Archive - Overfitting Dicta

Introducing Behavioral Analysis for Legal-AI

What session trajectories reveal about model behavior on twelve legal tasks

Sep 2 • Overfitting Dicta

July 2026

Investigating LAB, Part 4: RAG Impeded by the AI Refusals of Vibed Data Rooms

If Harvey plans to train agents in AI-generated RL data rooms, will those M&A diligence agents be able to identify AI refusal artifacts in their…

Jul 27 • Overfitting Dicta

Investigating LAB, part 3: Fabricated Rubric Criteria

LAB scoring rubrics include fabricated criteria possibly hallucinated during development

Jul 14 • Overfitting Dicta

Introducing a Twelve-Task Claude-for-Legal Benchmark

Defining a benchmark that tests twelve Claude-for-Legal skills from six practice areas on the Harvey LAB task that best exercises each skill

Jul 9 • Overfitting Dicta

June 2026

Investigating the LAB: Part 2, Judge Variation and the Case for Per-Criterion Scoring

How judge behavior affects the reliability and precision of legal AI assessment

Jun 9 • Overfitting Dicta

Investigating the Legal Agent Benchmark (LAB): Part 1

LAB should be studied as a benchmark, not merely consulted as a leaderboard

Jun 2 • Overfitting Dicta

May 2026

Benchmarking Claude for Legal

Testing Anthropic’s Claude for Legal toolkit against a litigation chronology benchmark

May 22 • Overfitting Dicta

#nojs-banner { position: fixed; bottom: 0; left: 0; padding: 16px 16px 16px 32px; width: 100%; box-sizing: border-box; background: red; color: white; font-family: -apple-system, "Segoe UI", Roboto, Helvetica, Arial, sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol"; font-size: 13px; line-height: 13px; } #nojs-banner a { color: inherit; text-decoration: underline; } This site requires JavaScript to run correctly. Please turn on JavaScript or unblock scripts