Constitutional Epistemology and the Epistemic Foundations of Governable Autonomous Systems
Assessment basis: Revised from the earlier technical assessment and read
alongside the two subsequent Constitutional Epistemology documents.
Subject: Michal Harcej's Executable Governance model, Intelligence From
Architecture (IFA), and TauDIL Governance Operating System (GOS).
Status: Updated analytical assessment; conceptual, formal and
implementation evidence are distinguished.
Date: 4 November 2026
Executive Summary
This assessment updates the earlier evaluation of two
subsequent works: Constitutional Epistemology and A Formal
Specification for the Epistemic Foundations of Governable Autonomous
Systems. These works materially change the assessment of the principal
conceptual limitation identified in the earlier report: the distinction
between enforcing a constitution and establishing the standing of the
constitution, evidence, authority, identity and meanings on which
enforcement depends.
Deterministic execution
cannot, by itself, establish the legitimacy of the rules being executed.
The new work does not claim that computation can manufacture ultimate
legitimacy. Instead, it specifies how a system should represent, expose
and evaluate the epistemic and constitutional dependencies on which
governance relies, including dependencies that must be established
through legal, institutional, contractual, multi-party or other external
processes.
The resulting architecture is best understood as two related levels.
Executable Governance evaluates whether a proposed action is
admissible under applicable governing conditions. Constitutional
Epistemology evaluates whether the claims, evidence, authorities,
identities, semantic references and constitutional premises
participating in that determination possess the required standing for
the particular domain, purpose and time.
The new formal specification provides more than a philosophical
response. It defines an epistemic object model, a contextual evaluation
function, three outcomes (ADMISSIBLE, REFUSED and INDETERMINATE),
principles for provenance and authority attribution, a non-circular
legitimacy dependency graph, explicit foundational assumptions, identity
continuity, semantic integrity, cross-domain recognition, temporal
validity and constitutional mutation control.
Accordingly, the earlier conclusion that the initial-constitution
problem remained wholly unaddressed should be revised. It is now
addressed at the level of architectural doctrine and formal
specification, although the specification does not itself decide
contested political, ethical or legal questions, nor does it prove the
legitimacy of every possible external foundation. Runtime integration,
conformance testing, security evaluation and independent validation
remain necessary before claims of complete operational resolution can be
made.
Overall assessment: the new work substantially strengthens the
theoretical and formal completeness of the Executable Governance model.
It provides a principled way to prevent hidden assumptions and circular
self-validation from silently acquiring authority. The remaining
challenge is to demonstrate these requirements in a working,
independently testable implementation.
1. Scope, Materials and Assessment Method
The assessment considers three principal works:
Beyond Self-Validation: Evaluating the Technical Applicability of
Michal Harcej's Executable Governance Model for AI (earlier
assessment).
Constitutional Epistemology (Michal Harcej).
A Formal Specification for the Epistemic Foundations of Governable
Autonomous Systems, identified in that document as Constitutional
Epistemology v0.1 (Michal Harcej).
Three evidence levels are distinguished throughout. A conceptual
response means that the problem has been recognised and addressed at
the level of doctrine or architecture. A formal response means that
definitions, relationships, evaluation rules or required distinctions
have been specified. An implementation claim requires working code,
test evidence, operational controls or independent reproduction. These
levels must not be conflated.
2. The Core Governance Model: Capability, Authority and Executable Enforcement
The central premise of Intelligence From Architecture (IFA) is the
separation of intelligence from authority. An intelligent component may
analyse, interpret, recommend or propose. Its capability to produce an
output or initiate a process does not, by itself, confer permission to
cause a governed consequence. Authority is assigned and constrained
through the governing architecture.
Executable Governance translates this separation into an operational
control model. A proposed action is evaluated against explicit governing
conditions before the protected action is allowed to proceed. The
governance decision is not inferred from the persuasiveness of a
model-generated explanation. It is established through the execution of
defined checks and the preservation of the relevant decision evidence.
The earlier assessment identified this as a strength: it addresses the
conflation of capability and authority and replaces unstructured
reliance on model confidence with explicit admissibility conditions.
That assessment remains valid, subject to the precise scope of the
implementation and the quality of the rules and inputs.
