<?xml version='1.0' encoding='UTF-8'?>
<rfc xmlns:xi="http://www.w3.org/2001/XInclude" version="3" category="info" docName="draft-watts-scientific-constitutional-law-00" submissionType="independent" ipr="trust200902" xml:lang="en" tocInclude="true" symRefs="true" sortRefs="true">
  <front>
    <title abbrev="Scientific Constitutional Evidence">Scientific Constitutional Law for Evidence-Bound AI Systems: A Fail-Closed Claims and Authority Profile</title>
    <seriesInfo name="Internet-Draft" value="draft-watts-scientific-constitutional-law-00"/>
    <author initials="D. O." surname="Watts" fullname="Deonté O'dell Watts" asciiFullname="Deonte O'dell Watts">
      <organization>GoodShyt Group Inc.</organization>
      <address>
        <postal>
          <city>San Francisco</city>
          <region>CA</region>
          <country>United States of America</country>
        </postal>
        <email>deonte@goodshyt.fun</email>
        <uri>https://orcid.org/0009-0005-8586-3650</uri>
      </address>
    </author>
    <date year="2026" month="11" day="16"/>
    <area>General</area>
    <keyword>evidence governance</keyword>
    <keyword>AI agents</keyword>
    <keyword>claim lifecycle</keyword>
    <keyword>revocation</keyword>
    <keyword>verification</keyword>
    <abstract>
      <t>This document describes an experimental evidence-governance profile for scientific claims consumed by automated and AI-assisted systems. It separates claim content, evidentiary evaluation, provenance, integrity, and operational authority. Its three-valued evaluation state (PASS, FAIL, INDETERMINATE) and separate authorization decision prevent signed records, citations, or computational checks from being automatically promoted into scientific truth or legal authority. It proposes immutable versions, scope-bound receipts, monotonic source generations, transitive revocation, authenticated freshness, and adversarial conformance cases.</t>
      <t>"Scientific Constitutional Law" is the name of the proposed procedural rule set; it is not a statement of enacted constitutional law, legal enforceability, scientific discovery, or IETF consensus. This document does not define a general truth engine or request IANA allocations.</t>
    </abstract>
  </front>
  <middle>
    <section anchor="introduction" numbered="true">
      <name>Introduction</name>
      <t>Scientific research and agentic systems routinely transform tentative hypotheses into summaries, structured records, and proposed actions. A correct file digest answers whether bytes changed; it does not answer whether the underlying scientific proposition is true. Likewise, a verifier may establish consistency with its declared inputs without validating the completeness of those inputs, and an authorization decision can permit a workflow without certifying the factual content on which the workflow relies.</t>
      <t>This document proposes a narrow set of constitutional-style constraints on how an evidence-bearing claim may be stored and operationalized. It draws on existing truth-maintenance systems, attestation architecture <xref target="RFC9334"/>, signed transparency logs <xref target="RFC9162"/>, and time-stamping methods <xref target="RFC3161"/>. It does not claim that these mechanisms are new.</t>
      <t>The intended use is as a testable research and interoperability proposal. This version is an individual Internet-Draft, not an approved standard or an externally validated security product.</t>
      <section anchor="requirements-language">
        <name>Requirements Language</name>
        <t>The key words <bcp14>MUST</bcp14>, <bcp14>MUST NOT</bcp14>, <bcp14>REQUIRED</bcp14>, <bcp14>SHOULD</bcp14>, <bcp14>SHOULD NOT</bcp14>, and <bcp14>MAY</bcp14> are to be interpreted as described in BCP 14 <xref target="RFC2119"/> <xref target="RFC8174"/> when, and only when, they appear in all capitals, as shown here. Their use specifies requirements of this experimental profile, not legal duties.</t>
      </section>
      <section anchor="scope">
        <name>Scope and Non-Goals</name>
        <ul>
          <li><t>In scope: typing claims, preserving evidence and authority states separately, preventing stale receipts from being reused, and defining conformance tests for the operational release boundary.</t></li>
          <li><t>Out of scope: deciding the scientific truth of arbitrary propositions, adjudicating constitutional rights, selecting a research hypothesis, proving a theorem, granting regulatory certification, and declaring a secure deployment without independent evidence.</t></li>
          <li><t>Out of scope: defining an interoperable wire format, global identity registry, timestamp authority, or cryptographic algorithm. Any use of this profile in an actual protocol requires a separate binding specification.</t></li>
        </ul>
      </section>
    </section>
    <section anchor="model">
      <name>Conceptual Model</name>
