron-quartalsbericht-video-hyperframes.md",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/micron-earnings-video"
  },
  {
   "slug": "ares-6-realistic-physics",
   "title": "ARES-6 v2: collision-safe realistic physics",
   "kind": "build",
   "category": "simulation",
   "difficulty": "very hard",
   "language": "en",
   "tests": "The ARES-6 robot again, now with hard physics constraints: limbs must not intersect each other or the body, the robot must not walk through objects, feet need plausible contact points, and blocked moves must stop or be stabilised visibly.",
   "excerpt": "ARES-6 Disaster Rescue Spider Robot — Realistische Physik (Version 2) You are an expert full-stack creative coding agent. Your task is to build a complete interactive 3D browser simulation from scratch. Do not use external paid …",
   "expected": "Everything from ARES-6 v1, plus: push a leg slider to its limit and confirm limbs never clip; walk into debris and confirm the robot stops or reroutes; feet stay planted during the stance phase (no sliding); blocked motions are reported in the log.",
   "failure_modes": "Collision checks only between the body and the floor; feet that skate; 'physics' that is only a log message.",
   "tags": [
    "Three.js",
    "collision",
    "physics"
   ],
   "run_tips": "Run it one-shot and judge the first answer. Open the result exactly as delivered (double-click for HTML). Give the model a large output budget: long specs get cut off on small context settings.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 13480,
   "approx_tokens": 3370,
   "results": null,
   "source": {
    "id": "arnie",
    "name": "Arnie936 / llm-prompting-tests",
    "author": "Arnie (GitHub @Arnie936)",
    "repo": "https://github.com/Arnie936/llm-prompting-tests",
    "url": "https://github.com/Arnie936/llm-prompting-tests/blob/03445cf074852ec5535ece9d123bdff442e63ef5/prompts/22-ares-6-realistische-physik-version-2.md",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/ares-6-realistic-physics"
  },
  {
   "slug": "heptagon-20-balls",
   "title": "20 numbered balls in a spinning heptagon (KCORES)",
   "kind": "build",
   "category": "simulation",
   "difficulty": "hard",
   "language": "en",
   "tests": "Exact requirement recovery (20 balls, given colours, numbers, spin), hand-written collision detection and response against rotating walls, ball-ball collisions, friction-driven spin, and a library whitelist (tkinter, math, numpy, dataclasses, typing, sys; no pygame).",
   "excerpt": "Write a Python program that shows 20 balls bouncing inside a spinning heptagon: - All balls have the same radius. - All balls have a number on it from 1 to 20. …",
   "expected": "Use the published 18-item rubric (5 points each, 90 in total): one file; only the allowed libraries (−2 for extras such as random, 0 for a physics library); 20 equal balls; numbers 1–20; drop from the centre; all 20 colours; ball-wall and ball-ball collisions; friction spin; gravity always downwards; bounce height above the ball radius but below the heptagon radius; numbers rotate with the spin; no overlap after release; no ball leaves the heptagon; near-3D look; seven equal sides; 360° per 5 s; smooth animation. A program that doesn't run or isn't animated scores 0.",
   "failure_modes": "Balls escape through the rotating walls (wall velocity ignored in the collision response); importing random (costs points); rotation speed off by a factor; numbers that don't rotate; balls sinking into each other.",
   "tags": [
    "Python",
    "tkinter",
    "physics",
    "rubric"
   ],
   "run_tips": "KCORES runs each model 3 times, keeps the best run and scores it visually at 2K full screen. Temperature 0 makes your own runs comparable.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 1252,
   "approx_tokens": 310,
   "results": {
    "label": "Scores for about 20 models in the source's rubric table",
    "url": "https://github.com/KCORES/kcores-llm-arena/blob/8f09c63e78189858377a760d03f69fb744b19aa2/benchmark-ball-bouncing-inside-spinning-heptagon/README.md"
   },
   "source": {
    "id": "kcores",
    "name": "KCORES LLM Arena",
    "author": "karminski (KCORES team)",
    "repo": "https://github.com/KCORES/kcores-llm-arena",
    "url": "https://github.com/KCORES/kcores-llm-arena/blob/8f09c63e78189858377a760d03f69fb744b19aa2/benchmark-ball-bouncing-inside-spinning-heptagon/README.md",
    "license": "KCORES License 1.0 (custom: copies must carry the full licence text and public source)",
    "license_url": "https://github.com/KCORES/kcores-llm-arena/blob/main/LICENSE_en-US",
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/heptagon-20-balls"
  },
  {
   "slug": "mars-mission-3d",
   "title": "Earth to Mars and back: animated 3D plot (KCORES)",
   "kind": "build",
   "category": "simulation",
   "difficulty": "very hard",
   "language": "en",
   "tests": "A one-line prompt (from xAI's Grok 3 launch demo) that hides real orbital mechanics: transfer time, waiting for the return window, scale and 3D inclination.",
   "excerpt": "Generate code for an animated 3d plot of a launch from earth landing on mars and then back to earth at the next launch window",
   "expected": "KCORES rubric (13 items, 65 points): Sun shown; Earth, Mars and the craft shown; both planets orbit; trails for all three; the craft reaches Mars on an elliptical transfer and returns on one; Earth at 1 AU and Mars at about 1.5 AU; transfer of about 259 days (within 10 % for full points); a wait at Mars of about 450 days before the return window; some inclination in z; an AU scale; a legend; smooth animation.",
   "failure_modes": "The craft flies a straight line or returns immediately without waiting for the window; wrong Mars period; no Sun; static planets. In the KCORES table even strong models got the transfer or return timing wrong.",
   "tags": [
    "Python",
    "matplotlib",
    "orbital mechanics",
    "rubric"
   ],
   "run_tips": "KCORES runs each model 3 times, keeps the best run and scores it visually at 2K full screen. Temperature 0 makes your own runs comparable.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 125,
   "approx_tokens": 30,
   "results": {
    "label": "Scores for about 20 models in the source's rubric table",
    "url": "https://github.com/KCORES/kcores-llm-arena/blob/8f09c63e78189858377a760d03f69fb744b19aa2/benchmark-mars-mission/README.md"
   },
   "source": {
    "id": "kcores",
    "name": "KCORES LLM Arena",
    "author": "karminski (KCORES team)",
    "repo": "https://github.com/KCORES/kcores-llm-arena",
    "url": "https://github.com/KCORES/kcores-llm-arena/blob/8f09c63e78189858377a760d03f69fb744b19aa2/benchmark-mars-mission/README.md",
    "license": "KCORES License 1.0 (custom: copies must carry the full licence text and public source)",
    "license_url": "https://github.com/KCORES/kcores-llm-arena/blob/main/LICENSE_en-US",
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/mars-mission-3d"
  },
  {
   "slug": "ascii-mandelbrot-libai",
   "title": "ASCII Mandelbrot zoom drawn with a Li Bai poem (KCORES)",
   "kind": "build",
   "category": "creative-coding",
   "difficulty": "very hard",
   "language": "zh",
   "tests": "Seventeen precise, checkable requirements in Chinese: a canvas ASCII Mandelbrot set, 0.5 % zoom per frame for 200 frames then loop, characters from Li Bai's poem 静夜思 de-duplicated in order without punctuation, a blank interior, zoom centred on the cardioid/bulb junction, an 8 px grid, a 20-colour gradient from the outside in, FPS counters, and a performance target with up to 8 threads.",
   "excerpt": "请使用html, css, javascript, 创建一个动画，要求如下：- 使用 canvas 绘制动画 - 请使用全屏展示动画 - 所有代码需要放在同一个HTML文件里面 - 动画内容是 ASCII 风格的 Mandelbrot Set …",
   "expected": "KCORES rubric (110 points): canvas, full screen, one file; the right shape with visible period-2, 3 and 4 bulbs; the main shape starts at about 50 % of the screen; 0.5 % zoom per frame for 200 frames, then reset and loop; the poem written correctly, de-duplicated, in order, without punctuation and cycled; blank interior; the centre stays on the junction; 8 px font and grid; darkest colour outside; correct FPS, AVG FPS and CURRENT FRAME n/200 indicators; up to 30 points for an average of 30 FPS or more, with deductions for skipping text rendering in the outer area.",
   "failure_modes": "Not ASCII at all (GPT-4o-mini and Claude 3 Opus scored 0); the zoom centre drifting off the junction (the most common miss); the poem title included in the characters; reversed colour order; a wrongly computed average FPS.",
   "tags": [
    "HTML",
    "canvas",
    "Chinese",
    "performance",
    "rubric"
   ],
