The doctrine map
Where the curve meets the 45° line, expectation and enforcement agree. Filled dots are stable, hollow dots are watersheds; the band marks the support of the case population.
The doctrinal potential
Wells are equilibria; the ridge between them is the barrier an accident must clear. Both wells sit on the edges of the doctrinal range when the population lies inside it.
The fold in c
Equilibrium ceiling as the case population shifts. Between the folds the map carries two stable answers and history decides. Hatched: the population would contain cases below one, which Definition 1 rules out — the lower fold lies there.
The fuse
Minimal duration after which a cover withdrawal of depth ℓ1 becomes irreversible, at adjustment speed η = 0.5. Two periods for a deep shock; without bound as ℓ1 approaches the watershed. A shock that keeps ℓ1 above the watershed is reversed at any duration.
What is computed. F is the exact distribution of
m_crit = 1/(θ0(1−ξκ̄)) with θ0 equiprobable on {0.78, 0.80, 0.82,
0.85} and ξκ̄ uniform on the sharing range, shifted so that its mean is c; the
potential U is its closed-form integral. Recovery and exit times are exact first-passage
quadratures of the perturbed adjustment with reflection at 1 and 1+λ̄ (not the Kramers
approximation), in periods, to be read as orders of magnitude. Adjustment speed η = 0.5.
The folds are computed from g0(u) = 1 + λ̄F0(u) − u; c* is
1 + λ̄/2 whenever the population lies inside the doctrinal range there, and found
numerically otherwise; c† is where the expected exit time of the permissive doctrine equals
30 periods.
Version. Version 6 of the paper (September 2026) replaced the logistic stand-in of earlier
versions by this population, which moved the watershed from 1.95 to 1.77 and made negotiated German contracting
multiple rather than determinate; this page was rebuilt on it and reproduces the paper's numbers at printed
precision. The page is a companion, not the source of record: every number in the paper comes from
population.py, make_doctrine.py and make_stochastic.py in the replication
package, doi:10.6084/m9.figshare.33212916.
Cite. Companion to Andreas Bauer, Doctrine as a Fixed Point: A Formal Model of the Enforceable Penalty Ceiling when Human Oversight of AI Must Remain Effective, working paper, version 6, September 2026. Source and verification script: github.com/Tafew/doctrine-fixed-point; licensed CC BY 4.0.