KAIROS Observatory¶
A seismograph for the intelligence explosion, built to be proven wrong.
KAIROS is an open early-warning instrument. It watches two curves, what AI can
do, C(t), and what humans can oversee, H(t), and keeps answering one
question: how much time is left while intervention is still cheap?
It reports four dials, a tripwire level agreed in advance, and a lead time with a 90% interval. Every weekly forecast is hashed into a public, timestamped ledger before its outcome is known.
| Dial | Symbol | Danger sign |
|---|---|---|
| Acceleration | δ, t_c |
δ significantly above 0 on several independent metrics |
| Recursion | κ |
κ rising across rolling windows |
| Vinge Horizon | h* |
h* shrinking |
| Jonas Gap | dG/dt |
dG/dt > 0, sustained |
KAIROS recommends; humans decide.
Where to go next¶
- Methodology: the mathematics and the decision rules.
- Weekly pipeline: what
kairos run weeklydoes. - API reference: generated from the docstrings.
- Design notes: the non-negotiables, toolchain pins, hardware plan and design notes D1–D12.
- The full idea lives in the repository as
KAIROS_Project_Idea.md.