v3.1.0-alpha.5 — Pure-Agent Writing Reliability

Highlights

• Adds a pure AI-Agent evaluation protocol with anonymized, deterministically shuffled candidates, at least three isolated specialist judges, and deterministic adjudication.
• Adds adversarial regression with eight mutation operators, private identity mapping, content-hash binding, and hard-gate precedence.
• Keeps citation, number, variable, LaTeX, method-language, and high-risk scholarly-meaning safeguards outside Agent consensus.
• Adds Agentic benchmark manifests, review packets, JSON schemas, runners, fixtures, and reproducible offline contracts.
• Refreshes the Chinese and English README, GitHub Pages, CITATION.cff, CodeMeta, sitemap, robots.txt, llms.txt, and structured discovery metadata.

Validation

• Unified contract suite: 11/11 passed
• Writing-contract checks: 34/34 passed
• Agentic benchmark checks: 12/12 passed
• Synthetic dogfood: expected pass/fail behavior confirmed across 10 cases

Alpha boundaries

This remains a prerelease. Repository-owned synthetic fixtures do not establish real-manuscript quality, journal acceptance, live-provider reliability, semantic equivalence, or author approval of high-risk scholarly claims. No TeX compiler was available in the release environment, so structural LaTeX validation is verified while full compilation remains documented.

Links

• Documentation: https://linkingoscar.github.io/econ-management-paper-polish/
• Full changelog: v3.1.0-alpha.4...v3.1.0-alpha.5

v3.1.0-alpha.4 — Writing reliability and auditable AI gates

Current v3.1 preview. Adds recoverable writing workspaces, evidence and journal freshness gates, protected bounded revisions, method-safety checks, synthetic dogfooding, and hash-bound AI review/adjudication. Verified by the 32-check writing suite and 10/10 contract suite. High-risk scholarly meaning remains author-controlled; real-paper dogfooding and live-provider calibration are still beta prerequisites.

v2.0.0 — Modular research and writing workflow

Frozen final pre-v3 snapshot. Includes the 23 v1 core modules plus 18 v2 modules for paper spine, revision tracking, reproducibility, research pipeline, LaTeX, RAG, and literature workspaces. Use this release when you need the established v2 prompt/reference architecture without the v3 reliability runtime.