The revised understanding is that Executable Governance is necessary but
not sufficient for autonomous constitutional systems. Even a perfectly
enforced rule set depends on the standing of the constitution, the
identity and authority of the rule-maker, the provenance and
completeness of evidence, the meaning assigned to terms, and the
legitimacy of changes to those conditions. Constitutional Epistemology
addresses this prior dependency layer.
3. Constitutional Epistemology: The Additional Governance Layer
Constitutional Epistemology asks what makes a proposition eligible to
participate in governance as authoritative knowledge. Its subject is not merely whether a rule has been followed, but whether the evidence, identity, authority, meaning and constitutional reference used in that determination possess the necessary standing.
The distinction can be stated as follows: verification establishes whether specified conditions have been satisfied; legitimacy concerns whether the relevant conditions and authorities have standing to govern. The formal specification expresses the point through the propositions:
Verified(x) does not imply Legitimate(x).
Admissible(x | C) does not imply Legitimate(C).
This distinction avoids two opposite errors. First, it prevents a system from treating successful execution as proof that its constitution is legitimate. Second, it avoids the unrealistic claim that all legitimacy questions can be settled by computation. The specification requires foundational dependencies to be made explicit, not disguised as computationally proven facts.
The two levels are complementary. Constitutional Epistemology supplies conditions for the standing of inputs and governing foundations. Executable Governance applies the constitution to proposed actions and enforces the resulting admissibility decision. The epistemic evaluation
does not replace legal, institutional or human authority; it makes the relevant authority and dependencies explicit and subjects their use to defined checks.
4. Reassessment of the Criticisms
4.1 The initial constitution and the source of authority
Common criticism: the architecture could enforce a constitution but did not provide a mechanism for establishing the legitimacy of that constitution. The question of who establishes the foundational rules and on what basis remained unresolved.
Updated assessment: this criticism is substantially addressed at the architectural and formal-specification levels. The new specification expressly recognises that not every constitutional foundation can be proven computationally. It allows foundations to be established through legal, institutional, contractual or other external processes and requires the system to identify them as explicit assumptions or dependencies.
The non-circular foundation requirement models legitimacy claims and their dependencies as a graph G=(N,R). A legitimacy determination must identify the originating claim, supporting evidence, responsible authority, applicable constitutional source, external or foundational basis, and unresolved dependencies. Circular dependency is not to be treated as independent proof.
The specification also distinguishes verified fact, authenticated
assertion, externally established authority, accepted constitutional premise, unresolved legitimacy dependency and unsupported assertion.
This is a meaningful technical response: it prevents an internal
representation from silently becoming its own source of authority.
Remaining boundary: the specification does not determine which external legal, institutional or political foundation is ultimately legitimate in every context. Nor does it settle disputes between competing authorities. It specifies how such dependencies must be represented and handled. The applicable institution or constitutional process must still establish the substantive foundation.
Conclusion: the earlier claim that the initial-constitution problem was wholly unaddressed is no longer accurate. The problem is now explicitly modelled and bounded, while the substantive establishment of external legitimacy remains context-dependent.
' 4.2 Determinism, truth and legitimacy
Earlier criticism: deterministic execution can reproduce an incorrect or illegitimate premise. In addition, hardware, software, parallel execution and numerical variation may undermine broad claims of identical outputs.
Updated assessment: the new specification directly incorporates the
first point. It explicitly states that deterministic evaluation is
conditional on formally specified, decidable rules and stable, complete inputs, and that determinism cannot eliminate unresolved external legitimacy questions. Such questions must remain explicit assumptions, dependencies or externally established determinations.
The three-valued epistemic result---ADMISSIBLE, REFUSED or
INDETERMINATE---is important. It prevents insufficient information from being silently converted into admissibility.
ADMISSIBLE: the applicable epistemic requirements have been met for the stated context.
REFUSED: a mandatory condition has failed.
INDETERMINATE: standing cannot be established from the available information.
This is a more precise account than treating determinism as a universal
guarantee. Deterministic governance concerns the reproducibility of
specified evaluation and enforcement under defined conditions; it does not establish truth, fairness, justice, or the legitimacy of the constitution.
