Internet-Draft Scientific Constitutional Constraints November 2026
Watts Expires 20 May 2027 [Page]
Workgroup:
Network Working Group
Internet-Draft:
draft-watts-scientific-constitutional-constraints-00
Published:
Intended Status:
Experimental
Expires:
Author:
D. O. Watts
GoodShyt Group Inc.

Scientific Constitutional Constraints for Evidence-Governed Agentic Internet Operations

Abstract

This document proposes SCC-1, an experimental profile for preserving evidentiary, authorization, and effect-enforcement boundaries in agent-mediated Internet operations. The term constitutional describes a locally adopted set of durable technical constraints; it does not refer to any national constitution, create law, or determine legal validity. SCC-1 separates authentication, attestation, evidence appraisal, policy authority, delegated scope, action approval, sink admission, execution, and observed outcomes. It specifies fail-closed state transitions, auditable claim promotion, and bounded closure statements for explicitly scoped sinks. This document does not introduce a new credential or OAuth grant. Its claims remain experimental: reported local Python/SQLite and Go/file-state experiments are not independent cross-host verification or production security certification.

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."

This Internet-Draft will expire on 20 May 2027.

▲

Table of Contents

1. Introduction and Scientific Constitutional Scope

Autonomous systems may present valid cryptographic credentials, accurate measurements, apparently qualified evidence, and an affirmative policy statement while lacking authority to commit one exact effect at a particular sink. This document describes experimentally testable constraints that prevent one state from silently standing in for another.

In SCC-1, constitutional means that a deploying organization has declared a durable set of technical invariants governing how the system reaches and reports its decisions. This terminology does not incorporate, interpret, modify, or supersede constitutional law in any jurisdiction; it does not confer legal standing on software, an AI agent, or this document.

The main design principle is non-collapse: evidence of identity is not permission, an approval is not an effect, a local revocation cut is not global closure, and passing a local test is not independent validation. The profile applies these separations to consequential Internet actions, but leaves existing authentication and authorization protocols in place.

This -00 source is a private research candidate prepared for possible future author-controlled review. It has not been submitted to the IETF or adopted by any working group. Its proposed status is Experimental and zero independently confirmed novel discoveries are claimed.

2. Conventions and Requirement Language

The key words MUST, MUST NOT, REQUIRED, SHALL, SHALL NOT, SHOULD, SHOULD NOT, RECOMMENDED, NOT RECOMMENDED, MAY, and OPTIONAL are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals.

Any deployment claiming SCC-1 conformance is assessed against the explicit trust scope, operation types, and verification stage stated in its conformance report. This experimental document does not extend the applicability of those claims beyond the declared scope.

3. Terminology

4. Constitutional Non-Collapse Boundaries

IdentityValid != ActionAuthorized
EvidenceAuthentic != ClaimTrue
AttestationValid != PolicySatisfied
ApprovalPresent != ApprovalCurrent
DecisionALLOW != EffectCommitted
RevocationRequested != RevocationClosed
LocalTestPASS != IndependentReproduction
DigestMatch != ScientificValidity
TechnicalAdmission != LegalAdjudication
Figure 1: State Separation

Each line is a non-implication rather than an assertion that the left-hand event is useless. An implementation MUST NOT promote the left-hand state into the right-hand state without the evidence and transition checks separately required for the latter.

A computational digest binds bytes under a chosen encoding; it does not demonstrate substantive correctness, legal authority, experimental validity, implementation security, authorship, or original invention.

5. Local Constitutional Rule Set

A deployment adopting SCC-1 MUST maintain an authenticated, versioned rule-set identifier and immutable digest over the rule set used for a decision. The effective rules MUST be selected by the enforcement domain, not by an untrusted agent or its proposed context.

The rule set MUST identify the protected effect operations, sink identities, authorized issuers, delegation rules, freshness limits, required witness classes, revocation and emergency barriers, and its treatment of unknown legal or jurisdictional applicability. A missing mandatory rule or unresolved rule version is INDETERMINATE for an otherwise well-formed request.

