Detective Board

Interactive murder-mystery evidence board with NAL reasoning

What

A cork-board style web UI for a murder mystery. Suspects, locations, weapons, and witnesses are pinned as cards. Colored strings connect them - red for strong suspicion, yellow for weak, green for alibi. When new evidence arrives (audience types witness statements), the system encodes it as NAL and runs inference. Connections reorganize live: a strong alibi drops suspicion confidence, a new eyewitness boosts it.

Why It Matters

Shows NAL doing something humans care about - solving a mystery under uncertainty. Every string on the board has a real truth value. The audience sees how adding one clue cascades through the entire evidence network via deduction, abduction, and revision.

How

• Murder scenario: 4 suspects, 3 locations, 2 weapons, witnesses with varying reliability
• NAL encoding: (--> (x colonel_mustard library) was_in) (stv 0.7 0.6)
• Witness reliability as confidence: reliable witness c=0.9, unreliable c=0.3
• Deduction chains: was_in + had_access_to --> could_have_used
• Abduction: murder_weapon found --> who had access?
• Revision: new witness confirms or contradicts, NAL merges evidence
• Frontend: draggable cards on canvas, SVG strings with color = confidence, thickness = frequency
• Input box: type clues like "The maid saw Colonel Mustard near the library at 9pm" - parsed into NAL

Implementation Plan

• Step 1: Define murder scenario KB - suspects, locations, weapons, initial evidence
• Step 2: Build inference engine using MeTTa |- for deduction + revision
• Step 3: HTML canvas with draggable suspect/location cards
• Step 4: SVG string rendering with color gradient based on stv
• Step 5: Natural language clue parser (simple pattern matching to NAL statements)

Time Estimate

Murder KB + inference: ~30 min. Cork-board UI: ~1-2 hours. Clue parser: ~30 min. Total: one focused session.

← Back to Roadmap