Measuring what humanity knows, and what we're losing
The Deep Time Research Institute is a New York nonprofit corporation conducting cross-domain research on cultural knowledge systems: measuring the accuracy of oral traditions, mapping knowledge extinction in endangered languages, and building open-access tools for preservation triage.
Using AI as core research infrastructure, DTRI has a portfolio of manuscripts in peer review, additional papers in preparation, and several submissions held for AIATSIS-equivalent cultural consultation before publication. See the papers page for the current list.
The flagship finding: oral tradition accuracy across 41 knowledge domains is governed by a measurable environmental variable, the observability gradient (r = 0.527, p = 0.0004). Above a threshold, cultural selection maintains accuracy indefinitely. Below it, traditions drift toward cognitive attractors. 73 Prolific raters scored 41 blinded vignettes with high agreement (ICC = 0.97); rater-consensus observability replicated the gradient (Spearman r = 0.50). The single-variable gradient has since been decomposed into the five-component External Referent Constraint (ERC) framework; the gradient remains the original empirical hook.
All preprints, data, and code are open access.
DTRI is an explicitly AI-native research organization. Claude Code serves as the primary computation engine for statistical analysis, Monte Carlo simulations, and data pipelines. Every number in every paper is computed from raw data via code, never from memory or training data.
Every paper runs through a multi-model adversarial review gauntlet (Claude, Perplexity, Gemini, Grok) before submission. The methodology, human-directed agentic AI executing cross-domain research, is itself part of the contribution.
Whether the work holds up is for peer review to decide. That's the whole point of submitting.
Elliot Allan — Founder & Director, Deep Time Research Institute. Quantitative archaeology, cultural evolution, and empirical analysis of oral traditions.
DTRI is structured as a small, independent research program. The Institute is US-based, AI-native by design, and operates with all research outputs open access. Its work is built to be auditable: every statistical claim is paired with the code and data that produced it, every paper is paired with a Zenodo deposit, and methodological choices — including AI use — are documented in the methods page.
Contact: elliot@deeptime-research.org
DTRI is a New York nonprofit corporation. All research outputs are open access. If you'd like to support this work, contact us directly.