Tech Meets Law

A narrow, auditable legal AI study.

This page summarizes one doctrine-specific research track: how factual framing shifts geometric positioning in a legal precedent corpus.

Scope is intentionally narrow and conservative. We report measured association, not legal outcome prediction.

Study Snapshot

Key numbers from the current run

The project computes a winning-proximity score (WPS) and a leave-one-out robustness variant (WPS_LOO). Both metrics show the same directional pattern.

Interpretation

What this does and does not mean

• It does: quantify narrative-position shifts within a defined doctrine corpus.
• It does: support draft comparison for legal fact sections.
• It does not: predict judicial outcomes.
• It does not: prove causality from language framing to case results.

Visuals

How to read the charts

In the narrative drift chart, the vertical lines are single-point scores for one case rewritten multiple ways. They are not distributions, so they naturally cluster.

For a simpler read, use three briefing charts: ranked variant WPS, outcome box plot, and WPS vs WPS_LOO effect comparison.

We use those three views in investor and technical briefings because they are easier to interpret than the dense single-case vertical-line plot.

Safe Route

Conservative product path

• Phase 1: keep doctrine-specific scoring only; no cross-doctrine claims.
• Phase 2: expose a JSON scoring endpoint for draft comparison workflows.
• Phase 3: add multi-doctrine expansion only after separate validation.
• Always: describe results as positioning signals, not legal advice or prediction.

Next

Technical packet and implementation notes

The full technical summary and verification checklist are available in the research packet, along with the rollout plan for a JSON scoring hook.