Remaining boundary: the formal distinction does not, by itself, resolve implementation-level reproducibility. The engineering system must define its execution environment, canonical input representation, versioning, arithmetic behaviour, concurrency model, external dependencies and replay conditions. Tests must demonstrate the claimed reproducibility envelope.
Conclusion: the conceptual criticism is directly addressed and the
formal specification now states the limits of deterministic evaluation.
Hardware- and implementation-specific determinism still requires
engineering evidence.
4.3 Static correctness and continuing legitimacy
Earlier criticism: formal verification may establish that a system is
correct at a particular time, while legitimacy can change as evidence,
law, institutional authority, context or social conditions evolve. A
static constitution may become unsuitable or lose standing.
Updated assessment: the new specification substantially develops this
area. Standing is conditional and time-sensitive; prior standing does
not automatically establish present standing. Evidence completeness and
independence are addressed. Material changes to authority, evidence,
meaning, identity or connections require re-evaluation.
The specification treats purpose, identity and constitutional invariants
as related conditions. It distinguishes current identity, historical
identity, inherited credentials, delegated authority, cloned identity,
successor identity and reconstructed identity. Technical continuity or
similarity alone does not establish constitutional continuity. Material
alteration of identity-defining invariants may require constitutional
succession rather than ordinary revalidation.
Meaning is also treated as a governed object. Semantic references
include vocabulary versions, definitions, contextual constraints,
interpretation rules, domain applicability and transformation history. A
semantic mutation that changes the effect of a rule, claim, authority or
evidence object is not an ordinary knowledge update; it requires
explicit evaluation.
The specification further distinguishes knowledge acquisition from
epistemic admission, and epistemic admission from constitutional
authorisation. Learning a proposition does not automatically make it
authoritative for consequential action.
Constitutional mutation control distinguishes operational
transformation, which occurs within the existing constitutional order,
from constitutional transformation, which changes that order. Ordinary
operational authority must not silently acquire authority to alter the
constitutional conditions from which it derives.
Remaining boundary: a formal process must be implemented for review
triggers, expiry, revocation, challenge, appeal, human or institutional
approval, emergency handling, constitutional succession and migration of
existing authorisations. The specification establishes the categories
and principles; deployment-specific procedures and evidence are still
needed.
Conclusion: the earlier claim that the model addresses only static
correctness is materially weakened. The new work specifies continuous
re-evaluation and constitutional change controls, while operational
lifecycle procedures remain to be demonstrated.
4.4 Implementation gap and independent verification
Earlier criticism: publicly available specifications, diagrams and
builder-authored demonstrations are not equivalent to an independently
reproduced implementation or audited large-scale deployment.
Updated assessment: the new documents improve the formal basis on which
an implementation can be assessed. They provide definitions, an
evaluation function, object relationships, explicit outcomes and a set
of Constitutional Epistemology principles. This makes the architecture
more testable and provides a clearer basis for conformance criteria.
However, a formal specification is not proof that a product implements
it correctly. The new documents do not, on their own, establish that a
production TauDIL GOS deployment implements every epistemic requirement,
that its security properties have been independently tested, or that it
has achieved large-scale operational performance.
A credible validation programme should include:
A traceability matrix from each normative requirement to implementation
components and tests.
Conformance and adversarial test suites.
Deterministic replay tests.
Tests for missing, stale, conflicting, derivative and self-generated
evidence.
Authority revocation and domain-boundary tests.
Semantic mutation and constitutional succession tests.
Independent security and architecture review.
Conclusion: the implementation gap remains open. The new specification
improves testability but does not substitute for executable evidence,
independent reproduction or audited deployment results.
5. Technical Contributions of the Formal Specification
The formal specification advances the earlier model in several concrete
ways.
5.1 Epistemic object model and contextual evaluation
The evaluation function is defined as Ε(Q, E, A, S, D, T, C), where Q is
the claim, E the evidence, A the authority, S the semantic reference, D
the domain, T the time and C the constitutional reference. Its result is
ADMISSIBLE, REFUSED or INDETERMINATE.
The contextual form matters: a claim may be admissible for one purpose,
receiving system, domain, constitution, time or semantic context, and
inadmissible for another.