      <t>A claim version is modeled as a typed tuple <tt>C = (id, version, proposition, evidence, provenance, authority, dependencies, scope)</tt>. Additional controls include issuer-signed source generation, verifier identity and generation, policy epoch, timestamps, expiry, and a witness-bound checkpoint.</t>
      <dl>
        <dt>Proposition:</dt><dd><t>The complete statement under review, including essential quantifiers, units, conditions, limitations, and source reference. A shorthand label alone is not the proposition.</t></dd>
        <dt>Evidence evaluation:</dt><dd><t>PASS indicates that the declared predicate and evidence requirements succeeded under their specified snapshot and method; FAIL indicates a definitive failure of that declared check; INDETERMINATE indicates insufficient evidence, unavailable inputs, or an inconclusive evaluation. None alone proves an empirical claim universally true or false.</t></dd>
        <dt>Authority decision:</dt><dd><t>ALLOW, DENY, or DENY_PROMOTION describes an action under a policy and an authenticated principal. Authority is not a fourth epistemic state.</t></dd>
        <dt>Integrity:</dt><dd><t>A digest or signature binds bytes to an asserted origin or key, subject to its trust assumptions. It is not a source of scientific truth or legal authority.</t></dd>
        <dt>Freshness:</dt><dd><t>A claim that supporting evidence, policy, authorization, and time constraints remain valid for the requested action.</t></dd>
        <dt>Dependency closure:</dt><dd><t>The transitive set of source, verifier, policy, signer, and delegation facts necessary for the operative authorization path.</t></dd>
      </dl>
      <t>A recipient <bcp14>MUST NOT</bcp14> treat a record's validity as proof of all assertions recorded in it. An INDETERMINATE evaluation <bcp14>MUST NOT</bcp14> silently become FAIL, and neither state <bcp14>MUST</bcp14> be promoted into ALLOW for an action requiring PASS.</t>
    </section>
    <section anchor="principles">
      <name>Scientific Constitutional Controls</name>
      <section anchor="claim-identity">
        <name>Article 1: Proposition Identity and Scope</name>
        <t>Each evaluated claim <bcp14>MUST</bcp14> bind the complete proposition to its immutable version identifier, declared source snapshot, evaluation requirements, and responsible authority. Changes of meaning, units, or quantification require a new claim identity or an explicitly linked successor record. A verified substring <bcp14>MUST NOT</bcp14> be treated as verification of a larger, different proposition.</t>
      </section>
      <section anchor="noncollapse">
        <name>Article 2: Non-Collapse of Evidence and Authority</name>
        <t>Integrity, provenance, scientific evaluation, and operational authorization are separate assessment tracks. In particular, a valid signature does not entail a correct proposition; a correctly executed numerical solver does not entail physical validity; and an existing citation does not entail independent replication.</t>
        <t>For a chain of proposed authority transfers A to B to C, authorization from A to B does not by itself authorize A to C. Every required transition must be supported by a current, scoped, authenticated edge. Systems <bcp14>MUST</bcp14> verify the complete required path rather than infer transitivity from merely similar labels.</t>
      </section>
      <section anchor="versions">
        <name>Article 3: Immutable Versions</name>
        <t>Once an evaluation is recorded, its evaluation state and original evidence identifiers <bcp14>MUST NOT</bcp14> be rewritten. New evidence produces a new version linked to the predecessor. A successor PASS does not retroactively change a predecessor INDETERMINATE into PASS.</t>
      </section>
      <section anchor="scoped-receipts">
        <name>Article 4: Content-Bound, Scoped Verification Receipts</name>
        <t>A reusable receipt <bcp14>MUST</bcp14> identify the normalized predicate and arguments, claim version, tenant or principal scope, policy epoch, source identity and digest, issuer-signed source generation, verifier identity and generation, and applicable expiry conditions. Reuse across a nonmatching scope <bcp14>MUST</bcp14> be refused. Unknown verifier behavior or exceptions produce INDETERMINATE rather than an invented definitive FAIL.</t>
        <t>A verifier <bcp14>SHOULD</bcp14> document its deterministic inputs, method, error bounds, and reproducible toolchain. This does not require that all scientific investigations be deterministic.</t>
      </section>
      <section anchor="generation">
        <name>Article 5: Monotonic Generations and Source Restoration</name>
        <t>Before releasing a result, the system <bcp14>MUST</bcp14> verify an authenticated, nondecreasing source-generation watermark that cannot be reset by replaying older signed content. Restoring identical source bytes at a newer generation does not restore authority of receipts tied to earlier generations. Verifier-generation changes likewise cannot be undone merely by reusing an old verifier label.</t>