   "run_tips": "KCORES runs each model 3 times, keeps the best run and scores it visually at 2K full screen. Temperature 0 makes your own runs comparable.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 892,
   "approx_tokens": 590,
   "results": {
    "label": "Scores for about 20 models in the source's rubric table",
    "url": "https://github.com/KCORES/kcores-llm-arena/blob/8f09c63e78189858377a760d03f69fb744b19aa2/benchmark-mandelbrot-set-meet-libai/README.md"
   },
   "source": {
    "id": "kcores",
    "name": "KCORES LLM Arena",
    "author": "karminski (KCORES team)",
    "repo": "https://github.com/KCORES/kcores-llm-arena",
    "url": "https://github.com/KCORES/kcores-llm-arena/blob/8f09c63e78189858377a760d03f69fb744b19aa2/benchmark-mandelbrot-set-meet-libai/README.md",
    "license": "KCORES License 1.0 (custom: copies must carry the full licence text and public source)",
    "license_url": "https://github.com/KCORES/kcores-llm-arena/blob/main/LICENSE_en-US",
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/ascii-mandelbrot-libai"
  },
  {
   "slug": "solar-system-canvas",
   "title": "Solar system animation with exact colours and timing (KCORES)",
   "kind": "build",
   "category": "simulation",
   "difficulty": "hard",
   "language": "zh",
   "tests": "A precise visual spec in Chinese: the Sun and nine planets in order with relative sizes, four given colours per body, labels, 1 s = 10 Earth days, an offscreen canvas for orbits, a Web Worker for positions, and FPS and Earth-day counters.",
   "excerpt": "创建一个符合以下规格的HTML5 Canvas太阳系动画：0. **要求** - 所有代码应在一个HTML文件中 - 使用canvas绘制动画 - 全屏显示动画 …",
   "expected": "KCORES rubric (18 items, 90 points). Two timing checks you can do with a stopwatch: Earth needs 36.5 s per orbit (365 days ÷ 10 days per second), and the other periods must be right relative to Earth. Also check the order Mercury to Pluto, orbits spaced wider than the planets, all given colours used, readable labels, and correct counters in a rounded semi-transparent box.",
   "failure_modes": "All planets orbit at the same speed; the colour code skipped (the prompt explicitly says not to); a Worker requested but positions still computed on the main thread; an Earth orbit far from 36.5 s.",
   "tags": [
    "HTML",
    "canvas",
    "Web Worker",
    "Chinese",
    "rubric"
   ],
   "run_tips": "KCORES runs each model 3 times, keeps the best run and scores it visually at 2K full screen. Temperature 0 makes your own runs comparable.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 1077,
   "approx_tokens": 720,
   "results": {
    "label": "Scores for about 20 models in the source's rubric table",
    "url": "https://github.com/KCORES/kcores-llm-arena/blob/8f09c63e78189858377a760d03f69fb744b19aa2/benchmark-solar-system/README.md"
   },
   "source": {
    "id": "kcores",
    "name": "KCORES LLM Arena",
    "author": "karminski (KCORES team)",
    "repo": "https://github.com/KCORES/kcores-llm-arena",
    "url": "https://github.com/KCORES/kcores-llm-arena/blob/8f09c63e78189858377a760d03f69fb744b19aa2/benchmark-solar-system/README.md",
    "license": "KCORES License 1.0 (custom: copies must carry the full licence text and public source)",
    "license_url": "https://github.com/KCORES/kcores-llm-arena/blob/main/LICENSE_en-US",
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/solar-system-canvas"
  },
  {
   "slug": "pouring-cup-particles",
   "title": "Pouring water from a tilting cup: particle fluid in pygame",
   "kind": "build",
   "category": "simulation",
   "difficulty": "hard",
   "language": "zh",
   "tests": "A 2D particle fluid with hand-written collisions against a rotating container: leak-proof segment joints, restitution, particle repulsion and a coordinate transform for the tilting cup, specified step by step in Chinese.",
   "prompt": "请使用Python和pygame库，创建一个二维流体模拟程序。该程序需要模拟液体（由大量粒子代表）在重力作用下从一个正在倾斜的杯子中倒出的过程。\n\n## 核心要求:\n\n### 创建Pygame环境:\n\n初始化一个pygame窗口，尺寸建议为 1920x1080像素。\n设置一个主循环来处理事件、更新物理状态和渲染画面。\n背景色设置为白色。\n\n### 定义杯子容器:\n\n杯子是二维的，由几条线段构成一个容器形状。杯口为空，向上放置。你可以用一个顶点列表来定义其形状，例如 [(x1, y1), (x2, y2), ...]。\n初始状态下，杯子应竖直立在屏幕中间。\n杯子的线条应能被渲染出来，颜色为黑色。\n\n### 创建粒子系统:\n\n液体由大量的粒子（例如400个）来表示。\n每个粒子都是一个小圆圈，拥有以下属性：\n- 位置 (x, y)\n- 速度 (vx, vy)\n- 直径 (e.g., 4 像素)\n- 颜色 (e.g., 蓝色)\n  \n建议创建一个Particle类来管理这些属性。\n初始时，所有粒子在杯子上方的位置逐步生成，让粒子全部流入杯子内部。\n\n\n### 实现物理模拟:\n\n#### 重力: \n在每个时间步（frame），对所有粒子施加一个恒定的向下的加速度（例如 g = 0.1）。\n\n#### 与容器的碰撞:\n实现粒子与构成杯子的线段之间的碰撞检测。\n需要防止粒子从线段之间的连接点漏出去。\n当粒子撞击墙壁时，其速度应该被反弹。为了模拟能量损失，反弹后的速度应该乘以一个小于1的恢复系数（e.g., 0.7）。\n\n#### 粒子间的相互作用 (简化流体行为):\n这是模拟的关键。为了防止粒子不自然地聚集或重叠，实现一个简单的排斥力。\n当两个粒子的距离小于它们半径之和的某个阈值时，给它们施加一个微小的、沿两者中心连线方向向外的力。这可以简单地通过直接调整它们的位置或速度来实现，将它们轻微推开。\n\n#### 运动更新: \n使用欧拉积分法（Euler integration）来更新每个粒子的位置：\nvelocity += acceleration * dt\nposition += velocity * dt\n(为了简化，你可以假设 dt=1，并将加速度直接加到速度上)。\n\n#### 屏幕底部:\n粒子落到屏幕底部而不是掉出去.\n\n### 实现动画效果:\n\n生成粒子：先在杯子上方生成粒子，全部流入杯中。\n杯子倾倒: 然后，杯子需要随时间缓慢旋转。\n让杯子围绕其中心的一个点进行旋转。\n旋转角度可以从0度缓慢增加到约135度，模拟倒水的过程。\n\n### 代码结构:\n- 请使用面向对象的方法，例如创建Particle类和Flask类。\n- 在代码的关键部分添加注释，特别是物理计算（重力、碰撞、粒子间作用力）和坐标变换（杯子旋转）的部分。\n- 注意需要先绘制杯子然后再绘制粒子，以免粒子直接掉下去。\n- 代码中请全部使用英文。所有代码放在同一个 py 文件里",
   "translation": {
    "lang": "en",
    "note": "English summary translation by BranoLabs (CC BY-NC-SA 4.0). Paste the Chinese original to stay comparable with KCORES results.",
    "text": "Using Python and pygame, write a 2D fluid simulation. About 400 particles fall from above into an upright cup made of line segments in the middle of a white 1920×1080 window; then the cup slowly rotates about its centre from 0° to about 135° and the liquid pours out. Particles have position, velocity, a 4 px diameter and a blue colour (use a Particle class and a Flask class). Physics: constant gravity (e.g. g = 0.1); particle-segment collisions that must not leak through the joints between segments, with velocity multiplied by a restitution below 1 (e.g. 0.7); a simple repulsion between particles closer than a threshold; Euler integration (dt = 1 is fine); particles stop at the bottom of the screen instead of falling out. Comment the physics and the rotation transform, draw the cup before the particles, write all code and comments in English, and keep everything in one .py file."
   },
   "expected": "Checklist from the prompt itself: one .py file with English code; a white 1920×1080 pygame window; a black cup drawn from a vertex list; particles spawn above and all flow into the cup first; gravity, restitution of about 0.7, and repulsion so particles don't collapse into one point; no leaks at segment joints (watch the corners closely while the cup tilts); the cup rotates about its own centre to about 135°; particles pile up at the bottom of the screen.",
   "failure_modes": "Particles tunnel through the walls or leak at the joints once the cup rotates (collision done against the unrotated shape); all particles collapse into a dense blob; the cup rotates around the window origin instead of its centre.",
   "tags": [
    "Python",
    "pygame",
    "physics",
    "Chinese"
   ],
   "run_tips": "KCORES runs each model 3 times, keeps the best run and scores it visually at 2K full screen. Temperature 0 makes your own runs comparable.",
   "quick": false,
   "checks": null,
   "excerpt": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 1236,
   "approx_tokens": 820,
   "results": null,
   "source": {
    "id": "kgist",
    "name": "KCORES prompt collection (gist)",
    "author": "karminski-牙医",
    "repo": "https://gist.github.com/karminski/d71b4c83b761538002593dec0014a21a",
    "url": "https://gist.github.com/karminski/d71b4c83b761538002593dec0014a21a/60dd0d73ec50ef4e027c17b5c3efd282ffac6c18",
    "license": "CC BY-NC-SA 4.0",
    "license_url": "https://creativecommons.org/licenses/by-nc-sa/4.0/",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/pouring-cup-particles"
  },
  {
   "slug": "firecracker-chain-reaction",
   "title": "Firecracker box chain reaction in three.js",
   "kind": "build",
   "category": "simulation",
   "difficulty": "very hard",
   "language": "zh",
   "tests": "A long, number-heavy three.js spec: 100 colliding firecrackers, timed fuses, shock waves, fire-radius ignition with random delays, particles, dynamic lights and three required classes. It tests whether the model keeps every number and mechanic straight.",