A legal-applicability predicate MAY be required by local policy. Its authenticated assessment, if used, is an input supplied by separately authorized qualified processes or professionals, not a legal conclusion computed by SCC-1. Required but unresolved applicability MUST NOT be treated as lawful authorization.

A rule-set transition MUST NOT reuse approvals, authorizations, or evidence receipts tied to an older digest unless their continued applicability is independently and explicitly established under the new rule set.

6. Scientific Claim and Evidence Ledger

For each claim used to justify a security or compliance promotion, the research record SHOULD separate (1) the exact falsifiable claim, (2) evidence source and immutable content digest, (3) evaluation procedure and version, (4) scope and assumptions, (5) result, (6) independent evaluator identity or an explicit absence thereof, (7) refuting counterexample, and (8) release status.

A verifier MUST NOT replace NOT RUN with PASS. It MUST NOT combine a synthetic oracle result with independently measured field results under a single denominator or certification label. A failing normative vector may be a test-oracle defect rather than an implementation defect; the defect and the verifier state must remain separate.

This record is informational for research auditability and does not replace an IETF erratum, independent peer review, or a legal evidentiary rule.

7. Action-Time Decision Procedure

SCC-1 adopts the CAG-1 decision boundary as a conceptual dependency and does not redefine its exact 88-vector synthetic context. The required facts for a consequential effect are authenticated identity (I), verified cryptographic provenance (K), scope and attenuation (S), delegation authority (D), independent approval when required (A), temporal and epoch freshness (T), revocation and runtime barriers (R), and trusted policy provenance (P). The local expression ALLOW iff I AND K AND S AND D AND A AND T AND R AND P is a conjunctive acceptance condition, not a proof that any verifier supplied correct facts.

Malformed fact context --> INVALID (no CAG decision)
Well-formed + required FAIL --> DENY
Well-formed + no FAIL + required UNKNOWN --> INDETERMINATE
Well-formed + all required PASS --> ALLOW
ALLOW + trusted sink transaction --> optional effect commit
Every other branch --> no effect admission
Figure 2: Decision and Admission Separation

The decision precedence for valid inputs is DENY if any mandatory condition is explicitly and reliably FAIL; otherwise INDETERMINATE if any mandatory condition is UNKNOWN; otherwise ALLOW. A failed cryptographic envelope that cannot be trusted is not a conformant CAG semantic DENY and MUST be distinguished from an evaluated rule failure.

SCC-1 implementations MUST independently compare the authenticated action audience to the configured sink identity. A request signed for another sink MUST NOT commit at the local sink even when its own grant and proof agree with the claimed audience.

Before evaluating effect bytes and before committing them, the sink MUST use a stable private request snapshot or equivalent immutable representation. The effect, audience, policy digest, delegation chain, evidence and approval bindings, and actor nonce MUST describe the same validated request state.

An evaluated ALLOW is a non-bearer local decision. The sink MUST atomically order nonce consumption, authority-epoch checks, and effect-commit state. A queued request MUST be re-evaluated at dequeue/commit time; a scheduling-time ALLOW MUST NOT bypass current barriers.

8. DICE, Remote Attestation, and Identity Assurance

Layered device identities are relevant only as evidence about an identified device or workload state. The Remote ATtestation procedureS (RATS) roles in [RFC9334] and the Entity Attestation Token in [RFC9711] separate attester evidence, verifier appraisal, and relying-party use. Signed COSE objects, if chosen, require the construction and verification semantics of [RFC9052].

Successful device quote verification or a measured model digest MUST NOT alone authorize an Internet action. The relying party MUST still evaluate its current policy, delegated grants, exact effect, audience, epoch, and sink barrier.

A prototype that has unverified PUF reconstruction, placeholder HMAC/SHA logic, secret readback registers, simulated signatures, or an incomplete verifier MUST NOT be described as a hardware-rooted production attester. SCC-1 does not specify a DICE key derivation implementation or assert that the submitted AI identity-engine RTL is secure.

