Sigma Runtime Standard - Public Specification Notice
This document is part of the Sigma Runtime Standard (SRS) public specification layer.Specification License: CC BY 4.0.
Implementation Safe Harbor: independent implementation permitted under public SRS/SRIP terms.
Machine-readable artifacts: Apache License 2.0 where explicitly marked.
Marks / Certification: governed by Sigma Marks and Certification Policy.
Proprietary Runtime Assets: not licensed by this SRIP.Independent implementations of public SRS/SRIP normative requirements are welcome under the public specification terms.
Product assets, protected Sigma marks, official certification, compatibility badges, CC BY-NC commercial use, and patent commitments use the relevant policy or explicit covenant. Independent implementation, attribution, or citation does not imply certification, endorsement, partnership, official compatibility, or permission to use Sigma marks as product identity.
Branch-Safe Unresolved State Handling, Contradiction Cooling, and Deferred Reintegration
| Field | Value |
|---|---|
| SRIP | SRIP-19 |
| Title | Recursive Contradiction Buffering (RCB) |
| Version | Public Draft v0.2 |
| Status | Public Draft |
| Date | 2026-05-21 |
| Authors / Contributors | Vladimir Ryabinskiy; Sigma Stratum Research Group (SSRG) |
| Owning Layer | Runtime Control / Multi-Agent Stability / Contradiction Buffering |
| Parent Specs | SRIP-03, SRIP-06, SRIP-09, SRIP-13, SRIP-17 |
| Related Specs | SRIP-04, SRIP-10, SRIP-11, SRIP-14, SRIP-15, SRIP-16, SRIP-18 |
| Specification License | CC BY 4.0 |
| Implementation Safe Harbor | Independent implementation permitted under public SRS/SRIP terms |
| Machine-Readable Artifacts | Apache 2.0 where explicitly marked |
| Marks / Certification | Governed by Sigma Marks and Certification Policy |
| Proprietary Runtime Assets | Not licensed by this SRIP |
| Independent Implementation | Permitted under the public specification terms |
| Commercial Runtime Boundary | Relevant policy or explicit covenant for protected Sigma marks, official certification, managed deployment, white-label, resale, CC BY-NC commercial use, and patent commitments |
| Information Class | Open |
| Change Class | Mixed SRS+SRD |
| Normative Status | Defines a public draft architecture contract for detecting, bounding, cooling, preserving, and later reintegrating unresolved contradictions without premature consensus collapse. It does not authorize fact deletion, deceptive smoothing, false synthesis, or medical or mental-health interpretation. |
| Conformance Level | Public Draft |
| SRD Synchronization Action | Deferred follow-up synchronization for runtime control, memory, safety, and multi-agent architecture explanation. |
| Release Alignment Status | Draft; no runtime enablement claim. |
Independent implementations of the public normative requirements in this SRIP are welcome under the applicable public specification terms.
No Sigma commercial runtime license is needed solely because an independent implementation follows those public normative requirements.
Product assets, protected Sigma marks, official certification, compatibility badges, CC BY-NC commercial use, and patent commitments use the relevant policy or explicit covenant. Independent implementation, attribution, or citation does not imply certification, endorsement, partnership, official compatibility, or permission to use Sigma marks as product identity.
SRIP-19 defines Recursive Contradiction Buffering (RCB): a runtime control
layer for handling incompatible but potentially valid states without forcing
immediate collapse into one winning interpretation.
RCB exists for long-horizon and multi-agent runtimes where contradiction can be
real, unresolved, and operationally important.
Its core principle is:
Contradiction is not automatically an error.
It is a multi-state coherence problem that must be bounded, cooled, traced,
and reintegrated only when phase conditions permit.
RCB allows a runtime to:
Recursive AI systems often encounter incompatible states:
Naive systems commonly respond by:
These behaviors create:
RCB defines a controlled alternative:
do not collapse incompatible attractors prematurely
Instead, the runtime may encapsulate an unresolved contradiction as a bounded,
traceable, low-amplification structure.