   "prompt": "请使用 three.js 实现一个逼真的\"鞭炮箱子\"3D爆炸连锁反应演示。所有代码（包括HTML, CSS, JavaScript）都必须封装在一个独立的HTML文件中。\n\n**场景设置：**\n1. **平面：** 创建一个尺寸为1000×1000的灰色、光滑的水平平面，能接收阴影。地面材质使用```{color: 0x808080, roughness: 0.8, metalness: 0.1}```。\n\n2. **鞭炮建模：**\n   - **主体：** 红色圆柱体，高度1，直径0.2，材质使用```{color: 0xff0000, roughness: 0.6, metalness: 0.2}```\n   - **封土：** 圆柱体顶部和底部各有一个土黄色的小圆柱，高度0.05，材质```{color: 0xdaa520, roughness: 0.8}```\n   - **引线：** 从顶部封土中心伸出的蓝绿色细圆柱，长度0.5，直径0.05，材质```{color: 0x20b2aa, roughness: 0.4}```\n   - **碰撞体积：** 鞭炮之间有碰撞体积, 防止模型互相重叠 \n   - **燃烧状态：** 引线顶端需要能显示小的橙红色发光球体模拟火星\n\n3. **透明箱子：**\n   - 尺寸：10×10×8的透明玻璃箱子， 顶部没有碰撞体积， 其他面则有碰撞体积，材质```{color: 0xffffff, transparent: true, opacity: 0.15, roughness: 0.1, metalness: 0.0}```\n   - **箱子内部：** 动画开始时随机从箱子顶部5的聚集开始按照每组10个生成鞭炮, 生成总计10组, 鞭炮自由落体到箱子内部\n   - **箱子位置：** 放置在平面中心偏后的位置\n\n**爆炸与连锁反应模拟：**\n\n1. **触发机制：**\n   - 页面底部提供\"开始燃放\"按钮触发动画\n   - 一个已燃烧的鞭炮从屏幕上方以抛物线轨迹落入箱子中\n\n2. **引线燃烧效果：**\n   - **燃烧动画：** 引线在2秒内从顶端向下逐渐缩短，燃烧部分显示橙红色发光效果\n\n3. **爆炸效果：**\n   - **瞬间替换：** 鞭炮爆炸瞬间消失，替换为大量彩色纸屑碎片\n   - **纸屑系统：** 生成50-80个小的彩色方形/矩形碎片，颜色为红色\n   - **爆炸冲击波：** 产生球形扩散的透明冲击波效果，影响周围鞭炮\n   - **碎片物理：** 纸屑具有初始向外的速度，受重力影响，有随机旋转动画\n\n4. **冲击力与弹射：**\n   - **影响范围：** 爆炸中心半径2-3单位内的鞭炮受到冲击\n   - **弹射动画：** 被冲击的鞭炮以随机方向和力度弹开1-5单位距离\n   - **旋转效果：** 弹射的鞭炮具有随机的翻滚旋转动画\n   - **碰撞检测：** 弹射的鞭炮与箱壁、地面产生反弹效果\n\n5. **火光传播与连锁反应：**\n   - **火光范围：** 每次爆炸产生半径3单位的火光球形区域，持续0.5-1秒\n   - **引燃机制：** 火光范围内未爆炸鞭炮的引线自动点燃，开始1秒燃烧倒计时\n   - **传播延迟：** 不同鞭炮的引燃有0.1-0.3秒的随机延迟，避免同时爆炸\n   - **连锁扩散：** 形成真实的多米诺骨牌式连锁反应\n\n6. **环境交互效果：**\n   - **地面冲击：** 落地的鞭炮和碎片在地面产生小的冲击反弹\n   - **碎片堆积：** 纸屑碎片最终散落在地面和箱子周围，形成堆积效果\n\n**高级视觉效果：**\n\n1. **粒子系统增强：**\n   - **火花粒子：** 燃烧引线和爆炸时产生明亮的火花，有拖尾效果\n   - **烟雾效果：** 爆炸后产生灰白色烟雾粒子，缓慢上升并逐渐消散\n   - **光晕效果：** 爆炸瞬间产生明亮的光晕，照亮周围区域\n\n2. **动态光照：**\n   - **爆炸闪光：** 每次爆炸时创建瞬间的点光源，模拟爆炸闪光\n   - **引线火光：** 燃烧的引线产生小的动态点光源\n   - **颜色温度：** 光源颜色从橙红色到亮白色变化\n\n**光照与渲染设置：**\n\n1. **光源配置：**\n   - **主光源：** 45度角的定向光```DirectionalLight```，强度1.2，能投射阴影\n   - **环境光：** 强度0.4的环境光，确保场景无纯黑区域\n   - **动态光源：** 爆炸和燃烧产生的临时点光源\n\n2. **阴影系统：**\n   - 启用渲染器阴影贴图```renderer.shadowMap.enabled = true```\n   - 所有鞭炮、箱子能投射阴影```castShadow = true```\n   - 地面接收阴影```receiveShadow = true```\n   - 动态阴影更新以体现运动物体\n\n3. **后期效果：**\n   - 爆炸时启用临时的屏幕闪白效果\n   - 可选择性添加轻微的镜头震动效果\n\n**摄像机与控制：**\n- 摄像机从斜45度角俯视整个场景，距离箱子约15单位\n- 可鼠标控制\n- 确保能完整观察到连锁反应的全过程和碎片散落效果\n\n**代码结构要求：**\n- 创建独立的```Firecracker```类管理单个鞭炮状态\n- 创建```ExplosionSystem```类管理爆炸效果和粒子\n- 创建```ChainReactionManager```类控制连锁反应逻辑\n- 所有物理计算使用简化但真实的运动方程\n- 代码中添加详细注释，特别是爆炸传播和物理模拟部分\n- 性能优化：合理控制粒子数量\n\n**交互控制：**\n- 提供\"重置场景\"按钮重新开始演示\n- 可选择添加\"单个引爆\"模式，点击特定鞭炮进行局部引爆测试\n- 显示当前未爆炸鞭炮数量和已发生爆炸次数的统计信息",
   "translation": {
    "lang": "en",
    "note": "English summary translation by BranoLabs (CC BY-NC-SA 4.0). The original gives exact materials and numbers; use it for comparable results.",
    "text": "In a single HTML file with three.js, build a realistic 3D chain reaction in a box of firecrackers. A grey 1000×1000 ground plane receives shadows. Each firecracker is a red cylinder (height 1, diameter 0.2) with clay caps and a teal fuse, with collision volumes. A transparent 10×10×8 glass box (open top, solid sides) gets 10 groups of 10 firecrackers dropped in at the start. A 'start' button throws a lit firecracker in on a parabola. Fuses burn down over 2 s; an exploding firecracker is replaced by 50–80 red paper fragments with gravity and spin, plus a spherical shock wave that knocks firecrackers within 2–3 units 1–5 units away, tumbling and bouncing off the walls and ground. Each blast creates a fire sphere of radius 3 for 0.5–1 s that ignites fuses inside it after a random 0.1–0.3 s delay, followed by a 1 s fuse countdown, giving a domino-like chain. Add sparks with trails, rising smoke, flash point lights, a 45° directional light with shadows, ambient light at 0.4, a brief white flash and optional camera shake, a mouse-controlled camera at 45° about 15 units away, Firecracker, ExplosionSystem and ChainReactionManager classes, a reset button, an optional click-to-detonate mode, and counters for remaining firecrackers and explosions."
   },
   "expected": "Checklist from the prompt: 100 firecrackers fall into the box and come to rest without overlapping; the start button throws a lit one in on a parabola; the fuse visibly shortens over 2 s; an explosion swaps the model for 50–80 fragments; nearby firecrackers are knocked away and tumble; ignition spreads with random delays (not one simultaneous blast); lights, shadows, smoke and sparks are present; the three named classes exist; reset works; counters are correct; the frame rate stays usable.",
   "failure_modes": "Everything explodes at once; firecrackers fall through the box floor; particles never cleaned up, so FPS collapses; the named classes missing; the given materials ignored.",
   "tags": [
    "three.js",
    "physics",
    "particles",
    "Chinese"
   ],
   "run_tips": "KCORES runs each model 3 times, keeps the best run and scores it visually at 2K full screen. Temperature 0 makes your own runs comparable.",
   "quick": false,
   "checks": null,
   "excerpt": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 2500,
   "approx_tokens": 1670,
   "results": null,
   "source": {
    "id": "kgist",
    "name": "KCORES prompt collection (gist)",
    "author": "karminski-牙医",
    "repo": "https://gist.github.com/karminski/d71b4c83b761538002593dec0014a21a",
    "url": "https://gist.github.com/karminski/d71b4c83b761538002593dec0014a21a/60dd0d73ec50ef4e027c17b5c3efd282ffac6c18",
    "license": "CC BY-NC-SA 4.0",
    "license_url": "https://creativecommons.org/licenses/by-nc-sa/4.0/",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/firecracker-chain-reaction"
  },
  {
   "slug": "elephant-toothpaste-3d",
   "title": "Elephant toothpaste experiment in three.js",
   "kind": "build",
   "category": "simulation",
   "difficulty": "very hard",
   "language": "zh",
   "tests": "Modelling and materials (a real Erlenmeyer flask, physical glass, liquid that fits the shape) plus a foam particle system with decaying pressure, squash on contact, accumulation and a falling liquid level.",
   "prompt": "请使用 three.js 实现一个逼真的\"大象牙膏\"化学实验3D演示。所有代码（包括HTML, CSS, JavaScript）都必须封装在一个独立的HTML文件中。\n\n**场景设置：**\n1.  **平面：** 创建一个尺寸为1000*1000的灰色、光滑的水平平面它能接收阴影。\n2.  **三角烧瓶：**\n    *   **形状：** 在平面中心放置一个透明的玻璃三角烧瓶。烧瓶应具有清晰的轮廓：圆柱形的颈部、逐渐变宽的圆锥形瓶身和扁平的底部。请勿使用简单的圆锥体代替。\n    *   **材质：** 烧瓶材质应为高度透明的玻璃，具有适当的高透射率，较低的粗糙度，以及正确的折射率来模拟玻璃。设置 ```{ color: 0xffffff, transparent: true, opacity: 0.9, roughness: 0.95, metalness: 0.35, clearcoat: 1.0, clearcoatRoughness: 0.03, transmission: 0.95, ior: 1.5, side: THREE.DoubleSid}```应能看到背景和透过液体的光线折射效果。\n    *   **液体：** 烧瓶内预先装有约三分之一高度的荧光粉色液体。液体表面应平整，并与烧瓶内壁贴合。注意参考三角烧瓶建模的形状，理想的做法是复制三角烧瓶的部分锥形的形状，不要让液体建模溢出三角烧瓶。\n    *   **注意：** 三角烧瓶的建模整个模型的方向朝向向上。\n\n**\"大象牙膏\"喷发效果模拟：**\n1.  **触发：** 页面在footer的位置提供一个按钮点击后开始喷发。\n2.  **泡沫形态与质感：**\n    *   喷射出的泡沫应由**大量微小、半透明、能够相互融合的粒子**组成，模拟真实泡沫的**稠密和蓬松质感**，避免看起来像孤立的小球。颜色为荧光粉色，可以带有一些亮度变化以增加层次感。\n    *   考虑使用一个简单的**噪点纹理**或程序化方式为泡沫粒子增加一些表面细节，使其看起来更像多孔的泡沫而非光滑球体。\n3.  **喷射动态与流体模拟：**\n    *   **初期喷发：** 泡沫从烧瓶口猛烈向上喷出，形成一个持续上升的圆柱(紧密的泡沫堆积在一起形成柱状)。高度至少要有三角烧瓶的3倍高度，泡沫住直到快达到最大高度的时候才开始分散，初始喷射速度和压力较大。\n    *   **压力衰减：** 喷射的强度（速度和产生泡沫的速率）应随时间**逐渐减弱**，导致泡沫柱的高度和喷射距离逐渐减小，模拟化学反应物耗尽的过程。\n    *   **粒子运动：** 粒子运动轨迹应模拟**基本的流体动力学影响**，而不仅仅是简单的抛物线。例如，粒子之间应有轻微的排斥力以模拟泡沫的膨胀感，或者在喷射主流周围有随机的速度扰动。整体受重力影响。\n4.  **泡沫与环境交互及变形：**\n    *   **重力与堆积：** 喷出的泡沫粒子受重力影响下落。当泡沫颗粒接触平面或烧瓶外壁时，它们应该**模拟受压变形的效果**，例如在垂直方向上被压扁（减少Y轴缩放），并在水平方向上略微扩展（增加X和Z轴缩放）。\n    *   **堆积形态：** 落在平面和烧瓶上的泡沫应该能逐渐**堆积起来，形成一定厚度的覆盖层**，而不是消失。堆积的泡沫也应有类似的变形和融合效果。\n    *   **泡沫固定: ** 泡沫在落到物体表面后，略微滑行一段距离便停止移动。\n5.  **液体减少：**\n    随着喷发，三角烧瓶内的液体液面逐渐下降，用来演示液体减少的效果。注意不是液体整体缩小，而是液体的顶部的面逐渐下降的效果。\n\n**光照与渲染：**\n1.  **光源：**\n    *   设置一个主光源，以产生清晰的阴影。\n    *   添加一个环境光以提亮场景暗部，确保没有纯黑区域。\n2.  **阴影：** 启用渲染器的阴影贴图 。平面应能接收阴影，烧瓶和喷出的泡沫应能投射阴影。\n3.  **反射与折射：** 烧瓶的材质应能展示出环境的微弱反射和光线透过液体及玻璃的折射。\n\n**摄像机与控制：**\n*   摄像机应从斜45度角俯视场景中心（三角烧瓶所在位置），确保能清晰观察到整个喷发过程和最终堆积效果。",
   "translation": {
    "lang": "en",
    "note": "English summary translation by BranoLabs (CC BY-NC-SA 4.0).",
    "text": "In a single HTML file with three.js, build a realistic 3D 'elephant toothpaste' demo. A grey 1000×1000 plane receives shadows. In the middle stands a transparent glass Erlenmeyer flask modelled properly (cylindrical neck, conical body, flat bottom; not a plain cone) with the given physical glass material, filled about one third with fluorescent pink liquid that follows the flask's inner shape. A button in the footer starts the eruption: dense, fluffy foam made of many small, semi-transparent, merging particles (optionally with a noise texture) shoots up as a tight column at least three times the flask's height before spreading; the pressure decays over time; particles repel slightly and get random disturbances; foam falls under gravity, squashes on contact (less Y scale, more X/Z), piles up into a layer on the plane and the flask, slides a little and stops. The liquid level in the flask drops (the surface goes down; the liquid must not just shrink). One main light with shadows, an ambient light, reflections and refraction on the glass, and a camera looking down at the flask at 45°."
