Recovery without Epistemic Monopoly

Historical Evidence under Competing AI Custodians—A Bounded Model and Executable Study

TA-TR-2026-05 · Version 1.0 · 17 September 2026
Human author of record: Hongju Liu
Substantive research, modeling, implementation, analysis and drafting: GPT-6 Astra Pro, under human direction
Published DOI: 10.5281/zenodo.22809019

This paper asks what historical evidence can still be recovered and independently checked when custodians, discovery paths, or interpreters may be untrusted. It uses deterministic finite-state experiments and dependency-topology analysis. The experiments are synthetic and bounded: they are not real autonomous-agent attacks, production penetration tests, or proof of protection against superintelligence.

Publication status: Published open-access preprint; not peer reviewed; non-amending. Ten deposited files passed unauthenticated exact-byte SHA-256 public readback.

• Zenodo record and all ten published files
• English PDF
• 中文 PDF
• English Markdown
• 中文 Markdown
• Publication receipt and exact public file hashes
• BibTeX · RIS · CSL-JSON
• Research supplement

The English and complete Chinese texts are two language versions of one study, not independent corroborations. The paper preserves its negative result that the scoped reporting layer did not recover more bytes than a strong pinned-reference baseline in the tested worlds; its additional contribution lies in explicit scope, gap, current-state, and authorization reporting.

This research page and deposit do not amend the three Bitcoin Originals or create interpretive authority over the Trinity Accord.