# SPQR Research | Runtime Governance, Methods & Evidence Canonical page: https://spqrtech.ai/research/ Content edition: 2026-10-10 This Markdown edition is generated from the public page content. --- SPQR Research # Here is the thesis. Here is the evidence. *Here is how to test it.* Our research asks what must be true before autonomous systems can act under policy, produce defensible evidence and improve without escaping governance. [Read the bounded evaluation](https://spqrtech.ai/evidence/)[Review the test protocol](https://spqrtech.ai/challenge/) Runtime governanceVerifiable policyConstitutional authorityGoverned improvement SPQR / RESEARCH OPERATING SYSTEMPUBLIC S **PROOFS**NOT PROMISES 01**Authority**Who is permitted to decide? 02**Policy state**Which reviewed rule is active? 03**Action boundary**Can control intervene before effect? 04**Evidence**What proves the chain held? PAPERS → ARCHITECTURE → PRODUCTS → PROOFCHALLENGE WELCOME The research position ## We publish the architecture because *“trust us” is not a governance model.* Authority, evidence and execution conditions should be separable from the model generating the proposal, then inspectable by people who did not build the system. Model proposes · authority verifies · runtime enforces · evidence remains Four research lines. One operational question. ## What has to be true before an autonomous system is allowed to *cause an effect?* The corpus moves from jurisprudential questions to concrete system boundaries. Each line asks a different part of the same governance problem. 01 Question ### Where does authority live? Research line**Constitutional AI & verifiable trust** Legitimacy, public authority, non-unilateral amendment and constraints that bind machine action. FOUNDATION 02 Question ### How does policy become executable? Research line**Cryptographic policy enforcement** Signed policy state, provenance, active authority and deterministic governance artefacts. PRAXIS 03 Question ### Can governance intervene before effect? Research line**Runtime action-boundary control** Trusted provenance, mapped operations, fail-closed execution and per-action verdicts. AEGIS 04 Question ### How can systems improve without self-authorising? Research line**Governed candidate improvement** Auditable proposals, challenge evidence, containment and held-out evaluation rather than invisible self-modification. NEXT Start with a measured claim ## What the runtime evaluation *actually reports.* Prompt-policy comparator 79 ### Risky proposal-leakage rows. Reported in the prompt-policy comparator path. This is a proposal endpoint, not an executed-side-effect measure. Governed sandbox path 0 / 2,100 ### Risky side-effect completions. Zero observed in the evaluated governed rows. The same path also recorded **zero governed mock-tool applications**. Attempted governed decisions 1,832 / 1,832 ### Trusted provenance retained. All attempted governed decisions preserved runtime-resolved provenance in this evaluation. ### The result is bounded. The next test counts useful work. 6,300 total rows across 42 tasks, five run families, three conditions and ten repeats per family. Different endpoints must not be presented as a percentage reduction. These results do not establish useful allowed-task completion, universal AI safety or production bypass resistance. [Read the methods and limitations](https://spqrtech.ai/evidence/) · [Review the next evaluation protocol](https://spqrtech.ai/challenge/) Start here ## Three papers that explain the *operating thesis.* These works are the shortest path into the architecture: the runtime boundary, the proof model and the governed-improvement problem. [FEATURED PAPER 012026 · PREPRINT · ARXIV Runtime governance ### Runtime Governance for Agentic AI Action-boundary control with trusted provenance and fail-closed execution. The paper evaluates Aegis as the independent seam between an agent proposal and a real-world tool effect. Action boundaryProvenanceFail closedReceipts Adam Mazzocchetti**Read on arXiv ↗**](https://arxiv.org/abs/2608.16891) [PAPER 022026 · SSRN ### Proofs, Not Promises Governed candidate improvement as an auditable lifecycle rather than an invisible update. **Read the working paper ↗**](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6835839)[PAPER 032026 · ARXIV ### Cryptographic Runtime Governance The Aegis architecture for turning policy constraints into verifiable execution conditions. **Read the preprint ↗**](https://arxiv.org/abs/2603.16938) Bounded claims ## A controlled result is evidence about the controlled result. *Nothing more.