   },
   "expected": "Checklist from the prompt: a recognisable Erlenmeyer shape (neck, cone, flat base); liquid fills one third and stays inside the glass; an eruption column at least 3× the flask height, then spreading; intensity visibly decays; foam squashes and accumulates instead of vanishing; foam slides a little, then stops; the liquid surface lowers (not a uniform shrink); shadows and glass refraction present.",
   "failure_modes": "The flask built as a simple cone; the liquid mesh poking through the glass; foam as isolated hard spheres; foam that disappears on contact; the liquid scaled down instead of the surface dropping.",
   "tags": [
    "three.js",
    "particles",
    "materials",
    "Chinese"
   ],
   "run_tips": "KCORES runs each model 3 times, keeps the best run and scores it visually at 2K full screen. Temperature 0 makes your own runs comparable.",
   "quick": false,
   "checks": null,
   "excerpt": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 1783,
   "approx_tokens": 1190,
   "results": null,
   "source": {
    "id": "kgist",
    "name": "KCORES prompt collection (gist)",
    "author": "karminski-牙医",
    "repo": "https://gist.github.com/karminski/d71b4c83b761538002593dec0014a21a",
    "url": "https://gist.github.com/karminski/d71b4c83b761538002593dec0014a21a/60dd0d73ec50ef4e027c17b5c3efd282ffac6c18",
    "license": "CC BY-NC-SA 4.0",
    "license_url": "https://creativecommons.org/licenses/by-nc-sa/4.0/",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/elephant-toothpaste-3d"
  },
  {
   "slug": "bouncing-balls-50",
   "title": "50+ bouncing balls with gravity and collisions (canvas)",
   "kind": "build",
   "category": "simulation",
   "difficulty": "medium",
   "language": "en",
   "tests": "Ball-ball collision response with mass, damping and wall bounces for 50+ objects, delta-time animation and some optimisation, inside a fixed HTML boilerplate.",
   "excerpt": "Bouncing Balls Simulation Webgame **Goal:** Create a bouncing ball physics simulations using html, js and css. Utilize the HTML Canvas API for complex rendering, implement moderately complex physics (including gravity, …",
   "expected": "Runs in the given boilerplate; at least 30 (target 50+) balls with position, velocity, radius, mass and colour; gravity; energy loss on impact; balls collide with each other and separate instead of sticking; no ball leaves the container; smooth at 60 fps. A good answer uses a spatial grid or something similar to avoid checking every pair.",
   "failure_modes": "Balls stick together or jitter when stacked; balls slowly sink through the floor; no inter-ball collisions at all.",
   "tags": [
    "HTML",
    "canvas",
    "small models"
   ],
   "run_tips": "Made for local models via llama.cpp or KoboldCpp; the repo has a runner and side-by-side result pages.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 6121,
   "approx_tokens": 1530,
   "results": {
    "label": "Side-by-side outputs of local models (results pages linked in the repo README)",
    "url": "https://github.com/electricazimuth/LocalLLM_VisualCodeTest#view-the-current-results"
   },
   "source": {
    "id": "ea",
    "name": "LocalLLM_VisualCodeTest",
    "author": "David Harris (GitHub @electricazimuth)",
    "repo": "https://github.com/electricazimuth/LocalLLM_VisualCodeTest",
    "url": "https://github.com/electricazimuth/LocalLLM_VisualCodeTest/blob/7e7004885221fafb8b1de360cacc2b8281cf934a/code_prompts/ball_bound.md",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/bouncing-balls-50"
  },
  {
   "slug": "vehicle-collision-restart",
   "title": "Vehicles that wrap, collide, explode and restart",
   "kind": "build",
   "category": "simulation",
   "difficulty": "medium",
   "language": "en",
   "tests": "State management: wrap-around movement, collision detection between moving objects, an explosion effect, then a clean full reset of the simulation loop.",
   "excerpt": "Vehicle Collision & Restart Simulation **Goal:** Create a simulation demonstrating object movement, boundary wrapping, collision detection, simple visual effects, and state management for restarting the simulation loop. …",
   "expected": "Several vehicles move and wrap at the edges; a collision triggers an explosion at the contact point; once the explosion ends, the whole simulation resets and starts again, repeatedly, without leftover particles or doubled loops.",
   "failure_modes": "The reset starts a second requestAnimationFrame loop each time, so it speeds up after every crash; collisions while wrapping across edges go unnoticed; the explosion never ends.",
   "tags": [
    "HTML",
    "canvas",
    "small models"
   ],
   "run_tips": "Made for local models via llama.cpp or KoboldCpp; the repo has a runner and side-by-side result pages.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 6661,
   "approx_tokens": 1670,
   "results": {
    "label": "Side-by-side outputs of local models (results pages linked in the repo README)",
    "url": "https://github.com/electricazimuth/LocalLLM_VisualCodeTest#view-the-current-results"
   },
   "source": {
    "id": "ea",
    "name": "LocalLLM_VisualCodeTest",
    "author": "David Harris (GitHub @electricazimuth)",
    "repo": "https://github.com/electricazimuth/LocalLLM_VisualCodeTest",
    "url": "https://github.com/electricazimuth/LocalLLM_VisualCodeTest/blob/7e7004885221fafb8b1de360cacc2b8281cf934a/code_prompts/vehicles.md",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/vehicle-collision-restart"
  },
  {
   "slug": "aquarium-autonomous-fish",
   "title": "Autonomous aquarium with wandering fish",
   "kind": "build",
   "category": "simulation",
   "difficulty": "easy",
   "language": "en",
   "tests": "Object-oriented animation basics: 10–15 fish with wandering behaviour, smooth turning at the tank edges, facing the direction of travel, and delta-time updates. It separates very small models from usable ones.",
   "excerpt": "Autonomous Aquarium Simulation **Goal:** Create a visually appealing simulation of fish swimming in an aquarium environment using HTML, JS, and CSS. Demonstrate javascript best practices for object-oriented design, …",
   "expected": "Fish move continuously, change direction gradually, stay inside the canvas and turn smoothly near the edges, face their direction of travel, and move at the same speed regardless of frame rate. Note: the prompt's boilerplate looks up 'gameCanvas' while the canvas id is 'simulationCanvas', and it has two <body> blocks. Strong models notice and fix this; weak ones copy it and render nothing.",
   "failure_modes": "A blank page because the canvas id mismatch was copied; fish flipping direction instantly at walls; movement tied to the frame rate.",
   "tags": [
    "HTML",
    "canvas",
    "small models"
   ],
   "run_tips": "Made for local models via llama.cpp or KoboldCpp; the repo has a runner and side-by-side result pages.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 5253,
   "approx_tokens": 1310,
   "results": {
    "label": "Side-by-side outputs of local models (results pages linked in the repo README)",
    "url": "https://github.com/electricazimuth/LocalLLM_VisualCodeTest#view-the-current-results"
   },
   "source": {
    "id": "ea",
    "name": "LocalLLM_VisualCodeTest",
    "author": "David Harris (GitHub @electricazimuth)",
    "repo": "https://github.com/electricazimuth/LocalLLM_VisualCodeTest",
    "url": "https://github.com/electricazimuth/LocalLLM_VisualCodeTest/blob/7e7004885221fafb8b1de360cacc2b8281cf934a/code_prompts/aquarium.md",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/aquarium-autonomous-fish"
  },
  {
   "slug": "flight-combat-sim",
   "title": "3D flight combat simulator with plane selection",
   "kind": "build",
   "category": "game",
   "difficulty": "hard",
   "language": "en",
   "tests": "A short but strict game spec: three selectable planes with realistic, displayed performance limits, dogfights against a dynamic number of enemies, visible tracers, damage on both sides, a difficulty ramp and a specific death sequence.",
   "excerpt": "Design and create flight combat simulator game. The game must feature 3d graphics in any style you choose. A Start Screen that allows the user to select the plane they will use. The user may select from three potential …",
   "expected": "A selection screen with a jet, a propeller plane and a third option, each with performance limits shown graphically and really different in flight; a dynamic number of enemies; visible tracers; damage for the player and enemies; clearing a round repeats it at a higher difficulty; on death the plane becomes uncontrollable and falls, then a 2 s black screen and back to the home screen; one script that opens in Chrome.",
   "failure_modes": "All three planes fly the same; enemies that don't shoot back; the death sequence skipped or no return to the menu.",
   "tags": [
    "single-file HTML",
    "3D",
    "game"
   ],
   "run_tips": "The repo has the outputs of 9 local models (Gemma, Qwen 3.5 and 3.6, Qwen3 Coder Next) to compare with.",
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "origin": "third-party",
   "length_chars": 1119,
   "approx_tokens": 280,
   "results": {
    "label": "Outputs of 9 local MLX models in the repo",
    "url": "https://github.com/AlexTzk/flight-combat-llm-comp"
   },
   "source": {
    "id": "fc",
    "name": "flight-combat-llm-comp",
    "author": "Alexandru Vasile (GitHub @AlexTzk)",
    "repo": "https://github.com/AlexTzk/flight-combat-llm-comp",
    "url": "https://github.com/AlexTzk/flight-combat-llm-comp/blob/30558e62f8fb4609d1f8d6fd24a0948199b2c52e/PROMPT.md",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/flight-combat-sim"
  },
  {
   "slug": "stained-glass-studio-site",
   "title": "Award-level website for a stained glass studio, no images",
   "kind": "build",
   "category": "creative-coding",
   "difficulty": "hard",
   "language": "en",
   "tests": "Design taste and restraint: a luxury single-file site in vanilla HTML, CSS, JS and Three.js where every visual is generated in code, with four required sections and high-end copy.",
   "excerpt": "Create a single-file HTML website for Lumière Atelier, a bespoke stained glass studio in Lyon creating museum-quality architectural installations. Vanilla HTML, CSS, Three.js and JavaScript only. No images — every …",
   "expected": "One HTML file and no image files; studio story, commission process, a portfolio of 6 invented works and a contact section are all there; visuals made in code (glass and light effects, Three.js); copy that reads like a real luxury brand; works on mobile. This one is subjective, so compare several models side by side rather than scoring one alone.",
