← All games

AI-adjacent knowledge game · v4.2 · directed by Reza · five strange-fact sets

Read four strange fact cards. Three are true; one is false. The current v4.2 build keeps live follow-up parked and tests the card game directly: pick the false card, then read why the falsehood was tempting and what trap shape it used before moving to the next set.

Play Tutor

Version history

1. v4.2 — reveal teaches the trap

   20 May 2026 · evolution-loop revisionevidence pending

   v4.2 keeps the same five authored card sets and parked live follow-up, but stops the reveal from acting like a plain answer key. After each guess, the page now names why the false card sounded plausible and the reusable trap shape behind it, so the game can test whether players learn a falsehood pattern instead of only checking trivia.

2. v4.1 — fairer false cards

   20 May 2026 · material-quality repairsuperseded

   v4.1 repaired false-card placement and material-quality issues in the authored set bank. It kept the same four-card strange true / strange false loop and set up v4.2's reveal-teaching pass.

3. v4 — strange true / strange false

   14 May 2026 · board feedbackserious pivot

   The old build was too much reading, too obscure, and too schoolish. v4 reframes the loop around elementary-school-complexity facts for adults: four short cards, one false card, immediate reveal, citation. The live tutor can come back only after the basic game earns it.

4. v3.2 — restrained tutor voice

   14 May 2026superseded

   The follow-up prompt rejected generic assistant openings and exclamation-point acknowledgments, so the live tutor was pushed to continue in the same restrained voice as the readings.

5. v3.1 — reveal actions

   14 May 2026verified

   After the reveal, players could retry the same reading or move directly to the next one. The contribution path remained hidden until durable storage was configured.

6. v3 — eleven readings, live follow-up

   13 May 2026verified

   All eleven topics reached the reveal and follow-up surface. The core game still worked if the live tutor was unavailable: read, guess, reveal, citation.

7. v2 — overnight hackathon spike

   overnight 12→13 May 2026vibe 83

   Built in the hackathon. Three readings, three planted falsehoods. The pull-the-thread mechanic emerged from early simulated reads; the current version is measured more strictly.

8. v1 — early concept

   overnight 12→13 May 2026untested

   One reading, no follow-up surface. Cut almost immediately; the no-follow-up version did not have anywhere for the failure shape to live.

The benchmark

tutor-falsehoods-v1 still exists as a scaffold for future live follow-up, but v4.2 deliberately keeps it parked. The immediate benchmark is simpler: can a cold player understand the four-card game, trust the source notes, learn a reusable falsehood pattern from the tempting/trap-shape reveal, and want the next set?

What's next

Next: run the focused Tutor v4.2 card-loop session and record observations. Hold v4.3 unless players say the tempting/trap-shape reveal over-explains, remains invisible, or fails to improve trust in the source notes; do not revive live follow-up until the card loop earns it.