| Internet-Draft | RKCS Claim-State Separation | October 2026 |
| Watts | Expires 25 April 2027 | [Page] |
A falsifiable, typed model separates ontological categories, epistemic states, evidence, consensus, uncertainty, provenance, and authority. The proposed protocol forbids consensus-only promotion of physical claims and requires traceable evidence for changes in verification status. Novelty and empirical advantage remain unestablished.¶
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 25 April 2027.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
This Internet-Draft is a faithful conversion of the author's research baseline "RKCS v0.1: Reality-Knowledge-Consensus Separation Framework", dated 20 October 2026, together with the preregistration draft of its 100-case synthetic pilot (EM-001) [RKCS-PKG]. Mathematical notation has been rendered in plain text. RKCS v0.1 is a frozen proposed specification plus a small deterministic reference checker, not a validated scientific discovery. It is published for discussion; it is not a standard.¶
The state of a proposition p in domain d at time t is the tuple¶
S(p,d,t) = (O, E, V, C, U, P, A)¶
Ontology O is one of {physical, institutional, subjective, mathematical, mixed}; epistemic status E is one of {unknown, proposed, supported, refuted, conflicting}. V denotes evidence records, C scoped consensus, U uncertainty, P provenance, A authority.¶
A promotion requires a nonempty rule identifier, authorized evaluator, applicable valid evidence, and no unresolved material contradiction. For physical claims, independent empirical evidence is required; for institutional claims, jurisdiction-matched institutional evidence; for mathematical claims, proof. This reference implementation is deliberately conservative and not a complete theory of knowledge.¶
Sample 100 claims, 20 each in physics, medicine, finance, law, and subjective experience. Use at least three independent reviewers per claim, with domain-specific adjudication and no ground-truth inference from majority agreement. Compare majority vote, evidence-weighted evaluation, and RKCS under equal information budgets. Primary endpoints: unauthorized promotion rate and claim-type-appropriate accuracy. Secondary: calibration (Brier score where binary ground truth is defined), abstention rate, coverage, contradiction preservation, and review latency. Freeze dataset, rules, baseline implementations, and analysis before testing. Confidence intervals and effect sizes must be reported. A 100-item pilot is not powered to establish small effects.¶
Adversarial cases: A01 unanimity without evidence; A02 reproducible minority contradiction; A03 institutional status change; A04 missing/stale/conflicting evidence; A05 unsupported derived promotion. Include positive controls to detect degenerate always-reject behavior. RKCS's empirical benefit is rejected if it fails preregistered endpoints against fair baselines.¶
Status: frozen synthetic pilot design, not independent empirical validation; the preregistration is NOT REGISTERED. The pilot uses 100 synthetic placeholders: 20 per domain, 10 scenario types repeated twice per domain. No real-world factual claims are used yet. A real benchmark requires independently sourced, versioned claims and source packets, with adjudicators blinded to model output.¶
Primary endpoint: invalid verification promotion rate on independently adjudicated ineligible cases, comparing RKCS with majority vote and evidence-weighted scoring under identical evidence access. Paired differences are reported with confidence intervals; significance thresholds and multiplicity correction are preregistered before collecting human judgments. Secondary endpoints: valid promotion recall, classification accuracy, abstention rate, uncertainty calibration (Brier score and ECE only when probability predictions exist), contradiction retention, latency and adjudicator disagreement, reported by domain and scenario.¶
Human review: at least three independent domain-qualified reviewers per claim, collecting separate classifications, supporting citations, uncertainty and conflict-of-interest declarations. Disagreements are resolved with a fourth blinded adjudicator or the contested labels are preserved. Leakage of intended outcomes, scenario codes and model predictions is prevented. The superiority claim is rejected if primary improvement is absent versus matched baselines, or if valid-promotion recall or abstention cost breaches preregistered noninferiority bounds. No threshold is selected post hoc.¶
Illustrative synthetic metrics (status ILLUSTRATIVE_SYNTHETIC_ONLY, n = 100):¶
| Model | TP | Invalid promotions (FP) | TN | Missed valid (FN) | Accuracy |
|---|---|---|---|---|---|
| Majority vote | 10 | 70 | 20 | 0 | 0.30 |
| Evidence-weighted | 10 | 30 | 60 | 0 | 0.70 |
| RKCS | 10 | 6 | 84 | 0 | 0.94 |
Caveats: labels are rule-derived, not independently adjudicated; RKCS is evaluated against an oracle aligned with its own rules (circularity risk); majority votes are simulated, not actual reviewer judgments; evidence weights and consensus proportions are illustrative; subjective claims require separate reference criteria; no probability calibration was tested. This is not a scientific performance claim: synthetic rule-oracle results are software smoke tests, not evidence that RKCS improves scientific reasoning. Independent adjudication NOT RUN.¶
Formal tests verify only the encoded rules, not empirical truth, metaphysical existence, external replication, priority, or patentability. No experiment with human reviewers has been performed. The reference code does not persist event logs, implement time validity, calibrate confidence, or establish evidence authenticity; these are required future work.¶
PI: Deonte O'dell Watts; AI-assisted formulation. First documented in this research record on 2026-10-20; historical priority unverified.¶
This document has no IANA actions.¶
This document defines no protocol. A system implementing these transition rules is a target for manipulation: an attacker who can inject fabricated evidence records, impersonate an authorized evaluator, or alter the append-only evidence store can cause unauthorized promotion of claims. Evidence authenticity, evaluator authentication and tamper-evident storage are therefore preconditions that the v0.1 reference code does not provide. Consensus signals are easy to manufacture (for example, coordinated or automated reviewers), which is the motivation for invariant I01; deployments should not weaken it. Retaining contradictory evidence (I03) may expose sensitive material and needs access control.¶