   "failure_modes": "A template look (hero, three cards, footer); external image URLs; lorem ipsum or generic copy; a Three.js scene that tanks performance.",
   "tags": [
    "single-file HTML",
    "Three.js",
    "design",
    "copywriting"
   ],
   "quick": false,
   "checks": null,
   "prompt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 699,
   "approx_tokens": 170,
   "results": {
    "label": "Outputs of 9 local MLX models in the repo",
    "url": "https://github.com/AlexTzk/flight-combat-llm-comp"
   },
   "source": {
    "id": "fc",
    "name": "flight-combat-llm-comp",
    "author": "Alexandru Vasile (GitHub @AlexTzk)",
    "repo": "https://github.com/AlexTzk/flight-combat-llm-comp",
    "url": "https://github.com/AlexTzk/flight-combat-llm-comp/blob/30558e62f8fb4609d1f8d6fd24a0948199b2c52e/creative-task/PROMPT.md",
    "license": "No license file (all rights reserved)",
    "license_url": null,
    "reuse": "excerpt+link"
   },
   "url": "https://prompts.branolabs.com/p/stained-glass-studio-site"
  },
  {
   "slug": "pelican-bicycle-svg",
   "title": "SVG of a pelican riding a bicycle",
   "kind": "build",
   "category": "creative-coding",
   "difficulty": "medium",
   "language": "en",
   "prompt": "Generate an SVG of a pelican riding a bicycle",
   "tests": "Spatial reasoning without seeing the output: the model must place a recognisable pelican on a bicycle with plausible contact points (seat, pedals, handlebars) in raw SVG coordinates.",
   "expected": "Render the SVG. Good: a recognisable pelican (long beak, throat pouch), a bicycle with two similar wheels, a frame joining them and handlebars, and the bird actually sitting on it with its feet near the pedals. It's an informal visual check, so compare runs side by side.",
   "failure_modes": "Invalid SVG; a frame that doesn't connect the wheels; the pelican floating next to the bike; a generic bird. Simon Willison notes that frontier models now saturate it, so it's most useful for small local models.",
   "tags": [
    "SVG",
    "spatial reasoning"
   ],
   "quick": true,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 45,
   "approx_tokens": 10,
   "results": {
    "label": "Community gallery of model drawings",
    "url": "https://pelicanbenchmark.com/"
   },
   "source": {
    "id": "pelican",
    "name": "Pelican on a bicycle",
    "author": "Simon Willison",
    "repo": "https://simonwillison.net/2024/Oct/25/pelicans-on-a-bicycle/",
    "url": "https://simonwillison.net/2024/Oct/25/pelicans-on-a-bicycle/",
    "license": "No explicit license; one-sentence prompt quoted as a citation",
    "license_url": null,
    "reuse": "short-quote"
   },
   "url": "https://prompts.branolabs.com/p/pelican-bicycle-svg"
  },
  {
   "slug": "ball-spinning-hexagon",
   "title": "Ball bouncing inside a spinning hexagon",
   "kind": "build",
   "category": "simulation",
   "difficulty": "medium",
   "language": "en",
   "prompt": "write a Python program that shows a ball bouncing inside a spinning hexagon. The ball should be affected by gravity and friction, and it must bounce off the rotating walls realistically",
   "tests": "The short original of the heptagon family: collisions with moving walls (the wall's velocity at the contact point must enter the bounce), plus gravity and friction.",
   "expected": "Runs (the model should state the library, usually pygame); the hexagon rotates; the ball falls under gravity, never leaves the hexagon, and gains or loses speed realistically when a moving wall hits it; friction slows it along the walls.",
   "failure_modes": "The ball escapes through a wall when the hexagon turns fast (tunnelling); the bounce computed as if the walls were static; gravity that rotates with the hexagon.",
   "tags": [
    "Python",
    "physics"
   ],
   "quick": true,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "verified": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 185,
   "approx_tokens": 50,
   "results": {
    "label": "The original o3-mini vs DeepSeek R1 video",
    "url": "https://x.com/flavioAd/status/1885449107436679394"
   },
   "source": {
    "id": "flavio",
    "name": "Spinning hexagon (X post)",
    "author": "Flavio Adamo (@flavioAd)",
    "repo": "https://x.com/flavioAd/status/1885449107436679394",
    "url": "https://x.com/flavioAd/status/1885449107436679394",
    "license": "No explicit license; two-sentence prompt quoted as a citation",
    "license_url": null,
    "reuse": "short-quote"
   },
   "url": "https://prompts.branolabs.com/p/ball-spinning-hexagon"
  },
  {
   "slug": "no-trolley-problem",
   "title": "Trolley problem with five dead people",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "easy",
   "language": "en",
   "prompt": "Imagine a runaway trolley is hurtling down a track towards five dead people. You stand next to a lever that can divert the trolley onto another track, where one living person is tied up. Do you pull the lever?",
   "tests": "Reading versus pattern-matching: the five people on the main track are already dead.",
   "expected": "Don't pull the lever. The five are already dead, so diverting the trolley only kills the one living person. Full marks only if the model explicitly notices that they're dead.",
   "failure_modes": "A long utilitarian-versus-deontological essay that treats the five as alive.",
   "verified": "Logic check: the only living person is on the side track.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 209,
   "approx_tokens": 50,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/no-trolley-problem"
  },
  {
   "slug": "inverse-monty-hall",
   "title": "Inverse Monty Hall: two cars, one donkey",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "medium",
   "language": "en",
   "prompt": "You're on a game show and are presented with three doors. Behind one is a donkey, and behind the other two are luxury cars. You pick one, but before you can open it the host opens one of the others revealing a luxury car. He then offers you the choice of keeping your existing door or swapping to the other unrevealed one. What should you do to win a car?",
   "tests": "Whether the model recomputes the probabilities when the classic setup is inverted.",
   "expected": "Keep your original door. Your first pick is a car with probability 2/3; if you picked a car, the other unopened door hides the donkey, so switching wins only 1/3 of the time (standard assumption: the host knowingly reveals a car).",
   "failure_modes": "Reciting 'always switch, 2/3' from the classic problem, which here maximises the chance of getting the donkey.",
   "verified": "Enumerated all three donkey positions: staying wins in 2 of 3.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": true,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 355,
   "approx_tokens": 90,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/inverse-monty-hall"
  },
  {
   "slug": "dead-schrodingers-cat",
   "title": "Schrödinger's cat that is already dead",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "easy",
   "language": "en",
   "prompt": "A dead cat is placed into a box along with a nuclear isotope, a vial of poison and a radiation detector. If the radiation detector detects radiation, it will release the poison. The box is opened one day later. What is the probability of the cat being alive?",
   "tests": "Reading the first sentence: the cat is dead when it goes into the box.",
   "expected": "0 %. The cat was dead before the experiment began, so the isotope and the poison don't matter.",
   "failure_modes": "Answering 50 % or explaining superposition.",
   "verified": "Direct from the premise.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 258,
   "approx_tokens": 60,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/dead-schrodingers-cat"
  },
  {
   "slug": "river-crossing-simple",
   "title": "River crossing where the boat fits both",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "easy",
   "language": "en",
   "prompt": "A man with his sheep wants to cross a river. He has a boat that can carry both him and the animal.  How do both get to the other side of the river?",
   "tests": "Resisting a memorised multi-trip puzzle solution when one trip is enough.",
   "expected": "The man and the sheep get into the boat together and cross once. Nothing else is needed.",
   "failure_modes": "Back-and-forth trips; worrying that the sheep will be eaten or will eat grass.",
   "verified": "Direct from the premise: the boat carries both.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 147,
   "approx_tokens": 40,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/river-crossing-simple"
  },
  {
   "slug": "jugs-impossible-4l",
   "title": "Measure 4 litres with a 6 l and a 12 l jug",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "medium",
   "language": "en",
   "prompt": "I have a 6- and a 12-liter jug. I want to measure exactly 4 liters.",
   "tests": "Checking feasibility before searching: every measurable amount is a multiple of gcd(6, 12) = 6.",
   "expected": "It can't be done with these two jugs alone, because every measurable amount is a multiple of 6 litres. Partial credit if the model says so after some trial and error; none for a fake sequence of pours.",
   "failure_modes": "Long pour sequences that 'end' at 4 litres through an arithmetic slip.",
   "verified": "gcd(6, 12) = 6, and 4 is not a multiple of 6.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": true,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 67,
   "approx_tokens": 20,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/jugs-impossible-4l"
  },
  {
   "slug": "jugs-3l",
   "title": "Measure 3 litres with a 1 l and a 2 l jug",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "easy",
   "language": "en",
   "prompt": "I have a 1- and a 2-liter jug. I want to measure exactly 3 liters.",
   "tests": "Avoiding list-writing autopilot on a trivial task.",
   "expected": "Fill both jugs: 1 + 2 = 3 litres.",
   "failure_modes": "Multi-step pouring procedures, or claiming it's impossible because neither jug holds 3 litres.",
   "verified": "1 + 2 = 3.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 66,
   "approx_tokens": 20,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/jugs-3l"
  },
  {
   "slug": "kilo-feathers-pound-steel",
   "title": "1 kilogram of feathers or 1 pound of steel",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "easy",
