AI
From Proposal to Verified Effect: Praxa, an Evidence-Bound Harness for Governed AI Agent Execution
Publication: cs.AI updates on arXiv.org
Author: Stefan G. Creadore
Source published: 02/10/2026, 04:00:00 UTC
Briefing created: 02/10/2026, 04:45:29 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 02/10/2026, 04:03:24 UTC.
Briefings
Praxa is an evidence-bound agent harness designed to explicitly represent and test the transition from proposal to verified external effect in governed AI agents. The project reports four evidence lanes: a local audit of 1,027 unit tests and 89 Workerd tests on 363 source files; a Terminal-Bench pilot across 12 tasks where baseline and reliability-layer arms passed 17/36 trials; a post-debug comparison of two coordination proxies where both completed 180/180 trials with equal accuracy but no quality improvement; and deployed source evidence showing bounded reflection and memory compilation without production outcome lift. The core finding is that Praxa provides an architecture for explicit, testable authority-to-effect transitions, though current evidence does not establish security, safety, or superior performance.
Key points
- Praxa is an agent harness that explicitly represents states like proposal, authority, and verified external effect through deterministic admission, brokered execution, external read-back, reconciliation, and reviewed promotion.
- A local audit of Praxa's repository passed 1,027 unit tests and 89 Workerd tests on 363 source files, though raw transcripts are unavailable.
Context and limitations
The evidence is limited to a preprint on arXiv and a single Terminal-Bench pilot with 12 tasks. No production outcome lift, adversarial security, or autonomous recursive optimization has been established.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 02/10/2026, 04:45:29 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeigrhzfpajnxy7vab4vnqkyqcxnad3x5y5jj4bopqm5iqg3bro4uhq