Agentic Context Sharing for Constraint Satisfaction Problems: solving a Murdoku across evidence troves

documents → extracted graph → clues → solution
How to read this
  1. Open the Spec (top-right) for the write-up behind this tool.
  2. Select evidence troves on the left and watch the graph and grid evolve.
  3. Click an evidence document to see how entities and clues are extracted from its text.
  4. Read the clue breakdown on the right — efficient / redundant / red herrings.
  5. With all troves selected, click each suspect to place it (or Show solution) and reveal the victim.
  6. Turn the density knob (bottom-left) up: distractor documents pile in, but the extracted graph stays clean — it isolates the answer from the noise.

Evidence troves

density N (distractor docs / trove)0