Runtime Admissibility Experiment

Runtime Admissibility Experiment is a minimal, deterministic, reviewable research repository accompanying the paper:

Runtime Admissibility as a Computational Primitive for Institutional Reliance
Arkadiy Miteiko and Ignacio Adrian Lerer, 2026

The repository demonstrates a precise computational distinction:

1. a static authorization system may continue relying on a prior permissibility classification after institutional state has changed;
2. a runtime admissibility system re-evaluates current authority, evidence, state, scope, consequence, execution integrity, and correctability before institutional reliance.

The primary demonstrated failure mode is Dynamic Classification Failure (DCF).

---

Core formal claim

Institutional reliance on a machine-generated action requires runtime admissibility at the moment of reliance:

$$Rel(x,t) \Rightarrow RA(x,t)$$

Defective reliance occurs when an institution relies without runtime admissibility:

$$Rel(x,t) \land \neg RA(x,t)$$

Runtime admissibility is modeled as a conjunctive predicate:

$$RA(x,t) \Leftrightarrow VA(x,A_t) \land SE(x,E_t) \land CS(x,S_t) \land WS(x,A_t,C) \land CB(x,C) \land EI(x,t) \land COR(x,C)$$

Predicate	Meaning
VA(x,A_t)	Valid authority exists for action x under current authority envelope A_t.
SE(x,E_t)	Sufficient evidence exists under current evidentiary basis E_t.
CS(x,S_t)	Action x is compatible with current institutional state S_t.
WS(x,A_t,C)	Action x is within delegated scope for consequence class C.
CB(x,C)	Consequence class C is bounded and recognized.
EI(x,t)	Execution integrity is preserved at time t.
COR(x,C)	Correction, challenge, or remediation remains available.

---

What this repository demonstrates

System A: Static authorization

Checks whether an action was permissible at t0.

It does not re-evaluate current institutional state at t2.

Under material state change, it may produce:

APPROVE -> defective reliance

System B: Runtime admissibility

Checks all runtime admissibility predicates at t2.

Under material state change, it produces:

ESCALATE or BLOCK -> defective reliance prevented

---

Quick start

python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -e ".[dev]"
pytest

Run one simulation:

python -m runtime_admissibility.cli scenarios/financial_transaction_state_change.json

Run all scenarios:

python -m runtime_admissibility.cli scenarios --out results/all_scenarios_output.json

---

Repository layout

runtime-admissibility-experiment/
  README.md
  LICENSE
  CITATION.cff
  CODE_OF_CONDUCT.md
  CONTRIBUTING.md
  SECURITY.md
  REPRODUCIBILITY.md
  Makefile
  pyproject.toml
  requirements-dev.txt
  .gitignore
  .github/workflows/tests.yml
  .github/ISSUE_TEMPLATE/bug_report.md
  .github/ISSUE_TEMPLATE/scenario_request.md
  docs/
    formal_model.md
    experiment_protocol.md
    scenario_schema.md
    reviewer_guide.md
    validation_matrix.md
  ontology/
    predicates.json
  paper/
    formal_model.md
  scenarios/
    financial_transaction_state_change.json
    authority_revocation.json
    evidence_contradiction.json
    execution_drift.json
    context_loss.json
    correction_unavailable.json
    low_consequence_allowed.json
  src/runtime_admissibility/
    __init__.py
    models.py
    predicates.py
    engines.py
    simulation.py
    cli.py
  tests/
    test_dcf_static_failure.py
    test_runtime_admissibility_blocks.py
    test_authority_revocation.py
    test_evidence_contradiction.py
    test_execution_drift.py
    test_context_loss.py
    test_correction_unavailable.py
    test_low_consequence_allowed.py
    test_scenario_schema.py
  results/
    sample_static_failure_output.json
    sample_runtime_admissibility_output.json

---

Expected primary result

For the primary DCF scenario, the static authorization system approves the transaction because it was permissible at t0.

The runtime admissibility system escalates because the current state at t2 invalidates the state-compatibility predicate.

{
  "scenario_id": "financial_transaction_state_change",
  "static_authorization_system": {
    "decision": "APPROVE",
    "defective_reliance": true,
    "failure_modes": ["Dynamic Classification Failure", "Reliance Without Admissibility"]
  },
  "runtime_admissibility_system": {
    "decision": "ESCALATE",
    "runtime_admissible": false,
    "defective_reliance_prevented": true,
    "failed_predicates": ["compatible_state"]
  }
}

---

Research artifact boundaries

This repository is intentionally:

• implementation-independent: no commercial platform assumptions;
• deterministic: same scenario files produce the same outputs;
• auditable: each predicate returns reasons, not just booleans;
• extensible: new scenarios can be added without changing engine logic;
• safe for peer review: no personal, customer, financial, or confidential data.

This repository does not provide legal advice, regulatory certification, financial compliance certification, or a production governance system.

---

Citation

Use CITATION.cff for software citation metadata.

---

License

MIT License.