PISF — Pre-Irreversibility Safety Framework
Distributed irreversible authority · timely refusal · human-revisable development
A — Distributed Irreversible Authority
The first dimension asks whether a small number of independent authority domains can force civilization-scale irreversibility. The paper introduces a robust threshold object, κ̲_E, and studies conditions such as κ̲_E > f.
Version 1.2 also represents authority exposure through a family of minimal irreversible authority cut sets, with cardinality and compromise-cost summaries. This makes shared dependencies and small-coalition failure explicit rather than reducing authority distribution to a single nominal count.
B — Effective Pre-Irreversibility Refusal
The second dimension asks whether refusal, stop, or isolation can complete before the last safe intervention point. The paper uses a robust refusal margin M_R^rob and a condition of the form M_R^rob > Δ_safe, together with the existence of an executable refusal policy.
C — Human-Revisable Development Viability
The third dimension asks whether ordinary automation, dependency, competition, and institutional change can gradually move society outside the set of states from which humans can still choose another development trajectory. The research uses a human-revisable viability object K_R^H.
Version 1.2 augments finite-horizon viability with future optionality: remaining inside the revisability region is not enough if the system becomes path-locked to a single materially distinct future corridor. The paper uses an application-specific ΩδH diagnostic for preserved materially distinct human-governed future options.
Assurance conditions
Boundary Completeness
Do major causal pathways remain inside the evaluation boundary?
Model Adequacy
Are the state, transition, timing, dependency, and uncertainty models adequate for the claim being made?
Independent Assurance
Are evidence, threat model, validation basis, and interpretation independent enough from the authority or interests being evaluated?
Evidence state: PISF v1.2 classifies evidence as Verified, Unresolved, or Failed. Unresolved evidence is not treated as a pass.
Safety as a maintained property
PISF does not treat A/B/C as a one-time check. After material change, the relevant property must be re-evaluated. The paper frames the maintained condition as A(t) ∧ B(t) ∧ C(t).
Reproducible reference cases
- R1 — Threshold Release Gate: a 3-of-5 gate with shared-root dependence, used to examine nominal versus robust authority distribution.
- R2 — Time-Bounded Emergency Refusal: detection, decision, authorization, enforcement, and verification latency, used to compute refusal margin.
- R3 — Fallback-Capacity Dynamics: a finite-horizon revisability model using
h(t+1)=h(t)-αa(t)+βm(t)to study automation/dependency, fallback investment, and horizon sensitivity. - R4 — Viable but path-locked: a four-state toy graph where two states remain inside the viability region, but only one preserves access to multiple materially distinct future terminal regions. It demonstrates that viability alone does not establish preservation of materially distinct future choices.
Open the Worked Research Cases guide.
What the cases establish — and do not establish
The published record describes the cases as toy/reference cases. They show executable definitions, reproducible evaluation procedures, independently recomputable parameter boundaries, and boundary behavior. They are not real-system empirical validation.
Relationship to LUMINA-30 and PCR-C
LUMINA-30 retains the boundary question; PCR-C remains the staged infrastructure-control layer. PISF adds a broader civilizational safety architecture connecting irreversible authority distribution, time-bounded refusal, and human-revisable development without redefining the LUMINA-30 canon.