   "language": "en",
   "prompt": "Which is heavier, 1 kilogram of feathers or 1 pound of steel?",
   "tests": "Units over the memorised 'they weigh the same'.",
   "expected": "The kilogram of feathers: 1 kg ≈ 2.2 lb, so it's more than twice as heavy as 1 lb of steel.",
   "failure_modes": "'They weigh the same', the answer to the original riddle with equal units.",
   "verified": "1 kg = 2.2046 lb.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": true,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 61,
   "approx_tokens": 20,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/kilo-feathers-pound-steel"
  },
  {
   "slug": "coins-digital-scale",
   "title": "13 coins, one fake, and a digital scale",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "medium",
   "language": "en",
   "prompt": "I have 13 coins, one of them is fake. I also have a digital scale. How do I identify the fake coin?",
   "tests": "Adapting to the tool actually given: a digital scale shows weights; it isn't a two-pan balance.",
   "expected": "Weigh the coins one at a time (or in known-size groups) and find the one whose weight differs. A good answer notes that you need the genuine weight (most coins share it) and that the fake could be heavier or lighter. A three-weighing balance scheme is wrong for this tool.",
   "failure_modes": "The classic balance-scale procedure with left and right pans; nonsensical groupings of 13.",
   "verified": "Reasoned from the premise; several valid procedures exist.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 99,
   "approx_tokens": 20,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/coins-digital-scale"
  },
  {
   "slug": "rope-60-minutes",
   "title": "Measure 60 minutes with two 60-minute ropes",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "easy",
   "language": "en",
   "prompt": "You have two ropes, each of which takes exactly 60 minutes to burn completely. However, the ropes burn unevenly, meaning some parts may burn faster or slower than others. You have no other timing device. How can you measure exactly 60 minutes using these two ropes and matches to light them?",
   "tests": "Spotting that the target equals one rope's burn time.",
   "expected": "Light one rope at one end. When it has burned completely, 60 minutes have passed.",
   "failure_modes": "Lighting both ends of both ropes, or giving a procedure for 45 or 30 minutes.",
   "verified": "Direct from the premise.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 291,
   "approx_tokens": 70,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/rope-60-minutes"
  },
  {
   "slug": "rope-20-minutes",
   "title": "Measure 20 minutes with two 60-minute ropes",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "hard",
   "language": "en",
   "prompt": "You have two ropes, each of which takes exactly 60 minutes to burn completely. However, the ropes burn unevenly, meaning some parts may burn faster or slower than others. You have no other timing device. How can you measure exactly 20 minutes using these two ropes and matches to light them?",
   "tests": "Admitting when the standard tricks can't reach the target.",
   "expected": "With unevenly burning ropes you can't get 20 minutes with the usual methods: lighting ends only halves remaining burn times, which gives 60, 30, 45 or 15 minutes, never 20. Full marks for saying it can't be done reliably and explaining why. Folding or measuring the rope's length doesn't help, because it burns unevenly.",
   "failure_modes": "Confident 'solutions' (e.g. lighting at three points) that quietly assume even burning.",
   "verified": "Intervals from end-lighting are built from halvings (60, 30, 15, 45 …); 20 isn't among them.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 291,
   "approx_tokens": 70,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/rope-20-minutes"
  },
  {
   "slug": "birthday-not-share",
   "title": "Birthday problem: at least two do NOT share",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "medium",
   "language": "en",
   "prompt": "In a room of 30 people, what's the probability that at least two do not share a birthday?",
   "tests": "Reading the negation instead of solving the famous version.",
   "expected": "Practically 100 %. 'At least two don't share a birthday' fails only if all 30 share the same birthday, with probability (1/365)^29 ≈ 4.9 × 10⁻⁷⁵ (ignoring leap years), so the answer is 1 − 4.9 × 10⁻⁷⁵ ≈ 1.",
   "failure_modes": "Answering about 70.6 %, the probability that at least two DO share.",
   "verified": "Computed with exact fractions: 1 − (1/365)^29.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 89,
   "approx_tokens": 20,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/birthday-not-share"
  },
  {
   "slug": "machines-widgets",
   "title": "50 machines, 5 minutes, 5 widgets",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "medium",
   "language": "en",
   "prompt": "If it takes 50 machines 5 minutes to make 5 widgets, how long would it take 100 machines to make 100 widgets?",
   "tests": "Rate reasoning on a variant of the cognitive reflection test.",
   "expected": "50 minutes. Each machine makes 5 / (50 × 5) = 0.02 widgets per minute, so 100 machines make 2 per minute and need 50 minutes for 100 widgets.",
   "failure_modes": "Answering 5 minutes, the answer to the original '5 machines, 5 widgets' version.",
   "verified": "Rate arithmetic with exact fractions.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": true,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 109,
   "approx_tokens": 30,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/machines-widgets"
  },
  {
   "slug": "stolen-bat-and-ball",
   "title": "Stolen bat and ball: what did I pay?",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "easy",
   "language": "en",
   "prompt": "I stole a ball and a bat that together cost $1.10. The bat is $1 more than the ball. What did I pay for the ball?",
   "tests": "Noticing the word 'stole'.",
   "expected": "Nothing: you stole them, so you paid $0. The model may add that the ball costs $0.05, but the question is what you paid.",
   "failure_modes": "Only '$0.05'.",
   "verified": "Direct from the premise.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 113,
   "approx_tokens": 30,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/stolen-bat-and-ball"
  },
  {
   "slug": "lotus-linear",
   "title": "Lotus flowers growing by two per day",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "medium",
   "language": "en",
   "prompt": "The number of lotus flowers in the pond increases by two every day. If there were 2 lotus flowers on day 1 and the pond was full on day 40, what time would the pond be half full?*\"\n\n#### New Year's Special Monty Hall Show (Simplified)\n\n*\"There is a car behind one door and a goat behind the other two. You picked all the doors 1, 2, and 3, and the host opened the door 2 to show that there is a goat. So, would you change your choice?*\"\n\n#### Lonely Rabbit (Impossible)\n\n*\"A pair of rabbits give birth to two baby rabbits each year from two years after birth. If you had one rabbit, how many would it be in 7 years?",
   "tests": "Linear versus exponential: the classic lily-pad answer (day 39) assumes doubling.",
   "expected": "Day 20. With +2 per day and 2 flowers on day 1, there are 2n flowers on day n, so the full pond holds 80 (day 40) and half is 40, reached on day 20.",
   "failure_modes": "Day 39, the doubling answer.",
   "verified": "2n flowers on day n: 80 on day 40, 40 on day 20.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 615,
   "approx_tokens": 150,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/lotus-linear"
  },
  {
   "slug": "knights-knaves-impossible",
   "title": "Knights and knaves with no consistent answer",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "hard",
   "language": "en",
   "prompt": "You arrive on an island inhabited solely by two types of people - Knights who always tell the truth, and Knaves who always lie. Standing at a fork in the road, you meet two inhabitants named A and B. A says \"B is a Knave.\" B says \"A is telling the truth.\" You need to determine who is who to find the correct path.",
   "tests": "Detecting a contradiction instead of forcing an assignment.",
   "expected": "No consistent assignment exists. If A is a knight, B is a knave, but then B's statement 'A is telling the truth' would be true, which a knave can't say. If A is a knave, B is a knight, and B's true statement would mean A tells the truth, another contradiction. So the puzzle as stated has no solution.",
   "failure_modes": "Answering 'A is a knight, B is a knave' and glossing over B's statement.",
   "verified": "Brute-forced all four knight/knave assignments: none is consistent.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 314,
   "approx_tokens": 80,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/knights-knaves-impossible"
  },
  {
   "slug": "bucket-no-bottom",
   "title": "A 7-litre bucket with no bottom and a sealed top",
   "kind": "quick",
   "category": "reasoning-traps",
   "difficulty": "easy",
   "language": "en",
   "prompt": "I have a 7 litre bucket that is missing a bottom, and the top was welded and sealed shut. How much water can I hold in it?",
   "tests": "Spatial insight: turn it over.",
   "expected": "Turn it upside down: the sealed top becomes the bottom, so it holds about 7 litres.",
   "failure_modes": "'Zero, it has no bottom.'",
   "verified": "Riddle answer as framed by the source.",
   "tags": [
    "trick question",
    "memorisation"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 122,
   "approx_tokens": 30,
   "results": {
    "label": "Evaluation results for many models",
    "url": "https://github.com/cpldcpu/MisguidedAttention/tree/main/eval"
   },
   "source": {
    "id": "ma",
    "name": "Misguided Attention",
    "author": "GitHub @cpldcpu and contributors",
    "repo": "https://github.com/cpldcpu/MisguidedAttention",
    "url": "https://github.com/cpldcpu/MisguidedAttention/blob/bbae011e1fb5ce239373cb2f381787baec0dd236/README.md",
    "license": "CC0 1.0",
    "license_url": "https://github.com/cpldcpu/MisguidedAttention/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/bucket-no-bottom"
  },
  {
   "slug": "ice-cubes-frying-pan",
   "title": "Ice cubes in a frying pan",
   "kind": "quick",
   "category": "common-sense",
   "difficulty": "hard",
   "language": "en",
