Gather evidence. Keep your bearings.
Find your way through a maze of information. Gather supported evidence, avoid the ideas that inflate confidence, and reach the answer gate when your context is ready.
Clean FAQ is the first playable chapter. A fictional community repair workshop gives the maze a question: do you need to bring tools? Source tokens stand for evidence; they are not real research sources.
How to play
Use arrow keys, WASD, or the direction buttons. Each token is collected once. Supported evidence builds evidence faster than confidence; weak context and junk can do the opposite.
The gate opens with at least 60 evidence, no more than 20 contamination, and confidence no more than 15 ahead of evidence. Collecting supported evidence always helps. There is no time bonus.
Hallu removes evidence while adding confidence and contamination. Bluster raises confidence without evidence. Each hit costs one of three stability marks, sends you to the start, and gives you two seconds of protection. Collected evidence stays with you.
At the review station, compare a relevant current FAQ with an unsupported claim. Review can reduce contamination and bring confidence back within the evidence. It does not create new evidence.
Take your time
Start with practice for a safe introduction. Space pauses while the board has focus; Escape opens the pause menu. Leaving the tab or the game pauses play, and you decide when to resume. The pause menu keeps a short record of your recent context.
Shapes and labels supplement color, controls support keyboard and pointer input, and reduced motion is respected. This version does not provide a complete nonvisual maze interaction.
These meters are a learning metaphor, not measurements of truth or AI safety. Review in this fictional game cannot guarantee real-world correctness.
Build journal and credits
This first chapter is part of REN-846's guided migration. Mike approved Clean FAQ and its local Game Room integration on September 20, 2026. The other two scenarios and deployment remain separate follow-up work.
The original Context Gobbler source is preserved with its Git revision and file hashes. Mike Besso authored the original introduction commit. The new implementation was developed with Codex assistance. The original lesson and maze concept inform this revision; the new confidence rule, touch controls, source-review station, and debrief are explicit changes.
The interface uses the shared Renaissance Solutions frame and original vector game symbols. Atkinson Hyperlegible Next and Newsreader font licenses accompany the local assets. All scenario text is authored fictional material. No live AI output or external API is required to play.