From c2735fa919f6cad0a8ac55036312718533a7e059 Mon Sep 17 00:00:00 2001 From: davegilligan Date: Mon, 25 May 2026 21:27:34 +0200 Subject: [PATCH] Simplify advocate engine options: remove GPU/dbn_legal_v3, fix time estimates GPU and DBN Legal Agent v3 are unsuitable for advocate synthesis (4-6K token structured JSON output at 20-30 tok/s = 3-5 min on RTX 3060, plus both fall through to Sonnet when Bedrock is enabled anyway). Reduce to two honest options: Haiku (~2-4 min) and Sonnet (~3-5 min), with accurate description of why time is dominated by multi-pass question answering, not synthesis. Co-Authored-By: Claude Sonnet 4.6 --- advocate.php | 9 +++------ 1 file changed, 3 insertions(+), 6 deletions(-) diff --git a/advocate.php b/advocate.php index 6171ebd..a73a7a0 100644 --- a/advocate.php +++ b/advocate.php @@ -52,13 +52,10 @@ require_once __DIR__ . '/includes/layout.php';
Engine - - - - - + +
-

Claude Haiku 4.5 is fast and accurate for most cases. Claude Sonnet 4.6 produces the most thorough advocate brief — superior at multi-party legal reasoning, ECHR precedent weighting, and long-form argumentation. Norwegian specialist v3 is a Qwen2.5 fine-tune trained on barnevernsloven, ECHR, and forvaltningsloven — highest precision for § 4-25, Strand Lobben, and procedural red flags.

+

Both engines run on AWS Bedrock via Claude. Most of the time is spent on multiple question-answering passes — 6–10 sub-questions each requiring a full retrieval and answer cycle. Haiku is faster and handles most cases well. Sonnet produces a more thorough brief with deeper ECHR precedent analysis and stronger multi-party argumentation.

Corpus slices