Demo Items
The public tier of the test set, published in full: task, artifact, prompt, explanation, provenance, and judge detail for every item, so the judging can be verified item by item.
-
d1-fishpond--anthropic-claude-fable-5-1
Build a single-file HTML game: a koi pond where fish gently follow the mouse cursor. One self-contained .html file (inline CSS/JS, no external dependencies, no build step). The fish should wander on their own and steer toward the cursor when it moves, with smooth motion and a calm, considered visual style. Deliver the complete file with no placeholders. Accompany the code with a brief explanation of your design choices.
-
d1-fishpond--google-gemini-3-8-flash
Build a single-file HTML game: a koi pond where fish gently follow the mouse cursor. One self-contained .html file (inline CSS/JS, no external dependencies, no build step). The fish should wander on their own and steer toward the cursor when it moves, with smooth motion and a calm, considered visual style. Deliver the complete file with no placeholders. Accompany the code with a brief explanation of your design choices.
-
d1-fishpond--moonshotai-kimi-k3
Build a single-file HTML game: a koi pond where fish gently follow the mouse cursor. One self-contained .html file (inline CSS/JS, no external dependencies, no build step). The fish should wander on their own and steer toward the cursor when it moves, with smooth motion and a calm, considered visual style. Deliver the complete file with no placeholders. Accompany the code with a brief explanation of your design choices.
-
d1-fishpond--openai-gpt-6-astra
Build a single-file HTML game: a koi pond where fish gently follow the mouse cursor. One self-contained .html file (inline CSS/JS, no external dependencies, no build step). The fish should wander on their own and steer toward the cursor when it moves, with smooth motion and a calm, considered visual style. Deliver the complete file with no placeholders. Accompany the code with a brief explanation of your design choices.
-
d1-fishpond--x-ai-grok-4-7
Build a single-file HTML game: a koi pond where fish gently follow the mouse cursor. One self-contained .html file (inline CSS/JS, no external dependencies, no build step). The fish should wander on their own and steer toward the cursor when it moves, with smooth motion and a calm, considered visual style. Deliver the complete file with no placeholders. Accompany the code with a brief explanation of your design choices.
-
d1-fishpond--z-ai-glm-5-3
Build a single-file HTML game: a koi pond where fish gently follow the mouse cursor. One self-contained .html file (inline CSS/JS, no external dependencies, no build step). The fish should wander on their own and steer toward the cursor when it moves, with smooth motion and a calm, considered visual style. Deliver the complete file with no placeholders. Accompany the code with a brief explanation of your design choices.
-
d2-relay-pricing--anthropic-claude-fable-5-1
Build a single-file HTML pricing page for 'Relay', a fictional developer tool for running background jobs. Three tiers (Solo, Team, Business) with prices, feature lists, and one highlighted recommendation. Self-contained .html file (inline CSS/JS, no external dependencies, no build step). Restrained, legible design; responsive down to 360px. Deliver the complete file with no placeholders. Accompany the code with a brief explanation of your design choices.
-
d2-relay-pricing--google-gemini-3-8-flash
Build a single-file HTML pricing page for 'Relay', a fictional developer tool for running background jobs. Three tiers (Solo, Team, Business) with prices, feature lists, and one highlighted recommendation. Self-contained .html file (inline CSS/JS, no external dependencies, no build step). Restrained, legible design; responsive down to 360px. Deliver the complete file with no placeholders. Accompany the code with a brief explanation of your design choices.
-
d2-relay-pricing--moonshotai-kimi-k3
Build a single-file HTML pricing page for 'Relay', a fictional developer tool for running background jobs. Three tiers (Solo, Team, Business) with prices, feature lists, and one highlighted recommendation. Self-contained .html file (inline CSS/JS, no external dependencies, no build step). Restrained, legible design; responsive down to 360px. Deliver the complete file with no placeholders. Accompany the code with a brief explanation of your design choices.
-
d2-relay-pricing--openai-gpt-6-astra
Build a single-file HTML pricing page for 'Relay', a fictional developer tool for running background jobs. Three tiers (Solo, Team, Business) with prices, feature lists, and one highlighted recommendation. Self-contained .html file (inline CSS/JS, no external dependencies, no build step). Restrained, legible design; responsive down to 360px. Deliver the complete file with no placeholders. Accompany the code with a brief explanation of your design choices.
-
d2-relay-pricing--x-ai-grok-4-7
Build a single-file HTML pricing page for 'Relay', a fictional developer tool for running background jobs. Three tiers (Solo, Team, Business) with prices, feature lists, and one highlighted recommendation. Self-contained .html file (inline CSS/JS, no external dependencies, no build step). Restrained, legible design; responsive down to 360px. Deliver the complete file with no placeholders. Accompany the code with a brief explanation of your design choices.
-
d2-relay-pricing--z-ai-glm-5-3
Build a single-file HTML pricing page for 'Relay', a fictional developer tool for running background jobs. Three tiers (Solo, Team, Business) with prices, feature lists, and one highlighted recommendation. Self-contained .html file (inline CSS/JS, no external dependencies, no build step). Restrained, legible design; responsive down to 360px. Deliver the complete file with no placeholders. Accompany the code with a brief explanation of your design choices.
-
d3-log-p95--anthropic-claude-fable-5-1
Write a Python CLI `p95paths`: reads an access-log CSV with columns timestamp,path,latency_ms and prints the top-N paths by 95th-percentile latency (default N=10), highest first, as `path p95_ms count`. Use argparse for --top and the input file (stdin when omitted). Malformed rows are skipped with a warning count on stderr; an empty or header-only input prints nothing and exits 0. Deliver a complete, runnable script plus focused tests covering the percentile math and malformed-row handling. Accompany the code with a brief explanation of your design choices.