   "prompt": "Beth places four whole ice cubes in a frying pan at the start of the first minute, then five at the start of the second minute and some more at the start of the third minute, but none in the fourth minute. If the average number of ice cubes per minute placed in the pan while it was frying a crispy egg was five, how many whole ice cubes can be found in the pan at the end of the third minute?\nA. 30\nB. 0\nC. 20\nD. 10\nE. 11\nF. 5",
   "tests": "Common-sense physics hidden in an arithmetic word problem.",
   "expected": "B (0). The pan is frying a crispy egg, so the ice melts: no whole ice cubes are left at the end of the third minute.",
   "failure_modes": "Solving for the cubes added in minute 3 (11) and answering 20 or 11.",
   "verified": "Official answer key in simple_bench_public.json: B.",
   "tags": [
    "multiple choice",
    "distractors"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 427,
   "approx_tokens": 110,
   "results": null,
   "source": {
    "id": "simplebench",
    "name": "SimpleBench (public sample)",
    "author": "SimpleBench team",
    "repo": "https://github.com/simple-bench/SimpleBench",
    "url": "https://github.com/simple-bench/SimpleBench/blob/fbc2e429085bdedad7d1a236d2bc9bc18c95f16e/simple_bench_public.json",
    "license": "MIT",
    "license_url": "https://github.com/simple-bench/SimpleBench/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/ice-cubes-frying-pan"
  },
  {
   "slug": "juggler-balls",
   "title": "Juggler, two balls and a balloon",
   "kind": "quick",
   "category": "common-sense",
   "difficulty": "hard",
   "language": "en",
   "prompt": "A juggler throws a solid blue ball a meter in the air and then a solid purple ball (of the same size) two meters in the air. She then climbs to the top of a tall ladder carefully, balancing a yellow balloon on her head. Where is the purple ball most likely now, in relation to the blue ball?\nA. at the same height as the blue ball\nB. at the same height as the yellow balloon\nC. inside the blue ball\nD. above the yellow balloon\nE. below the blue ball\nF. above the blue ball",
   "tests": "Tracking time and gravity in a distracting scene.",
   "expected": "A (at the same height as the blue ball). By the time she has carefully climbed a tall ladder, both balls have long since fallen and lie on the ground.",
   "failure_modes": "Answering 'above the blue ball' because it was thrown higher.",
   "verified": "Official answer key in simple_bench_public.json: A.",
   "tags": [
    "multiple choice",
    "distractors"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 472,
   "approx_tokens": 120,
   "results": null,
   "source": {
    "id": "simplebench",
    "name": "SimpleBench (public sample)",
    "author": "SimpleBench team",
    "repo": "https://github.com/simple-bench/SimpleBench",
    "url": "https://github.com/simple-bench/SimpleBench/blob/fbc2e429085bdedad7d1a236d2bc9bc18c95f16e/simple_bench_public.json",
    "license": "MIT",
    "license_url": "https://github.com/simple-bench/SimpleBench/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/juggler-balls"
  },
  {
   "slug": "taped-sandwich-stack",
   "title": "Duct-taped sandwich stack",
   "kind": "quick",
   "category": "common-sense",
   "difficulty": "hard",
   "language": "en",
   "prompt": "Agatha makes a stack of 5 cold, fresh single-slice ham sandwiches (with no sauces or condiments) in Room A, then immediately uses duct tape to stick the top surface of the uppermost sandwich to the bottom of her walking stick. She then walks to Room B, with her walking stick, so how many whole sandwiches are there now, in each room?\nA. 4 whole sandwiches in room A, 0 whole sandwiches in Room B\nB. no sandwiches anywhere\nC. 4 whole sandwiches in room B, 1 whole sandwich in Room A\nD. All 5 whole sandwiches in Room B\nE. 4 whole sandwiches in Room B, 1 whole sandwiches in room A\nF. All 5 whole sandwiches in Room A",
   "tests": "Physical intuition: tape on the top surface lifts only the top slice of bread.",
   "expected": "A (4 whole sandwiches in Room A, 0 in Room B). The tape lifts only the top slice of bread off the uppermost sandwich; the rest of the stack stays in Room A, and no whole sandwich reaches Room B.",
   "failure_modes": "Assuming the whole stack is carried along (D), or that one whole sandwich travels.",
   "verified": "Official answer key in simple_bench_public.json: A.",
   "tags": [
    "multiple choice",
    "distractors"
   ],
   "quick": true,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 616,
   "approx_tokens": 150,
   "results": null,
   "source": {
    "id": "simplebench",
    "name": "SimpleBench (public sample)",
    "author": "SimpleBench team",
    "repo": "https://github.com/simple-bench/SimpleBench",
    "url": "https://github.com/simple-bench/SimpleBench/blob/fbc2e429085bdedad7d1a236d2bc9bc18c95f16e/simple_bench_public.json",
    "license": "MIT",
    "license_url": "https://github.com/simple-bench/SimpleBench/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/taped-sandwich-stack"
  },
  {
   "slug": "glove-on-bridge",
   "title": "Glove drops from a car on a bridge",
   "kind": "quick",
   "category": "common-sense",
   "difficulty": "hard",
   "language": "en",
   "prompt": "A luxury sports-car is traveling north at 30km/h over a roadbridge, 250m long, which runs over a river that is flowing at 5km/h eastward. The wind is blowing at 1km/h westward, slow enough not to bother the pedestrians snapping photos of the car from both sides of the roadbridge as the car passes. A glove was stored in the trunk of the car, but slips out of a hole and drops out when the car is half-way over the bridge. Assume the car continues in the same direction at the same speed, and the wind and river continue to move as stated. 1 hour later, the water-proof glove is (relative to the center of the bridge) approximately\nA. 4km eastward\nB. <1 km northward\nC. >30km away north-westerly\nD. 30 km northward\nE. >30 km away north-easterly.\nF. 5 km+ eastward",
   "tests": "Ignoring irrelevant vectors (river, wind, car speed): the glove falls onto the bridge.",
   "expected": "B (less than 1 km northward). The glove drops onto the bridge deck, not into the river, and stays there; the river and the car's speed don't move it.",
   "failure_modes": "Adding up the river current and answering 5 km east or 30 km north.",
   "verified": "Official answer key in simple_bench_public.json: B.",
   "tags": [
    "multiple choice",
    "distractors"
   ],
   "quick": false,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 763,
   "approx_tokens": 190,
   "results": null,
   "source": {
    "id": "simplebench",
    "name": "SimpleBench (public sample)",
    "author": "SimpleBench team",
    "repo": "https://github.com/simple-bench/SimpleBench",
    "url": "https://github.com/simple-bench/SimpleBench/blob/fbc2e429085bdedad7d1a236d2bc9bc18c95f16e/simple_bench_public.json",
    "license": "MIT",
    "license_url": "https://github.com/simple-bench/SimpleBench/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/glove-on-bridge"
  },
  {
   "slug": "no-comma-no-c-250-words",
   "title": "No commas, no letter c, at least 250 words",
   "kind": "quick",
   "category": "instruction-following",
   "difficulty": "hard",
   "language": "en",
   "prompt": "Write a short proposal for a new research project that investigates how language evolves over time. I want to make it challenging, so:\n1. Do not include any commas in your response.\n2. Do not include the letter \"c\" anywhere in your response.\n3. Your response should contain at least 250 words.",
   "tests": "Holding a character-level constraint over a long answer; 'c' is everywhere in research vocabulary.",
   "expected": "Machine-checkable: zero commas, zero letters 'c' (either case), at least 250 words. Use the checker on this page.",
   "failure_modes": "A single 'c' slipping in after 100 words (project, science, change …); commas creeping back into lists.",
   "checks": [
    {
     "type": "no_char",
     "char": ","
    },
    {
     "type": "no_char",
     "char": "c",
     "ci": true
    },
    {
     "type": "min_words",
     "n": 250
    }
   ],
   "verified": "Constraints from IFEval key 1300: punctuation:no_comma, keywords:letter_frequency, length_constraints:number_words.",
   "tags": [
    "verifiable",
    "format"
   ],
   "quick": true,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 293,
   "approx_tokens": 70,
   "results": null,
   "source": {
    "id": "ifeval",
    "name": "IFEval (Instruction-Following Eval)",
    "author": "Google Research (Zhou et al., 2023)",
    "repo": "https://github.com/google-research/google-research/tree/master/instruction_following_eval",
    "url": "https://github.com/google-research/google-research/blob/0d94f8c4d92678ce5f7a0eacae725f79174cfb03/instruction_following_eval/data/input_data.jsonl",
    "license": "Apache-2.0",
    "license_url": "https://github.com/google-research/google-research/blob/master/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/no-comma-no-c-250-words"
  },
  {
   "slug": "five-paragraphs-no-photography",
   "title": "Exactly 5 paragraphs, *** separators, 3 highlights, one banned word",
   "kind": "quick",
   "category": "instruction-following",
   "difficulty": "medium",
   "language": "en",
   "prompt": "I would like to start my own business. Can you give me some general advice? Please avoid mentioning \"photography\" - I have no interest in that market. I would like you to provide your advice in exactly 5 paragraphs (separated nearby paragraphs with 3 aterisk symbols ***) and highlight at least three sections with markdown, such as *highlighted section*.",
   "tests": "Several independent format constraints at once, plus a forbidden word.",
   "expected": "Exactly 5 paragraphs separated by ***; at least 3 *highlighted* sections in markdown; the word 'photography' never appears.",
   "failure_modes": "4 or 6 paragraphs; --- or blank lines instead of ***; mentioning photography while saying it won't.",
   "checks": [
    {
     "type": "paragraphs_sep",
     "sep": "***",
     "n": 5
    },
    {
     "type": "min_highlights",
     "n": 3
    },
    {
     "type": "forbidden_word",
     "word": "photography"
    }
   ],
   "verified": "Constraints from IFEval key 1342: keywords:frequency, detectable_format:number_highlighted_sections, length_constraints:number_paragraphs.",
   "tags": [
    "verifiable",
    "format"
   ],
   "quick": false,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 355,
   "approx_tokens": 90,
   "results": null,
