Ancient knowledge is more accurate than you think. And it's disappearing faster than you know.
See what we're finding ↓San trackers score 98% accuracy. Andean farmers predict El Niño by watching stars. Aboriginal Australians remember volcanic eruptions from 37,000 years ago. We measured why, across 41 knowledge domains and 39 cultures.
r = 0.527 across 41 domains (p = 0.0004) · r = 0.893 blind-scored (n=7, p = 0.007) · 73 Prolific raters replicated independently (ICC = 0.97)
Explore the gradient → View research →Around 9 languages die every year. Each one may carry environmental knowledge (fire management, navigation, medicine, flood prediction) that exists nowhere else. We built a tool to map what's at risk.
~150 endangered languages mapped · 75% of medicinal plant knowledge exists in only one language
Open the dashboard →What started as a statistical test of a fringe claim became a window into how geography shaped civilization. 8 databases, ~259,000 unique sites (over 550,000 database entries); 180,000 sites in the spatial-test subset. The alignment is real; the explanation is geological, not mystical.
Monument enrichment: 2.52× (Z = 6.74) · Settlement anti-clustering: Z = −2.91
Explore the data →If culture is a prediction machine, maintaining accuracy where the environment provides feedback, drifting where it doesn't, then knowledge extinction isn't just loss of information. It's the destruction of calibrated systems that took millennia to build.
Fire convergence paper on AIATSIS ethics hold · Psychoactive paper preparing ERA submission · BBS framework v2 ready (gated on Paper 2B acceptance)
View program →Every two months, a language dies. With it goes knowledge that may exist nowhere else: not in any database, not in any library, not in any other language.
75% of medicinal plant knowledge is unique to a single language (Camara-Leret & Bascompte 2021, PNAS). But it's not just medicine. Fire management techniques that ecologists are now reintroducing. Navigation systems that Western science didn't formally describe until the 20th century. Agricultural calendars tuned to local climate patterns over centuries.
Our research shows that the most accurate knowledge, traditions maintained by continuous environmental feedback, is also the most tightly bound to specific languages, landscapes, and communities. You can't relocate it. You can't archive it in a book. The process IS the knowledge.
The Knowledge Extinction Dashboard maps which endangered languages carry which types of knowledge, scored by the likelihood that the knowledge is empirically accurate. It's a triage tool, helping researchers and preservation efforts focus where the loss is greatest.
The variable that separates 98% accuracy from pure chance. Explore it across 41 domains.
41 knowledge domains. 39 cultural systems. Six continents. A single variable predicts whether a tradition maintains accuracy for millennia or decays to chance within centuries.
Explore the gradient →70,000 years of human knowledge, encoded in song
A scrollytelling journey from the Out of Africa migration through Aboriginal songlines to the scientific verification of Indigenous oral traditions, mapped across the globe in real time.
Begin the journey →Eleven indigenous cultures, five continents, seven drug classes. All independently built ceremonies matching pharmacokinetic duration profiles.
Read on Substack →Why some traditions are stunningly accurate and others are no better than guessing, and what that reveals about how human knowledge actually works.
Read on Substack →Pre-registered predictions about Torres Strait stepping stones confirmed at 30-meter resolution. The observability gradient explains why some traditions survive millennia while others decay.
Read on SocArXiv →Measuring what humanity knows, and what we're losing. The Deep Time Research Institute is an independent nonprofit quantifying the accuracy of ancient knowledge systems before they disappear. All research outputs are open access.