Attestation evidence MAY be absent when the local policy does not require it. When required, absent or unqualified attestation yields INDETERMINATE or a more restrictive local rejection; no untrusted source may self-advance its assurance stage.

9. Scoped Revocation and Closure

Closure is always qualified by a declared epoch and an authenticated scope manifest. A local cut acknowledgement from sink A cannot establish enforcement at sink B. A trusted scope manifest MUST enumerate the member sink identifiers and their independently provisioned verification keys or authorized references.

S := authenticated, enumerated set of sinks
e := authority epoch being cut
Ack(s,e) := verified sink-owned signature over a durably applied cut
Closed(S,e) only if S is established, complete for the claim,
  and for every s in S, Ack(s,e) is verified and current,
  with pending-effect obligations discharged or revalidated.
Any missing required Ack --> UNKNOWN (not CLOSED)
Known stale in-scope effect after cut --> NOT_CLOSED
Figure 3: Scoped Closure Predicate

A cut acknowledgement MUST be generated after the sink durably applies its cut and MUST bind sink identity, scope-manifest digest, applied epoch, previous epoch, cut identifier, local apply-event identifier, and a trustworthy ordering or timestamp statement. Receipt of a message is not evidence of enforcement. A verifier MUST NOT trust public keys or membership replacements supplied solely inside an untrusted acknowledgement bundle.

A root-signed monotonic fencing notice MAY block old-epoch actions without conveying a complete cut; this may improve safety but does not by itself supply a valid applied-cut acknowledgement. Likewise, lease expiration is distinct from a signed monotonic fence and is sensitive to clock assumptions.

Any unobserved descendant, unenumerated sink, uncut queued operation, unknown propagation state, or unsupported enforcement path MUST limit the closure claim to the scope that is actually evidenced. If complete scope coverage is required and cannot be established, a positive closure claim MUST NOT be emitted.

In local synthetic research, a Go/file-state sink accepted a correctly addressed epoch-7 descendant action after a Python/SQLite sink applied epoch 8 but before the Go sink learned the cut. This is a counterexample to A-only global closure, not to the possibility of more complete coordination. Later local experiments used sink-signed acknowledgements and signed fencing notices; both operated under one host and a single research administration. Neither established independently audited cross-host closure.

10. Evidence Types and Experimental Receipts

Experimental receipts and policy records SHOULD identify their type and origin to prevent evidence laundering. Type values below are descriptive and are not registered media types, OAuth claims, or interoperable credentials: ACTION_DECISION, EFFECT_COMMIT, CUT_APPLIED, SINK_ACK, SCOPE_MANIFEST, EXPERIMENT_RUN, and QUARANTINED_VECTOR.

A receipt documenting one type MUST NOT be reinterpreted as a different type. In particular, ACTION_DECISION is not EFFECT_COMMIT, SINK_ACK is not universal graph coverage, and EXPERIMENT_RUN is not formal proof.

For a machine-readable local experimental interchange, source records SHOULD bind the exact digest algorithm, canonical encoding rules, issuer identity, signer role, scope digest, epoch, correlation identifier, source state, and evaluation procedure. A deployment MUST use an existing, reviewed authenticity mechanism for signed records; this document specifies no new signature algorithm or token.

{
  "profile": "scc-1-research",
  "receipt_type": "SINK_ACK",
  "sink_id": "sink-bravo",
  "scope_digest": "sha256:<64 lowercase hex digits>",
  "prev_epoch": "7",
  "applied_epoch": "8",
  "cut_id": "cut-demo-08",
  "apply_event_id": "42",
  "verification_state": "unknown"
}
NOTE: An unsigned JSON example is NEVER a valid acknowledgement.
Figure 4: Non-Normative Experimental Receipt Example

11. Conformance and Falsification Matrix

A conformance report MUST identify which tests were actually executed and at which assurance stage. The following are independently falsifiable test obligations for experimental implementations, not claims that the present source has passed them.

