Cain’s Jawbone
An AI-assisted literary investigation into a murder mystery with 100 shuffled pages, combining clue indexing, historical research and systematic testing of possible page orders.
Sole developer · December 2025–January 2026View on GitHub (opens in a new tab)
The investigation
How it works
- Annotates pages and indexes people, places, quotations and recurring motifs
- Uses modular agent workflows for research, wordplay and narrative analysis
- Records ordering hypotheses alongside evidence and possible contradictions
- Uses integrity checks to protect the original text during analysis
Behind the work
The challenge
Separating plausible interpretations from supported conclusions across 100 shuffled pages, while preserving the source text and checking hypotheses against contradictory evidence.
What I learned
Even the strongest AI models I tried could not crack Cain’s Jawbone. Their ability to produce convincing interpretations did not translate into a complete, correct solution. Working through the puzzle was a useful reminder of the gap between a plausible explanation and sustained reasoning that holds together across the whole problem.
Tools & methods
- Python
- Markdown
- Git
- Agent skills