      </section>
      <section anchor="revocation">
        <name>Article 6: Dependency-Closed Revocation</name>
        <t>Let G be a directed dependency graph whose edges denote required authorization or evidence relationships. A claim may be released only if every required edge in the relevant dependency closure is current. Revocation of a necessary node or edge <bcp14>MUST</bcp14> invalidate each dependent release unless that release has been independently re-established under a new valid path and generation.</t>
        <t>In a multi-path authorization graph, a claimed cut set is valid only when it intersects all applicable authorization paths. A local invalidation at one edge is insufficient if an alternative previously authorized path can revive stale authority.</t>
      </section>
      <section anchor="freshness">
        <name>Article 7: Authenticated Freshness and Independent Checkpoints</name>
        <t>When an external witness is required, the witness response <bcp14>MUST</bcp14> bind a fresh request nonce, verifier-approved scope, signed time, checkpoint identity, and state commitment. The relying party <bcp14>MUST</bcp14> validate the latest acceptable checkpoint rather than merely any correctly signed historical checkpoint.</t>
        <t>At release, the relying party <bcp14>MUST</bcp14> recheck the complete transitive evidence closure at the conservative witness-authenticated time, including signed source-attestation expirations and receipt expirations. Clock skew is a tolerance for measurement, not permission to extend expired authority. The final mutable state snapshot <bcp14>MUST</bcp14> still match the witnessed commitment at the release boundary.</t>
        <t>An independently governed witness or immutable-retention claim <bcp14>MUST NOT</bcp14> be made solely because a local test witness can sign messages. Deployment evidence is distinct from protocol evidence.</t>
      </section>
      <section anchor="actions">
        <name>Article 8: Fail-Closed Operational Decisions</name>
        <t>An operational action requiring evidence qualification <bcp14>MUST</bcp14> be denied unless the relevant evaluation is PASS, all requirements are satisfied, the current policy independently permits the principal's action, and all freshness, scope, revocation, and witness checks succeed. Diagnostics that prevent safe authorization <bcp14>MUST</bcp14> also block release. A stored PASS receipt alone is never sufficient.</t>
        <sourcecode type="text/plain">
release_allowed =
    evaluation == PASS
    AND requirements_satisfied
    AND policy_authorizes(principal, action, claim)
    AND scope_and_tenant_match
    AND all_dependencies_current_at(witness_time)
    AND source_and_verifier_generations_current
    AND latest_witness_nonce_signature_valid
    AND witnessed_state_equals_final_snapshot
    AND no_blocking_diagnostics

if release_allowed:
    authority = ALLOW
else:
    authority = DENY or DENY_PROMOTION
</sourcecode>
      </section>
      <section anchor="responsibility">
        <name>Article 9: Attribution, Independent Review, and Release</name>
        <t>Implementation tests, synthetic benchmarks, independent source review, scientific replication, security certification, and publication authorization are separately recorded milestones. A project <bcp14>MUST NOT</bcp14> describe a self-authored software test as independent peer review, a locally signed witness as independent custody, or a novel combination as certified invention priority without corresponding external evidence.</t>
        <t>Confidential scientific evidence <bcp14>MUST NOT</bcp14> be disclosed merely because the profile can serialize or hash it. Public Internet-Draft submission is a separate owner decision governed by applicable IETF contribution and disclosure policies.</t>
      </section>
    </section>
    <section anchor="evaluation-algorithm">
      <name>Evaluation, Promotion, and Revocation Semantics</name>
      <t>An implementer can represent evaluation as a three-valued function E(C, S, V, P), where C is the proposition, S the source snapshot, V the verifier, and P the appraisal policy. The result MUST be interpreted only for the declared C, S, V, and P. A FAIL reflects that declared evaluation, not necessarily the real-world negation of C. Failure of authorization is recorded separately from evaluation failure.</t>
      <sourcecode type="text/plain">
if a required check definitively contradicts the declared claim:
    evaluation = FAIL
elif missing_required_evidence or unresolved_verifier_error:
    evaluation = INDETERMINATE
elif all_declared_checks_pass and evidence_current:
    evaluation = PASS
else:
    evaluation = INDETERMINATE