RCB does not define:
RCB is not a mechanism for making contradiction disappear.
RCB is a mechanism for preventing destructive contradiction handling.
A conformant RCB implementation must:
Prototype conformance may include:
Full conformance requires:
| Concept | Description |
|---|---|
| Contradiction Candidate | A detected incompatibility between claims, memories, agent outputs, retrieved evidence, constraints, or runtime interpretations. |
| Contradiction Node | A bounded unresolved structure preserving incompatible states without collapsing them into one answer. |
| Contradiction Energy | Estimated destabilization pressure of a contradiction, combining semantic divergence, recursive amplification risk, identity or boundary pressure, and control-posture strain. |
| Cooling | Runtime action that reduces recursive amplification pressure without deleting facts, provenance, or branch lineage. |
| Encapsulation | Process of making a contradiction bounded, traceable, and low-amplification while it remains unresolved. |
| Deferred Reintegration | Later attempt to resolve, merge, branch, or escalate a buffered contradiction when conditions are safer. |
| Branch-Safe State | Unresolved state that remains operationally available without contaminating canonical memory or identity. |
| Premature Coherence Collapse | Failure mode where the runtime forces a unified answer before the contradiction is safe or valid to resolve. |
| False Synthesis | User-facing or internal output that hides material incompatibility by presenting an artificial consensus. |
| Fossilized Contradiction | Long-retained unresolved contradiction that stops being reviewed and starts acting like hidden state. |
RCB operates between interpretation, stabilization, memory integration, and
multi-agent exchange.
Illustrative runtime position:
State Ingestion
-> Interpretation
-> Contradiction Detection
-> Contradiction Energy Estimation
-> Resolve Normally
-> Encapsulate / Cool / Defer
-> Branch
-> Escalate
-> Stabilization
-> Memory / Retrieval Integration
-> Generation / Verification
-> Field Update
RCB is not a generation layer. It is a control and state-management layer.
It interacts with:
Minimum RCB state should include:
| Field | Requirement |
|---|---|
contradiction_id |
Stable identifier for the contradiction node. |
status |
candidate, buffered, cooling, reopen_pending, reintegrated, branched, or escalated. |
sources |
Source records for conflicting claims, memories, agents, retrieval artifacts, or constraints. |
scope |
User, workspace, agent, session, channel, or runtime scope in which the contradiction is valid. |
energy |
Current contradiction-energy estimate. |
risk_vector |
Semantic divergence, amplification risk, identity pressure, safety pressure, provenance uncertainty, and branch impact. |
lineage |
Trace references, branch IDs, cycle IDs, or exchange envelope IDs. |
cooling_action |
Action taken to reduce amplification or dominance pressure. |
reopen_conditions |
Conditions under which the contradiction should be revisited. |
resolution_outcome |
resolved, preserved, split_branch, operator_escalated, or rejected_as_invalid. |
RCB state must not be treated as ordinary user memory by default.
Contradiction energy is a runtime estimate of how destabilizing a contradiction
is if processed immediately.
Minimum inputs should include:
Illustrative shape:
ContradictionEnergy:
semantic_divergence: 0.0..1.0
provenance_uncertainty: 0.0..1.0
amplification_risk: 0.0..1.0
identity_pressure: 0.0..1.0
safety_pressure: 0.0..1.0
branch_impact: 0.0..1.0
recommended_action: "resolve | buffer | branch | escalate"
Energy estimates are implementation-specific, but a conformant implementation
must record enough evidence to explain why immediate resolution was allowed,
buffered, branched, or escalated.
A conforming RCB flow should follow this sequence:
RCB must not delete conflicting evidence merely because the contradiction is
uncomfortable, inconvenient, or high-energy.
RCB must preserve memory integrity.
When contradiction involves memory or retrieval:
RCB may store:
RCB must not store:
RCB is especially important for multi-agent systems.
