How are Human or AI questions made?

What the labels mean

Human: a short sample drawn from a linked Hacker News comment attributed to its displayed username. The comment records used here were published in 2015 or 2020. On October 18, 2026 we checked the linked records' author names, comment type and publication years through the public Hacker News API. Public attribution and an early publication date are context, not independent proof of a real-world identity.

AI: a synthetic passage written for the set and labelled as such. It is paired editorially for topic, length and informal forum style. The game does not claim to identify an unknown model or represent all human and AI writing.

Follow the original comment link to inspect the source in its full context. The source list below is an attribution index, not an independent authorship certificate.

How a round is chosen

• Classic contains 20 human-labelled and 20 AI-labelled passages. The 2020 Mix contains 12 of each; Full Pool combines them.
• A round selects four from each side and shuffles all eight. There is no duplicate passage within the round.
• New random rounds try different seeds to prefer passages outside the last eight shown on the current browser. The pool is finite and repeats can still occur. Local history can be cleared on the homepage.
• Daily challenges use a UTC date to choose a fixed round. A shared challenge link recreates its fixed round and order, ignoring the recipient's local history.
• Local history is stored on the device, not in a player account. If storage is blocked, the current tab can still keep recent passages until it closes.

How scoring works

Matching the prepared label earns one point. Each round has eight possible points, with no speed bonus. A score describes that round only. Replaying a fixed link can make its answers familiar, so scores and the device leaderboard are not verified rankings or research results.

What the other practice games contain

Memory Check, Story Check, Correction Check and Typo Trap use prepared fictional situations to ask what evidence supports. Their characters and dialogue are not quotations from human participants. They award points for the supported reasoning choice, not for discovering a real person's identity. Each page explains its own set and scoring.

Try Memory Check → · Try Story Check →

Limitations worth knowing

• The human-labelled samples come from one technical community and two publication years; they do not represent every audience or writing style.
• AI samples are deliberately prepared for this game, rather than collected as a representative model benchmark.
• Remembering a detail, making a typo or writing cleanly cannot by itself establish identity.
• This page is not a detector. Nothing here validates high-stakes decisions about someone else's writing.

Human-labelled source index

Metadata checked October 18, 2026 using the official Hacker News API. The username and year below are the public record values. We do not reproduce the full comments here.

Change notes and corrections

October 12, 2026: prepared question set expanded with the 2020 Mix. October 17: local scores and same-round sharing added, followed by daily challenges. October 18: local recent-passage preference and this source index added.

Question text and the existing challenge shuffle were retained in this update. Metadata or code changes must not silently change what an old challenge link presents. If you find an attribution or wording problem, send the source URL and the passage to our contact page; corrections should be documented.