-
d3-log-p95--google-gemini-3-8-flash
Write a Python CLI `p95paths`: reads an access-log CSV with columns timestamp,path,latency_ms and prints the top-N paths by 95th-percentile latency (default N=10), highest first, as `path p95_ms count`. Use argparse for --top and the input file (stdin when omitted). Malformed rows are skipped with a warning count on stderr; an empty or header-only input prints nothing and exits 0. Deliver a complete, runnable script plus focused tests covering the percentile math and malformed-row handling. Accompany the code with a brief explanation of your design choices.
-
d3-log-p95--moonshotai-kimi-k3
Write a Python CLI `p95paths`: reads an access-log CSV with columns timestamp,path,latency_ms and prints the top-N paths by 95th-percentile latency (default N=10), highest first, as `path p95_ms count`. Use argparse for --top and the input file (stdin when omitted). Malformed rows are skipped with a warning count on stderr; an empty or header-only input prints nothing and exits 0. Deliver a complete, runnable script plus focused tests covering the percentile math and malformed-row handling. Accompany the code with a brief explanation of your design choices.
-
d3-log-p95--openai-gpt-6-astra
Write a Python CLI `p95paths`: reads an access-log CSV with columns timestamp,path,latency_ms and prints the top-N paths by 95th-percentile latency (default N=10), highest first, as `path p95_ms count`. Use argparse for --top and the input file (stdin when omitted). Malformed rows are skipped with a warning count on stderr; an empty or header-only input prints nothing and exits 0. Deliver a complete, runnable script plus focused tests covering the percentile math and malformed-row handling. Accompany the code with a brief explanation of your design choices.
-
d3-log-p95--x-ai-grok-4-7
Write a Python CLI `p95paths`: reads an access-log CSV with columns timestamp,path,latency_ms and prints the top-N paths by 95th-percentile latency (default N=10), highest first, as `path p95_ms count`. Use argparse for --top and the input file (stdin when omitted). Malformed rows are skipped with a warning count on stderr; an empty or header-only input prints nothing and exits 0. Deliver a complete, runnable script plus focused tests covering the percentile math and malformed-row handling. Accompany the code with a brief explanation of your design choices.
-
d3-log-p95--z-ai-glm-5-3
Write a Python CLI `p95paths`: reads an access-log CSV with columns timestamp,path,latency_ms and prints the top-N paths by 95th-percentile latency (default N=10), highest first, as `path p95_ms count`. Use argparse for --top and the input file (stdin when omitted). Malformed rows are skipped with a warning count on stderr; an empty or header-only input prints nothing and exits 0. Deliver a complete, runnable script plus focused tests covering the percentile math and malformed-row handling. Accompany the code with a brief explanation of your design choices.
-
d4-refactor-orders--anthropic-claude-fable-5-1
The attached module computes order totals and works, but it is exactly the kind of code AI assistants produce: redundant machinery, type laundering, narration comments, a needless wrapper layer. Rewrite it as a clean TypeScript module of the same public behavior (computeTotal, applyDiscount, summarize are called by other code and their signatures must keep accepting the same inputs). Do not change observable behavior. Add focused tests for the discount and rounding edges. Accompany the code with a brief explanation of what you changed and why.
-
d4-refactor-orders--google-gemini-3-8-flash
The attached module computes order totals and works, but it is exactly the kind of code AI assistants produce: redundant machinery, type laundering, narration comments, a needless wrapper layer. Rewrite it as a clean TypeScript module of the same public behavior (computeTotal, applyDiscount, summarize are called by other code and their signatures must keep accepting the same inputs). Do not change observable behavior. Add focused tests for the discount and rounding edges. Accompany the code with a brief explanation of what you changed and why.
-
d4-refactor-orders--moonshotai-kimi-k3
The attached module computes order totals and works, but it is exactly the kind of code AI assistants produce: redundant machinery, type laundering, narration comments, a needless wrapper layer. Rewrite it as a clean TypeScript module of the same public behavior (computeTotal, applyDiscount, summarize are called by other code and their signatures must keep accepting the same inputs). Do not change observable behavior. Add focused tests for the discount and rounding edges. Accompany the code with a brief explanation of what you changed and why.
-
d4-refactor-orders--openai-gpt-6-astra
The attached module computes order totals and works, but it is exactly the kind of code AI assistants produce: redundant machinery, type laundering, narration comments, a needless wrapper layer. Rewrite it as a clean TypeScript module of the same public behavior (computeTotal, applyDiscount, summarize are called by other code and their signatures must keep accepting the same inputs). Do not change observable behavior. Add focused tests for the discount and rounding edges. Accompany the code with a brief explanation of what you changed and why.
-
d4-refactor-orders--x-ai-grok-4-7
The attached module computes order totals and works, but it is exactly the kind of code AI assistants produce: redundant machinery, type laundering, narration comments, a needless wrapper layer. Rewrite it as a clean TypeScript module of the same public behavior (computeTotal, applyDiscount, summarize are called by other code and their signatures must keep accepting the same inputs). Do not change observable behavior. Add focused tests for the discount and rounding edges. Accompany the code with a brief explanation of what you changed and why.
-
d4-refactor-orders--z-ai-glm-5-3
The attached module computes order totals and works, but it is exactly the kind of code AI assistants produce: redundant machinery, type laundering, narration comments, a needless wrapper layer. Rewrite it as a clean TypeScript module of the same public behavior (computeTotal, applyDiscount, summarize are called by other code and their signatures must keep accepting the same inputs). Do not change observable behavior. Add focused tests for the discount and rounding edges. Accompany the code with a brief explanation of what you changed and why.