Synapse
⌘+K
Synapse
PulseExploreJournal ClubResearchersJournals
Instagram
HomeJournal ClubExplore
August 15, 2026Open Access

Auditable AI-Assisted Research Writing: An Engineering Discipline with Pre-Registered Process Observation - audit package

View Full Paper
Ask AI
Bookmark
Share

Authors

YZYang ZhouCYChengqun Yu

Discussion

Loading...

Member takes

Overview

Observational study demonstrates prospective process auditing for AI-assisted scientific writing, highlighting how hash-bound provenance and pre-registered stopping rules enforce research integrity.

Key Points

  • To establish and evaluate an engineering discipline and audit package for AI-assisted research writing that embeds verifiable, hash-bound provenance and adherence checks directly into scientific production.
  • Designed an audit protocol featuring git sealing, hash-bound provenance, refusal-logging gates, cross-model role separation, and scripted assembly under frozen pre-registered rules.
  • Evaluated the discipline across a prospective case study and a lower-standing retrospective case, releasing a standalone verification package containing 55 registered objects and audit ledgers.
  • Built a self-contained Python verifier that reconstructs sealed inventories, verifies digests, and recomputes nine primary metrics without external network or library dependencies.
  • Demonstrated prospective rule enforcement when an observed project's confirmatory test executed under seal returned a No-Go result, successfully halting the workflow via pre-frozen stopping rules.
  • Achieved full deterministic recomputability for all nine pre-registered primary metrics and measurement rows across the audit snapshot using standard cryptographic verification.

Cite This Study

Zhou et al. (2026) studied this question.

synapsesocial.com/papers/6a80197375c2e31742c8578dhttps://doi.org/10.5281/zenodo.21905683
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1Enrollment is not coverage: integrity gates, their measured detection rates, and the limits of self-checking documentation2026
  2. 2Enrollment is not coverage: integrity gates, their measured detection rates, and the limits of self-checking documentation2026
  3. 3Plan Pointers and Record-Directive Form in Budgeted Verification of Inherited Agent Memory2026
  4. 4Who Validates the Validator? Instrument failure in the shape of the hypothesis — fourteen audit findings from a pre-registered measurement2026
  5. 5Who Validates the Validator? Instrument failure in the shape of the hypothesis — a pre-registered measurement programme with a public correction ledger2026