Pre-registrations¶
The full set of pre-registration YAML files lives under v4/preregistration/ in the repository. Each file is a falsifiable prediction committed to git before data acquisition; once pushed, the predicted band cannot be silently widened post hoc.
See Pre-registration methodology for the schema and rationale.
Active pre-registrations¶
| System | Class | Predicted band | Verdict | Artifact |
|---|---|---|---|---|
cve-vulnerabilities | SOC / PA | \([1.5, 2.5]\) | FAIL | v4/validation/cve-vulnerabilities/ |
nyc-fdny-fires | SOC threshold cascade | \([1.3, 2.0]\) | INCONCLUSIVE (primary), FAIL (strict) | v4/validation/nyc-fdny-fires/ |
wsb-posts | Preferential attachment | \([1.8, 3.0]\) | (pending) | (pending) |
Additional historical systems (earthquakes, S&P 500, DeFi, neural, GitHub stars, FDIC, NIFC wildfires, GOES flares, Wikipedia views, power grid) were validated under earlier pipeline phases and are documented in the unified preprint (paper/v0-unified-pipeline-2026-05-13.md).
Verdict distribution¶
At the time of writing the pipeline has produced:
- PASS: 13 systems (earthquakes, S&P 500, Aave V2, Compound V2, MakerDAO Dog, mouse ALM cortex, GitHub stars, FDIC bank failures, NIFC wildfires, GOES solar flares, Wikipedia pageviews, power grid, plus the universal-collapse polish).
- INCONCLUSIVE: 1 system (NYC FDNY fires, primary series).
- FAIL: 1 system (CVE disclosures), 1 secondary series (FDNY strict).
The asymmetric distribution of outcomes is itself an argument that the pipeline is not over-claiming: a pipeline that produced only PASS would be suspect, and a pipeline that produced only FAIL would not justify the cross-domain universality claim. The 13/1/1 split, with mechanism-level explanations for the non-PASS verdicts, is the methodological target.