5.2 Evidence provenance, independence and completeness
The specification distinguishes independent, replicated, derivative,
correlated, self-generated evidence and evidence generated by an
authority under evaluation. Multiple records derived from one source do
not automatically provide independent corroboration. Mandatory evidence
that is absent, contradictory, expired or unverifiable must not be
treated as positive evidence.
5.3 Authority attribution and domain-bounded standing
An authority claim must identify its source, scope, domain, temporal
validity and authorisation basis. Possession of a credential is not
itself proof that the represented authority has legitimate standing.
Standing in one domain does not imply standing in another. Cross-domain
recognition requires an explicit recognition basis under the receiving
domain's constitutional conditions.
5.4 Identity, meaning and knowledge admission
The specification treats identity and meaning as governance-relevant
conditions rather than incidental metadata. It requires semantic
reference integrity and treats material semantic mutation as a governed
transformation. It also separates acquiring knowledge from admitting it
as authoritative and from authorising its use in consequential action.
5.5 Connections, authority transfer and mutation
External connections are governed objects with identity, source and
destination, provenance, authority, standing, scope, direction, semantic
contract, temporal validity, revocation conditions and mutation
authority. Technical connectivity or credential transfer does not
independently establish cross-domain standing or legitimate transfer of
authority.
5.6 Evidence before consequence
The requirement that evidence needed for admission must exist before the
protected consequence is a significant temporal safeguard. It prevents
retrospective evidence or interpretation from being used to manufacture
justification after an action has already occurred.
6. Implications for IFA and TauDIL GOS
The new work supports an expanded interpretation of the relationship
between IFA and TauDIL GOS. IFA provides the normative governance
architecture and the separation of capability from authority. TauDIL GOS
is intended to operationalise governance controls around intelligent
processes. Constitutional Epistemology specifies the epistemic
conditions that must be considered before claims, evidence, authority
and constitutional references are allowed to participate in those
controls.
A practical integration pattern would be:
Receive a claim, evidence object, authority assertion or proposed
constitutional reference.
Authenticate and preserve its provenance.
Evaluate source identity, authority, scope, domain and temporal
validity.
Evaluate evidence independence and completeness.
Bind the evaluation to a versioned semantic and constitutional
reference.
Return ADMISSIBLE, REFUSED or INDETERMINATE.
Permit only admissible material to proceed to the applicable IFA/TauDIL
governance checks.
Preserve the inputs, versions, decision, dependencies and audit
evidence.
This is an integration pattern implied by the formal model, not a claim
that every step is already implemented in a released TauDIL GOS build.
The exact interface, data schema, trust anchors, policy precedence,
state transitions, failure handling and evidence receipt format need to
be specified and tested.
The epistemic result should not be treated as a substitute for the final
governance decision. An epistemically admissible claim may still fail
constitutional, policy, role, relationship, separation-of-duties,
signature, prerequisite or state checks. Conversely, an indeterminate
epistemic result should not be silently promoted to permission. The
system's treatment of each outcome must be explicit in the applicable
policy and runtime contract.
For complex organisations, the architecture should support
domain-specific and federated governance arrangements while preserving
applicable organisation-wide constitutional constraints. Cross-domain
recognition and coordination should be explicit rather than assumed from
technical integration.
7. Remaining Research and Engineering Requirements
The revised assessment identifies the following work as necessary to
move from formal architecture to independently assessable operational
assurance.
Requirement-to-implementation traceability: map every normative
clause to code modules, configuration objects, state transitions, audit
events and tests.
Canonical epistemic data model: define versioned schemas for claims,
evidence, provenance, authorities, domains, semantic references,
constitutional references and dependency graphs.
Trust-anchor and foundational-authority procedures: define how
external foundations are registered, authenticated, challenged, renewed,
revoked and superseded in each deployment context.
Evaluation semantics: specify precedence and exact behaviour for
ADMISSIBLE, REFUSED and INDETERMINATE, including missing, conflicting,
stale, correlated and unverifiable inputs.
Lifecycle and mutation control: implement review triggers, expiry,
revocation, semantic version changes, identity succession,
constitutional amendment and emergency procedures.
Reproducibility envelope: define what must remain identical for
deterministic replay, including software versions, policy versions,
canonical inputs, execution environment and external data references.