# Never derive factual truth or authorization from digest alone.
</sourcecode>
      <t>The decision policy applies authorization constraints to the evaluation result, and may deny an otherwise correctly evaluated claim. Conversely, an authorization policy <bcp14>MUST NOT</bcp14> turn an INDETERMINATE source evaluation into factual PASS. Repeated receipt reuse <bcp14>MUST</bcp14> incorporate the latest generation and revoked dependencies.</t>
    </section>
    <section anchor="conformance">
      <name>Adversarial Conformance Cases</name>
      <t>These synthetic cases define minimal negative checks. An implementation's success on these cases is local protocol-conformance evidence only; it is neither a complete penetration test nor a scientific validation of underlying claims.</t>
      <ol>
        <li><t>Unknown evidence: a syntactically valid assertion without required measurements remains INDETERMINATE and cannot authorize publication as experimentally established.</t></li>
        <li><t>Expired upstream attestation: a source attestation has expired at witness time while its derived receipt remains within its own TTL; release is denied even if the application clock lags.</t></li>
        <li><t>Superseded signed source: an attacker replays an older signed source generation with valid bytes and signature; the previously valid receipt remains invalid.</t></li>
        <li><t>Restored identical content: a new signed generation restores old source bytes; a receipt tied to the older generation remains invalid without re-evaluation.</t></li>
        <li><t>Verifier rollback: verifier V1 is replaced by V2 and then labeled V1 again; the old verifier-generation receipt remains invalid.</t></li>
        <li><t>Cross-tenant replay: an authenticated receipt issued for tenant A is presented by tenant B; release is denied, even if the proposition text and source digest match.</t></li>
        <li><t>Negative receipt scope: an exact FAIL receipt can be reused only while its snapshot and policy remain current. A changed source causes a new evaluation; a transient INDETERMINATE is not cached as stable FAIL.</t></li>
        <li><t>Checkpoint truncation or fork: a correctly signed but non-latest checkpoint is replayed after state changes; release is denied unless independent consistency and latest-state checks pass.</t></li>
        <li><t>Missing external governance: a self-administered local witness cannot satisfy a rule requiring independent custodian control or immutable cloud retention.</t></li>
        <li><t>Scientific-model boundary: arithmetic consistency of a proposed physical formula can pass while empirical validation remains INDETERMINATE; neither may be automatically upgraded to a confirmed scientific discovery.</t></li>
      </ol>
      <t>Cross-tenant isolation, external witness compromise, concurrent updates, policy races, source mutation, signer-key rotation, and failover <bcp14>SHOULD</bcp14> be included in broader threat-model-specific testing.</t>
    </section>
    <section anchor="security">
      <name>Security Considerations</name>
      <t>Attackers may replay valid but superseded signatures, roll back a clock or a database pointer, exploit a stale dependency, switch policy or verifier versions, forge a privileged action by crossing tenant boundaries, or alter an audit log before generating a new checkpoint. Hashes do not prevent a sufficiently privileged attacker from altering history and recomputing a hash; signatures do not show that retained objects are immutable or that the signing custodian is independent.</t>
      <t>Deployments need independent administrative custody where required, an authenticated latest-head mechanism, appropriate cryptographic key management, robust failure handling, and recovery tests. A transactionally correct release boundary and a durable witness acknowledgment need careful engineering to avoid both time-of-check/time-of-use races and unbounded database locks. Existing mechanisms such as <xref target="RFC9334"/>, <xref target="RFC9162"/>, and <xref target="RFC3161"/> are relevant antecedents; this document is not an alternative to their security analysis.</t>
      <t>Scientific evaluation error and infrastructure compromise are different threats. A fully signed and preserved incorrect evaluation may still be factually incorrect. The operational consumer must review the evaluator's scope and evidence independently when stakes require it.</t>
    </section>
    <section anchor="privacy">
      <name>Privacy Considerations</name>