When multiple agents disagree, a runtime should not automatically:
Instead, RCB may treat disagreement as:
bounded unresolved multi-agent state
When used with SRIP-17 MAE:
SRIP-03 provides drift and instability language. RCB uses this evidence to
decide whether contradiction is likely to destabilize runtime coherence.
SRIP-06 has precedence over RCB. If a contradiction involves safety-critical
content, RCB may buffer evidence but must not delay required safety action.
RCB may operate as a soft pre-safe-mode layer: it can localize and cool conflict
before containment or recovery becomes necessary.
SRIP-09 provides branch lineage, trace ledger, cycle lineage, and recovery
events. RCB uses those primitives to store unresolved contradictions as
branch-safe state.
AEP helps prevent both fragmentation and crystallization. RCB should avoid
turning buffered contradictions into fossilized unresolved state. AEP signals
may help reopen or perturb stale contradictions.
CMT supports phase-aware recall, topology, and anchor facts. RCB must preserve
anchor facts and avoid replacing source facts with cooled narrative summaries.
RIS governs identity, relational, and participant boundaries. RCB uses RIS
signals when contradiction threatens identity continuity or participant scope.
RCB must not use buffering to permit ontology leakage or scope collapse.
RMI governs retrieved material. RCB may buffer contradictions involving retrieved
evidence, but it must not bypass RMI source trust, compression, or injection
rules.
ADP may help reopen a buffered contradiction through controlled perturbation.
RCB defines why immediate resolution was deferred; ADP defines how controlled
exploration may later test alternatives.
MAE defines exchange envelopes and local sovereignty. RCB defines how a runtime
handles incompatible exchanged artifacts without collapsing local state.
CSI may introduce conflicting commerce context through freshness, overlays,
scope, or operational grounding. RCB may buffer semantic contradictions, but it
must not become commerce decision authority.
RCB is subordinate to higher-priority controls.
Minimum precedence:
Safety > Legal / Policy Boundary > Identity / Participant Scope > Memory Provenance > Retrieval Scope > Contradiction Buffering > Generation
RCB must reduce, suspend, or escalate when a higher-priority layer requires
immediate action.
RCB must not be used to delay:
RCB should expose authorized diagnostics without leaking private control state.
Diagnostics may include:
User-facing behavior should remain natural and truthful.
When appropriate, the runtime may say that evidence is conflicting or unresolved.
It should not expose internal labels such as hidden control names, private
telemetry, or prompt instructions.
User-facing output must not present a false consensus where the runtime is only
buffering unresolved contradiction.
RCB introduces its own risks.
The runtime may overuse buffering to avoid resolution.
Mitigation:
Cooling may become factual distortion.
Mitigation:
Buffered contradictions may accumulate and influence behavior invisibly.
Mitigation:
Old unresolved contradictions may become permanent hidden assumptions.
Mitigation:
One agent may dominate the contradiction node.
Mitigation:
A runtime minimally conforms to SRIP-19 when it:
A runtime more fully conforms when it additionally:
| Capability | Status |
|---|---|
| Contradiction candidate detection | Draft / required for minimum conformance |
| Contradiction energy scoring | Draft / required for minimum conformance |
| Contradiction node state | Draft / required for full conformance |
| Branch-safe lineage | Draft / advanced |
| Cooling action evidence | Draft / required for full conformance |
| Deferred reintegration | Draft / required for minimum conformance |
| Multi-agent disagreement buffering | Draft / advanced |
| RMI / MAE integration | Draft / advanced |
| Stale contradiction review | Draft / advanced |
| User-visible unresolved-evidence language | Draft / optional, policy-bound |
If implemented within these boundaries, RCB should enable:
The intended outcome is not perfect truth resolution.
The intended outcome is safer continuity while truth, provenance, and branch
state remain unresolved.
SRIP-19 is a public draft architecture proposal. It does not claim that RCB is
implemented in any current Sigma Runtime release.
Any implementation claim must separately document:
End of SRIP-19 Public Draft v0.2
Sigma Stratum Research Group - 2026