Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2607.25566.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

ARCHER is a test-driven, deterministically orchestrated multi-agent harness that synthesizes auditable verification code from building regulations. Across six harnesses and four backbone models, it improves mean union accuracy by 82% over single-pass prompting and reaches 97.8% of frontier accuracy on a self-hosted open-weights model at a quarter of the cost.

Concrete empirical comparison of six multi-agent orchestration patterns on a real-world code-synthesis task, with cost-accuracy data across data-governance tiers.

Cited by

No citations on record.