Each test MUST preserve actual expected and observed results, failed cases, model or implementation identifiers, provenance, and the exact input corpus digest. A tester MUST NOT claim independent reproduction when the candidate implementation and verifier share the same engineering administration.

12. Verification Promotion Boundaries

SCC-1 recommends stage labels for research claims: SPEC_VALIDATED, LOCAL_CONFORMANT, INDEPENDENTLY_REPRODUCED, SECURITY_TESTED, and RELEASE_ELIGIBLE. These labels describe distinct evidence requirements rather than an automatically increasing numeric trust score.

An implementation MUST NOT promote LOCAL_CONFORMANT to INDEPENDENTLY_REPRODUCED without genuinely independently controlled authorship/execution evidence. It MUST NOT promote deterministic conformance to SECURITY_TESTED without appropriately scoped adversarial assessment. A mechanically verified ZIP manifest establishes integrity of available bytes, not implementation validity or legal eligibility.

The current user-supplied CAG-1 record contains 88 frozen semantic vectors, Python/TypeScript agreement, single-database sink testing, and simulated two-sink acknowledgements. These observations MAY be recorded as local tests only. A missing cross-host study, real network partition, external key custody, or physical DICE validation remains NOT RUN.

13. Security Considerations

Threats include malicious source assertions, actor self-signing of a purported root, hidden delegation descendants, client-supplied trust pins, changing request data between validation and commit, cross-audience submission, captured proof replay, epoch rollback on restart, forged cut acknowledgement, network partition, crash inconsistency, clock skew, and compromised signing keys.

The protected sink MUST enforce actions after authenticated validation rather than merely log decisions. Implementations SHOULD use authenticated control channels, narrowly scoped pins, revocation-resistant key rotation, durable monotonic epoch state, replay protection, explicit time assumptions, no public debug paths to key material, and controlled zeroization. A hash chain does not alone establish immutability or tamper-proof event storage.

For hardware identity, genuine cryptographic primitives, verified code-offset PUF reconstruction, measured conditional entropy, properly protected CDIs, and a complete COSE signature path are prerequisites to security claims. The earlier design-level RTL contains identified gaps; this document does not rehabilitate them through textual requirements.

Denial of service may result if missing acknowledgements keep closure UNKNOWN. Deployments SHOULD provide bounded expiry and recovery workflows without transforming UNKNOWN into CLOSED. Reporting uncertainty is intentional and necessary to prevent unsafe promotion.

Receipts may reveal model identifiers, workload identities, tenant relationships, jurisdiction, policy versions, internal sink topology, or proprietary research materials. Implementations SHOULD minimize exposed fields, use scoped opaque identifiers, restrict retention, and avoid replaying raw confidential context into logs or unauthenticated telemetry.

A local legal applicability rule may constrain an action, but this protocol has no legal jurisdiction and offers no determination of consent, fairness, constitutional validity, enforceability, or statutory compliance. The term constitutional MUST NOT be used by an implementation to imply government certification or legal review. Legal opinions and regulated decisions require separately authorized processes.

15. IANA Considerations

This private experimental candidate makes no IANA requests. It defines no registered HTTP header, media type, token, OAuth grant type, error registry, CBOR tag, or Internet Standards Track requirement.

16. Relationship to Existing Work and Novelty

The relevant prior art includes OAuth token revocation [RFC7009], token exchange [RFC8693], DPoP [RFC9449], the OAuth Security Best Current Practice [RFC9700], RATS architecture [RFC9334], and the Entity Attestation Token [RFC9711].

The WIMSE AI Identity Management System draft [I-D.ietf-wimse-aims] and the individual Revocation Closure draft [I-D.watts-oauth-agent-revocation-closure] already address neighboring identity, authority, and revocation questions. SCC-1 adds a research-facing synthesis and explicitly falsifiable claim-promotion boundary; it does not claim that the constituent mechanisms or their conjunction are historically new.

