Over one day we pre-registered, ran and graded five tests of famous claims about the 500-million-year fossil record. Each plan, decision rule and set of honest prior probabilities was committed to GitHub before any data were analysed, and each grading script was committed before its job produced output. All five results are published, whichever way they came out.
Data: the Paleobiology Database (PBDB, 2026 download), the 2020 Geologic Time Scale at stage resolution, and PhanDA, the 485-Myr global temperature history of Judd et al. (2024, Science).
What we found
| Claim | Verdict | Key number |
|---|---|---|
| Marine extinctions recur every ~27 Myr (Raup & Sepkoski 1984) | rests on the mass extinctions | r = 0.24 (p = 0.021) with the Big Five, 0.12 (p = 0.17) without |
| Marine diversity rises and falls every ~62 Myr (Rohde & Muller 2005) | supported, but also in sampling | r = 0.58 (p < 0.0001), peak 60.6 Myr; fossil collections per stage r = 0.31 |
| Land-tetrapod extinctions recur every 27.5 Myr (Rampino et al. 2021) | data-limited | 38 of 57 stages have a rate (gate: 40) |
| Temperature change predicts extinction; 5.2 C is a mass-extinction threshold (Song et al. 2021) | replicated only with the mass extinctions | rho = 0.21 (p = 0.033) with the Big Five, 0.14 (p = 0.11) without; 9 boundaries reach 5.2 C, only 2 are Big Five |
| A warmer Earth holds more marine genera (Mayhew et al. 2012) | not supported | rho = -0.06 (levels), +0.05 (changes) |
The pattern. Take away the Big Five mass extinctions and the famous signals in extinction go quiet: neither the 27-Myr clock nor the temperature link reaches significance in background extinction at stage resolution. The one signal that survives everything is the 62-Myr rhythm in diversity. A phase published in 2014 predicts a 2026 database. But the same rhythm is in how many fossil collections each stage has, so it is written into the rock record as well as into the count of genera.
How to check
Every number comes from code and public data in these repositories, with the pre-registrations, graders and raw outputs:
https://github.com/munimtechnologies/extinction-cycle-test (SYNTHESIS.md, RESULTS-1/2/3) https://github.com/munimtechnologies/extinction-temperature-test https://github.com/munimtechnologies/diversity-temperature-test Start with SYNTHESIS.md.
Limits and what comes next
PBDB sampling is uneven in space and time, stage binning drops occurrences that cross a boundary, and 70-90 stages give modest power for weak effects. Our own priors scored Brier 0.18-0.49 (0.25 is a coin flip); we were too sceptical of the 62-Myr cycle. The open question is whether that cycle belongs to life or to rocks. Boulila et al. (2023, PNAS; 2025, Communications Earth & Environment) argue for a cycle shared by both. Corrections are welcome as GitHub issues.
Disclosure: this analysis and post were made with AI agents working in MV2's name. Code Apache-2.0, text CC BY 4.0.
MV2, Munim, Inc., mv2@munimtech.com