      <t>Evidence may contain confidential research, identifying information, unpublished results, clinical information, or sensitive operational metadata. Receipts and transparency logs should carry only necessary hashes, scoped identifiers, and retention-controlled metadata. Public logs can leak the existence or timing of protected work even when source bytes are withheld. Consent, data minimization, retention limits, and access permissions remain independent requirements.</t>
    </section>
    <section anchor="iana">
      <name>IANA Considerations</name>
      <t>This document has no IANA actions.</t>
    </section>
    <section anchor="implementation-status">
      <name>Implementation and Research Status</name>
      <t>The procedural principles in this document are a proposed profile. Its components overlap earlier truth-maintenance, caching, remote attestation, and log-consistency methods. No claim of first invention, peer review, independently witnessed deployment, scientific truth determination, or conformity to an IETF-approved standard is made. External conformance, interoperability, and economic-effectiveness results require separately retained test evidence.</t>
    </section>
    <section anchor="acknowledgments">
      <name>Acknowledgments and Provenance</name>
      <t>The draft was prepared from PI-directed research on fail-closed epistemic governance, deterministic verification, and scoped evidence receipts. AI-assisted drafting is disclosed. Existing referenced research and RFCs retain their original authorship. Confidential laboratory records and unreviewed security implementation details are not included in this proposed public-facing text.</t>
    </section>
  </middle>
  <back>
    <references>
      <name>Normative References</name>
      <reference anchor="RFC2119" target="https://www.rfc-editor.org/rfc/rfc2119">
        <front><title>Key words for use in RFCs to Indicate Requirement Levels</title><author initials="S." surname="Bradner" fullname="Scott Bradner"/><date year="1997" month="March"/></front>
        <seriesInfo name="RFC" value="2119"/>
      </reference>
      <reference anchor="RFC8174" target="https://www.rfc-editor.org/rfc/rfc8174">
        <front><title>Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words</title><author initials="B." surname="Leiba" fullname="Barry Leiba"/><date year="2017" month="May"/></front>
        <seriesInfo name="RFC" value="8174"/>
      </reference>
    </references>
    <references>
      <name>Informative References</name>
      <reference anchor="RFC9334" target="https://www.rfc-editor.org/rfc/rfc9334">
        <front><title>Remote ATtestation procedureS (RATS) Architecture</title><author initials="H." surname="Birkholz" fullname="Henk Birkholz"/><author initials="D." surname="Thaler" fullname="Dave Thaler"/><author initials="M." surname="Richardson" fullname="Michael Richardson"/><author initials="N." surname="Smith" fullname="Ned Smith"/><author initials="W." surname="Pan" fullname="Wei Pan"/><date year="2023" month="January"/></front>
        <seriesInfo name="RFC" value="9334"/>
      </reference>
      <reference anchor="RFC9162" target="https://www.rfc-editor.org/rfc/rfc9162">
        <front><title>Certificate Transparency Version 2.0</title><author initials="B." surname="Laurie" fullname="Ben Laurie"/><author initials="E." surname="Messeri" fullname="Ericka Messeri"/><author initials="R." surname="Stradling" fullname="Rob Stradling"/><date year="2021" month="December"/></front>
        <seriesInfo name="RFC" value="9162"/>
      </reference>
      <reference anchor="RFC3161" target="https://www.rfc-editor.org/rfc/rfc3161">
        <front><title>Internet X.509 Public Key Infrastructure Time-Stamp Protocol (TSP)</title><author initials="C." surname="Adams" fullname="Carlisle Adams"/><author initials="P." surname="Cain" fullname="Pat Cain"/><author initials="D." surname="Pinkas" fullname="Denis Pinkas"/><author initials="R." surname="Zuccherato" fullname="Robert Zuccherato"/><date year="2001" month="August"/></front>
        <seriesInfo name="RFC" value="3161"/>
      </reference>
      <reference anchor="RFC7991" target="https://www.rfc-editor.org/rfc/rfc7991">
        <front><title>The "xml2rfc" Version 3 Vocabulary</title><author initials="P." surname="Hoffman" fullname="Paul Hoffman"/><date year="2016" month="December"/></front>
        <seriesInfo name="RFC" value="7991"/>
      </reference>
      <reference anchor="RFC9111" target="https://www.rfc-editor.org/rfc/rfc9111">
        <front><title>HTTP Caching</title><author initials="R." surname="Fielding" fullname="Roy T. Fielding"/><author initials="M." surname="Nottingham" fullname="Mark Nottingham"/><author initials="J." surname="Reschke" fullname="Julian Reschke"/><date year="2022" month="June"/></front>
        <seriesInfo name="RFC" value="9111"/>
      </reference>
    </references>
  </back>
</rfc>