What question did this study set out to answer?

This research aims to introduce a streamlined framework for making interpretive claims about ancient languages auditable and verifiable.

February 10, 2026Open Access

Falsifiability Sheet v5 (VAI2) with Ramesses II Sample

Key Points

This research aims to introduce a streamlined framework for making interpretive claims about ancient languages auditable and verifiable.
Developed a Falsifiability Sheet to document interpretive claims and their context.
Employs AI as an advisory tool rather than an authoritative source, recording its output with risk codes.
Includes peer checks to foster structured outcomes and mitigate disputes.
Demonstrates the framework using a known example of Ramesses II for calibration and training purposes.
Establishes a repeatable workflow that enhances the credibility of interpretive claims.
Promotes clear documentation of evidence types and confidence ratings to prevent misinterpretations.
Facilitates DOI-linked publishing and promotes the reuse of interpretations in community verification.

Abstract

AI-assisted interpretation is now moving faster than verification norms in epigraphy, ancient-language study, and heritage research. This paper introduces the Collaboration-First AI Falsifiability Framework: a lightweight, repeatable workflow that makes interpretive claims auditable at the moment they are made. The core deliverable is the Falsifiability Sheet (v5.vai2)—a one-page claim ledger that records (1) a narrowly bounded claim, (2) object identification and context (CISI, provenance, period, stable DOI/URL), (3) the key structural decisions that shape any reading (e.g., ORI orientation/reading direction, and G/OVR mode: glyph-first vs. overlay/alignment when applicable), (4) the evidence types actually consulted, (5) component-level confidence ratings that prevent weak links from hiding inside a strong narrative, and (6) two independent peer checks (R1/R2) that convert disagreement into structured outcomes rather than informal debate. The framework explicitly treats AI as advisory, logged input—not authority: AI output is recorded in a dedicated block and labeled with a traffic-light risk code (G/Y/R), while confidence can only rise through cited evidence and peer outcomes. A worked already-deciphered baseline example (Ramesses II) demonstrates calibration and reviewer training before applying the workflow to contested scripts (e.g., Indus). The result is a practical pathway for DOI-linked publishing, reuse, and community verification—starting with human-readable PDFs and extending (optionally) to machine-readable records for indexing at scale. If you’re interested in supporting this work or partnering on pilots, you can view the 501(c)(3) fiscal sponsorship page for Echoes of the Script here:

Read Full Paperexternally

Ask AI

Helpful

Bookmark

View Full Paper