Adversarial testing: test circular dependencies, compromised
issuers, copied or derivative evidence, authority escalation, domain
confusion, semantic substitution, replay attacks and retrospective
justification.
Independent validation: commission independent review of the
architecture, implementation, security controls, test coverage and
operational evidence. Clearly distinguish internal conformance results
from external assurance.
Human and institutional governance: define who has standing to
approve foundational changes, resolve contested claims, authorise
succession and review indeterminate cases. The interface must preserve
accountable authority without allowing informal overrides to bypass the
constitutional process.
8. Revised Overall Assessment
The earlier assessment's central insight remains correct: deterministic
execution cannot establish the legitimacy of the constitution it
executes. What changes is the conclusion that the architecture had no
sufficiently developed response to that problem.
Constitutional Epistemology now identifies the epistemic root as a
distinct governance concern. The formal specification gives that concern
structure through explicit objects, dependencies, principles and
contextual evaluation outcomes. It makes foundational assumptions
visible, rejects circular dependency as independent proof, conditions
standing on domain and time, governs semantic and identity change, and
requires evidence to precede consequential action.
The initial-constitution criticism is therefore substantially addressed
as an architectural and formal-specification problem, but not eliminated
as a substantive political, legal or institutional question. Determinism
is more carefully bounded, but implementation-level reproducibility
still requires evidence. The static-correctness criticism is
substantially addressed in the formal model, but operational lifecycle
mechanisms remain to be demonstrated. The independent implementation
criticism remains open.
The revised conclusion is that the work has progressed from a model
primarily concerned with enforcing legitimate rules to a more complete
governance architecture that also specifies how the standing of
evidence, authority, identity, meaning and constitutional dependencies
must be represented and evaluated. Its success in practice will depend
on faithful runtime integration, testable semantics, accountable
external foundations and independent validation.
9. Conclusion
The new materials materially strengthen the technical applicability of
Michal Harcej's Executable Governance model. They do not claim that a
machine can create its own legitimacy or settle every normative dispute.
Instead, they establish a disciplined boundary between what can be
computationally verified and what must be externally established, while
requiring the latter to remain explicit within the governance
architecture.
The central architecture can now be understood as a relationship between
Constitutional Epistemology and Executable Governance. The former
evaluates the standing of the claims, evidence, authorities, identities,
meanings and constitutional references on which governance depends. The
latter determines whether an action is admissible under the applicable
governing conditions and enforces the resulting decision.
This is a substantive development, not merely a change in terminology.
It gives the architecture a formal route for handling non-circular
foundations, evidence provenance and independence, conditional standing,
temporal validity, cross-domain recognition, identity continuity,
semantic mutation and constitutional succession.
The remaining test is operational. The requirements must be mapped into
TauDIL GOS, implemented with explicit state and data contracts,
exercised through adversarial and conformance testing, and subjected to
independent technical review. Until that evidence exists, the strongest
defensible conclusion is that the conceptual and formal architecture has
been substantially strengthened, while full implementation and
independent assurance remain matters for demonstration.
Appendix A. Summary of Revised Findings
• Earlier assessment issue • Revised position • Status •
• --- • --- • --- •
• Initial constitution / source of authority • Explicit foundational
assumptions, external authority dependencies and non-circular legitimacy
graph • Substantially addressed conceptually and formally; external
processes remain •
• Determinism does not establish truth or legitimacy • Explicit
distinction between verification and legitimacy; conditional
determinism; three-valued epistemic result • Addressed conceptually and
formally; reproducibility testing remains •
• Static correctness versus ongoing legitimacy • Conditional standing,
temporal validity, identity continuity, semantic mutation and
constitutional mutation control • Substantially addressed in formal
model; runtime lifecycle remains •
• Implementation - Implemented AND Tested in ARCHONT GOS https://github.com/tauguard/ARCHONT-GOS
Appendix B. Primary Materials Considered
Harcej, M. (2026). Constitutional Epistemology. Zenodo. https://doi.org/10.5281/zenodo.23932741
Harcej, M. (2026). A Formal Specification for the Epistemic Foundations of Governable Autonomous Systems (Version 1.0). Zenodo. https://doi.org/10.5281/zenodo.23851240
Top comments (0)