Priority, novelty, patentability, scientific effectiveness, IETF consensus, and independent replication are NOT ESTABLISHED. No public rights or standards adoption are inferred from manuscript authorship or from passing local tests.

17. Decisive Open Experiments

Future experimental reports SHOULD include negative controls, intentionally late acknowledgements, unobserved descendants known to the test operator, clock skew, key compromise, power-loss recovery, and sink event traces whose ordering is independently audited. All unrun gates remain HOLD.

18. Changes in -00

Initial unsubmitted experimental source. Combines constitutional non-collapse constraints, scientific evidence-promotion rules, CAG-1 exact-effect admission, attestation-to-authorization separation, and scoped revocation acknowledgement boundaries. Derived from limited author-provided research artifacts; not an all-project survey or an originality determination.

19. Normative References

[RFC2119]
Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", RFC 2119, , <https://www.rfc-editor.org/info/rfc2119>.
[RFC8174]
Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", RFC 8174, , <https://www.rfc-editor.org/info/rfc8174>.
[RFC8259]
Bray, T., "The JavaScript Object Notation (JSON) Data Interchange Format", RFC 8259, , <https://www.rfc-editor.org/info/rfc8259>.
[RFC8785]
Rundgren, A., "JSON Canonicalization Scheme (JCS)", RFC 8785, , <https://www.rfc-editor.org/info/rfc8785>.

20. Informative References

[RFC7009]
Lodderstedt, T., Dronia, M., and M. Scurtescu, "OAuth 2.0 Token Revocation", RFC 7009, , <https://www.rfc-editor.org/info/rfc7009>.
[RFC8693]
Jones, M., Nadalin, A., Campbell, B., Bradley, J., and C. Mortimore, "OAuth 2.0 Token Exchange", RFC 8693, , <https://www.rfc-editor.org/info/rfc8693>.
[RFC9449]
Fett, D., Campbell, B., Bradley, J., Lodderstedt, T., and M. Jones, "OAuth 2.0 Demonstrating Proof of Possession (DPoP)", RFC 9449, , <https://www.rfc-editor.org/info/rfc9449>.
[RFC9700]
Lodderstedt, T., Bradley, J., Labunets, A., and D. Fett, "Best Current Practice for OAuth 2.0 Security", RFC 9700, , <https://www.rfc-editor.org/info/rfc9700>.
[RFC9334]
Birkholz, H., Thaler, D., Richardson, M., Smith, N., and W. Pan, "Remote ATtestation procedureS (RATS) Architecture", RFC 9334, , <https://www.rfc-editor.org/info/rfc9334>.
[RFC9711]
Lundblade, L., Mandyam, G., O'Donoghue, J., and C. Wallace, "The Entity Attestation Token (EAT)", RFC 9711, , <https://www.rfc-editor.org/info/rfc9711>.
[RFC9052]
Schaad, J., "CBOR Object Signing and Encryption (COSE): Structures and Process", RFC 9052, , <https://www.rfc-editor.org/info/rfc9052>.
[I-D.ietf-wimse-aims]
Kasselman, P., Lombardo, J., Rosomakho, Y., Campbell, B., Steele, N., and A. Parecki, "AI Identity Management System", Work in Progress, Internet-Draft, draft-watts-scientific-constitutional-constraints-00, , <https://datatracker.ietf.org/doc/draft-ietf-wimse-aims/>.
[I-D.watts-oauth-agent-revocation-closure]
Watts, D., "Revocation Closure and Authority Epochs for Agent Delegation", Work in Progress, Internet-Draft, draft-watts-scientific-constitutional-constraints-00, , <https://datatracker.ietf.org/doc/draft-watts-oauth-agent-revocation-closure/>.

Author's Address

Deonté O'dell Watts
GoodShyt Group Inc.
San Francisco, CA
United States of America