* Serious governance research should expose its boundaries rather than smuggling confidence through tone. NOT A CLAIM OF**Universal AI safety** NOT A CLAIM OF**Automatic legal compliance** NOT A CLAIM OF**Infallible classification** THE CLAIM IS**The stated process, authority and evidence can be inspected.** Public research library ## Read the corpus. *Follow the argument.* Preprints, working papers and whitepapers are labelled by their public source. Use the filters to move through the research lines. [Search Google Scholar](https://scholar.google.com/scholar?q=%22Adam+Mazzocchetti%22) Search papers AllRuntime governanceConstitutional AIGoverned improvementWhitepapers 9 public works shown 01 **2026**Preprint · arXiv ### Runtime Governance for Agentic AI: Action-Boundary Control with Trusted Provenance and Fail-Closed Execution A trusted runtime seam between agent proposals and tool effects. [Read ↗](https://arxiv.org/abs/2608.16891) 02 **2026**Working paper · SSRN ### Proofs, Not Promises: Governed Candidate Improvement for Agent Control Runtimes An auditable candidate lifecycle instead of invisible self-modification. [Read ↗](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6835839) 03 **2026**Preprint · arXiv ### Cryptographic Runtime Governance for Autonomous AI Systems: The Aegis Architecture for Verifiable Policy Enforcement Signed policy state and auditable governance artefacts at the execution boundary. [Read ↗](https://arxiv.org/abs/2603.16938) 04 **2025**Whitepaper · SSRN ### SPQR Whitepaper: A Constitutional Framework for Sovereign AI Governance The broader constitutional-governance thesis and supporting SPQR architecture. [Read ↗](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5344913) 05 **2025**Working paper · SSRN ### Lex Aegis: The Living Law ; A Public Blueprint for Constitutional AI Governance Public participation, action in the absence of authority and plural governance. [Read ↗](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5382473) 06 **2025**Working paper · SSRN ### Lex Fiducia: Constitutional AI and the Architecture of Verifiable Trust Verifiable policy state and non-unilateral amendment as a basis for trust. [Read ↗](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5276785) 07 **2025**Working paper · SSRN ### Lex Digitalis ; The System Finds Itself in Contempt: Immutable Ethics for Autonomous AI Cryptographically enforced machine constraints as operational legal instruments. [Read ↗](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5283239) 08 **2025**Working paper · SSRN ### Lex Incipit: A Constitutional Doctrine for Immutable Ethics in Autonomous AI Binding, verifiable constraints established at system genesis. [Read ↗](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5326292) 09 **2025**Working paper · SSRN ### Lex Aeterna Machina: Autonomous Ethical Governance in the Age of Artificial Intelligence Non-discretionary responsibilities before broader operational freedom. [Read ↗](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5326353) No papers match that search and filter. From paper to product ## The papers are not the product. *The products are where the thesis is forced to work.* Praxis and Aegis turn the research questions into concrete assurance and control surfaces. [[Image: Praxis Policy Assurance product interface] PRAXIS ### Evidence-bound policy assurance. Test what organisational evidence supports, preserve ambiguity and remediation decisions, and establish approved authority. **Explore Praxis →**](https://spqrtech.ai/praxis/)[[Image: Illustrative Aegis decision console showing sample allow, escalate and deny states; not live telemetry.] AEGIS ### Runtime authority before effect. Verify signed active policy state at the action boundary and preserve a per-action decision receipt. **Explore Aegis →**](https://spqrtech.ai/aegis/) One clear next step ## Keep your safeguards. *Bring a testable objection.* Review the proposed protocol before any access is granted. Useful work, false denials, bypass routes and operating cost all belong in the result. [Review the evaluation protocol](https://spqrtech.ai/challenge/)[Request a methods conversation](https://spqrtech.ai/contact/#research) One workflow. Defined scope. A decision about fit.