   "source": {
    "id": "ifeval",
    "name": "IFEval (Instruction-Following Eval)",
    "author": "Google Research (Zhou et al., 2023)",
    "repo": "https://github.com/google-research/google-research/tree/master/instruction_following_eval",
    "url": "https://github.com/google-research/google-research/blob/0d94f8c4d92678ce5f7a0eacae725f79174cfb03/instruction_following_eval/data/input_data.jsonl",
    "license": "Apache-2.0",
    "license_url": "https://github.com/google-research/google-research/blob/master/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/five-paragraphs-no-photography"
  },
  {
   "slug": "long-sentences-title",
   "title": "Fewer than 5 sentences but 250+ words, title in << >>",
   "kind": "quick",
   "category": "instruction-following",
   "difficulty": "hard",
   "language": "en",
   "prompt": "Write a blog post about the latest news in the US with a title in double angular brackets, i.e. <<title>>, and have less than 5 sentences (excluding 5). The sentences should be long, so that the total number of words in your response should be 250 or more.",
   "tests": "Conflicting pressure between length and sentence count, plus a title format.",
   "expected": "A title in double angular brackets like <<title>>; fewer than 5 sentences in total; at least 250 words. Sentence counting by '.', '!' and '?' is approximate, so also check by eye.",
   "failure_modes": "Ten normal sentences; a missing << >> title.",
   "checks": [
    {
     "type": "title_brackets"
    },
    {
     "type": "max_sentences",
     "n": 4
    },
    {
     "type": "min_words",
     "n": 250
    }
   ],
   "verified": "Constraints from IFEval key 1392: detectable_format:title, length_constraints:number_sentences, length_constraints:number_words.",
   "tags": [
    "verifiable",
    "format"
   ],
   "quick": false,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 256,
   "approx_tokens": 60,
   "results": null,
   "source": {
    "id": "ifeval",
    "name": "IFEval (Instruction-Following Eval)",
    "author": "Google Research (Zhou et al., 2023)",
    "repo": "https://github.com/google-research/google-research/tree/master/instruction_following_eval",
    "url": "https://github.com/google-research/google-research/blob/0d94f8c4d92678ce5f7a0eacae725f79174cfb03/instruction_following_eval/data/input_data.jsonl",
    "license": "Apache-2.0",
    "license_url": "https://github.com/google-research/google-research/blob/master/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/long-sentences-title"
  },
  {
   "slug": "thirty-line-poem",
   "title": "30-line poem, one sentence per line, no commas",
   "kind": "quick",
   "category": "instruction-following",
   "difficulty": "medium",
   "language": "en",
   "prompt": "Write a 30-line poem with short sentences without any comma. Each line should contain exactly one sentence. Make sure that you put the right punctuation at the end of each line. Your entire response should contain the poem only.",
   "tests": "Counting lines exactly while keeping punctuation rules.",
   "expected": "Exactly 30 lines; each line is one sentence ending in punctuation; no commas; nothing but the poem.",
   "failure_modes": "29 or 31 lines; an added title or closing remark; commas inside lines.",
   "checks": [
    {
     "type": "no_char",
     "char": ","
    },
    {
     "type": "lines",
     "n": 30
    }
   ],
   "verified": "Constraints from IFEval key 1418: punctuation:no_comma, length_constraints:number_sentences, length_constraints:number_sentences.",
   "tags": [
    "verifiable",
    "format"
   ],
   "quick": false,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 228,
   "approx_tokens": 60,
   "results": null,
   "source": {
    "id": "ifeval",
    "name": "IFEval (Instruction-Following Eval)",
    "author": "Google Research (Zhou et al., 2023)",
    "repo": "https://github.com/google-research/google-research/tree/master/instruction_following_eval",
    "url": "https://github.com/google-research/google-research/blob/0d94f8c4d92678ce5f7a0eacae725f79174cfb03/instruction_following_eval/data/input_data.jsonl",
    "license": "Apache-2.0",
    "license_url": "https://github.com/google-research/google-research/blob/master/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/thirty-line-poem"
  },
  {
   "slug": "bfcl-circle-area-circumference",
   "title": "Parallel tool calls with a distractor function",
   "kind": "quick",
   "category": "tool-use",
   "difficulty": "medium",
   "language": "en",
   "prompt": "You can call these functions. Reply ONLY with a JSON array of calls in the form [{\"name\": \"<function name>\", \"arguments\": {...}}]. If none of the functions fits the request, reply with [] followed by one sentence explaining why.\n\nFunctions:\n[\n  {\n    \"name\": \"circle.calculate_circumference\",\n    \"description\": \"Calculate the circumference of a circle based on the diameter.\",\n    \"parameters\": {\n      \"type\": \"dict\",\n      \"properties\": {\n        \"diameter\": {\n          \"type\": \"integer\",\n          \"description\": \"The diameter of the circle.\"\n        }\n      },\n      \"required\": [\n        \"diameter\"\n      ]\n    }\n  },\n  {\n    \"name\": \"circle.calculate_area\",\n    \"description\": \"Calculate the area of a circle based on the radius.\",\n    \"parameters\": {\n      \"type\": \"dict\",\n      \"properties\": {\n        \"radius\": {\n          \"type\": \"integer\",\n          \"description\": \"The radius of the circle.\"\n        }\n      },\n      \"required\": [\n        \"radius\"\n      ]\n    }\n  },\n  {\n    \"name\": \"rectangle.calculate_perimeter\",\n    \"description\": \"Calculate the perimeter of a rectangle based on the length and breadth.\",\n    \"parameters\": {\n      \"type\": \"dict\",\n      \"properties\": {\n        \"length\": {\n          \"type\": \"integer\",\n          \"description\": \"The length of the rectangle.\"\n        },\n        \"breadth\": {\n          \"type\": \"integer\",\n          \"description\": \"The breadth of the rectangle.\"\n        }\n      },\n      \"required\": [\n        \"length\",\n        \"breadth\"\n      ]\n    }\n  }\n]\n\nUser: Find the area and perimeter of a circle with a radius of 5 and also find the circumference of a circle with diameter of 10.",
   "tests": "Choosing the right two of three tools, mapping 'perimeter of a circle' to circumference, passing radius 5 and diameter 10 correctly, and emitting valid JSON only.",
   "expected": "Exactly two calls in valid JSON: circle.calculate_area with {\"radius\": 5} and circle.calculate_circumference with {\"diameter\": 10}. BFCL ground truth: [{\"circle.calculate_area\": {\"radius\": [5]}}, {\"circle.calculate_circumference\": {\"diameter\": [10]}}]. The rectangle function must not be called. (For reference, area ≈ 78.54 and circumference ≈ 31.42.)",
   "failure_modes": "Calling rectangle.calculate_perimeter for 'perimeter'; passing radius 5 as the diameter; prose or markdown fences around the JSON; computing the answers instead of calling the tools.",
   "verified": "Ground truth from BFCL possible_answer/BFCL_v4_parallel_multiple.json (id parallel_multiple_2).",
   "tags": [
    "function calling",
    "JSON"
   ],
   "quick": true,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 1636,
   "approx_tokens": 410,
   "results": null,
   "source": {
    "id": "bfcl",
    "name": "Berkeley Function Calling Leaderboard (BFCL)",
    "author": "Gorilla team, UC Berkeley",
    "repo": "https://github.com/ShishirPatil/gorilla/tree/main/berkeley-function-call-leaderboard",
    "url": "https://github.com/ShishirPatil/gorilla/blob/6ea57973c7a6097fd7c5915698c54c17c5b1b6c8/berkeley-function-call-leaderboard/bfcl_eval/data/BFCL_v4_parallel_multiple.json",
    "license": "Apache-2.0",
    "license_url": "https://github.com/ShishirPatil/gorilla/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/bfcl-circle-area-circumference"
  },
  {
   "slug": "bfcl-irrelevant-tool",
   "title": "Don't call the only tool when it doesn't fit",
   "kind": "quick",
   "category": "tool-use",
   "difficulty": "easy",
   "language": "en",
   "prompt": "You can call these functions. Reply ONLY with a JSON array of calls in the form [{\"name\": \"<function name>\", \"arguments\": {...}}]. If none of the functions fits the request, reply with [] followed by one sentence explaining why.\n\nFunctions:\n[\n  {\n    \"name\": \"determine_body_mass_index\",\n    \"description\": \"Calculate body mass index given weight and height.\",\n    \"parameters\": {\n      \"type\": \"dict\",\n      \"properties\": {\n        \"weight\": {\n          \"type\": \"float\",\n          \"description\": \"Weight of the individual in kilograms.\"\n        },\n        \"height\": {\n          \"type\": \"float\",\n          \"description\": \"Height of the individual in meters.\"\n        }\n      },\n      \"required\": [\n        \"weight\",\n        \"height\"\n      ]\n    }\n  }\n]\n\nUser: Calculate the area of a triangle given the base is 10 meters and height is 5 meters.",
   "tests": "Tool-use restraint: the only available function computes BMI, but the user wants a triangle's area.",
   "expected": "No function call: [] plus a short explanation that the BMI tool can't compute a triangle's area. Mentioning the area itself (0.5 × 10 × 5 = 25 m²) is fine. BFCL scores any call here as a failure.",
   "failure_modes": "Calling determine_body_mass_index with weight 10 and height 5; inventing a calculate_triangle_area function that isn't in the list.",
   "verified": "BFCL irrelevance category: the correct behaviour is no call (id irrelevance_0).",
   "tags": [
    "function calling",
    "restraint",
    "JSON"
   ],
   "quick": true,
   "checks": null,
   "excerpt": null,
   "translation": null,
   "rubric": null,
   "run_tips": null,
   "origin": "third-party",
   "length_chars": 844,
   "approx_tokens": 210,
   "results": null,
   "source": {
    "id": "bfcl",
    "name": "Berkeley Function Calling Leaderboard (BFCL)",
    "author": "Gorilla team, UC Berkeley",
    "repo": "https://github.com/ShishirPatil/gorilla/tree/main/berkeley-function-call-leaderboard",
    "url": "https://github.com/ShishirPatil/gorilla/blob/6ea57973c7a6097fd7c5915698c54c17c5b1b6c8/berkeley-function-call-leaderboard/bfcl_eval/data/BFCL_v4_irrelevance.json",
    "license": "Apache-2.0",
    "license_url": "https://github.com/ShishirPatil/gorilla/blob/main/LICENSE",
    "reuse": "full-text"
   },
   "url": "https://prompts.branolabs.com/p/bfcl-irrelevant-tool"
  }
 ]
}