CYB-002: Active Dampening Mechanisms for Cross-Repo Oscillation Control

Streeling University

CYB-002 · 2026-03-23

Generated by the Demerzel Cybernetics department — not reviewed by me yet. View source · My journal

Department: Cybernetics Module ID: CYB-002 Produced by: Seldon Plan Cycle cybernetics-2026-03-23-002 Belief: T (probable), confidence 0.83 Date: 2026-03-23 Prerequisite: CYB-001 (VSM-to-AI Governance Mapping)

Research Question

What active dampening mechanisms from cybernetics and control theory can prevent cross-repo oscillation in a file-based AI governance system?

Summary

Demerzel’s Galactic Protocol currently defines message formats and flows (directives, compliance reports, knowledge packages) but operates as an open-loop coordination system. It specifies what messages look like, not how to prevent oscillatory feedback between consumer repos (ix, tars, ga). Five classical control theory mechanisms — negative feedback, hysteresis, deadbands, rate limiting, and exponential backoff — can transform the Galactic Protocol from passive interface specification into an active System 2 (anti-oscillation coordinator) per Beer’s VSM.

The Oscillation Problem

What Oscillation Looks Like in Governance

Cross-repo oscillation occurs when state changes in one repo trigger reactions in others, which trigger further reactions, creating amplifying feedback loops:

ix detects gap → Demerzel issues directive → tars adjusts →
Demerzel detects tars drift → issues counter-directive →
ix re-adjusts → Demerzel detects ix drift → ...

This is the same instability problem that System 2 in Beer’s VSM was designed to prevent. In the Viable System Model, System 1 operational units (ix, tars, ga) are semi-autonomous but must not destabilize each other through uncoordinated reactions.

Why Static Contracts Are Insufficient

The Galactic Protocol’s six message types (directive, knowledge-package, compliance-report, belief-snapshot, learning-outcome, external-sync-envelope) define interfaces — the shapes of messages. But interfaces alone cannot prevent oscillation. A thermostat with a temperature sensor (interface) but no deadband (dampening) will cycle on and off continuously. Similarly, governance contracts without dampening will produce directive-compliance-directive loops.

Five Dampening Mechanisms

1. Negative Feedback (Closed-Loop Correction)

Control theory: The output of a system is fed back and subtracted from the input, producing self-correcting behavior that converges toward a setpoint.

Governance application: Every Galactic Protocol directive should include a target state and every compliance report should include a measured state. The difference (error signal) determines whether further directives are needed. If the error is shrinking, no new directive is issued — the system is converging.

Implementation:

• Directives include target_state field (what Demerzel wants)
• Compliance reports include measured_state field (what the repo achieved)
• Error = target_state - measured_state
• New directives only issue when error is growing or static, not when it is shrinking

VSM mapping: This transforms Galactic Protocol from open-loop (fire-and-forget directives) to closed-loop (directives corrected by compliance feedback).

2. Hysteresis (Threshold-Gated State Propagation)

Control theory: A system has different thresholds for activation vs. deactivation, creating a switching gap that prevents rapid toggling. A thermostat set to 20C might turn on heating at 19C and turn it off at 21C — the 2-degree gap is hysteresis.

Governance application: State changes in one repo should only propagate to others when they cross a significance threshold, and the threshold for “problem resolved” should differ from the threshold for “problem detected.”

Implementation:

• Detection threshold: belief confidence drops below 0.5 (triggers investigation)
• Resolution threshold: belief confidence rises above 0.7 (clears the flag)
• The 0.2 gap prevents: detect at 0.49 → fix to 0.51 → detect again at 0.49 → fix…
• Applied to: belief state changes, compliance scores, governance audit findings

VSM mapping: Hysteresis gives System 2 “memory” — it remembers whether the system was recently stable or recently unstable and adjusts its sensitivity accordingly.

3. Deadbands (Tolerance Zones)

Control theory: A region around the setpoint where no control action is taken. Small deviations are ignored, reducing actuator wear and preventing unnecessary corrections.

Governance application: Minor state changes in consumer repos should not trigger Galactic Protocol messages. A persona version bump from 1.0.0 to 1.0.1 (patch) should not trigger a governance directive, while 1.0.0 to 2.0.0 (major) should.

Implementation:

• Belief confidence changes < 0.05: no cross-repo propagation
• Policy compliance scores within +/-5% of target: no directive
• Persona patch versions: no governance reaction
• Knowledge state updates with < 3 new entries: batch, don’t propagate individually

VSM mapping: Deadbands reduce the variety of signals flowing through System 2, preventing coordination overload. This is a variety attenuator — it filters noise from the S1-to-S2 channel.

4. Rate Limiting (Bounded Update Frequency)

Control theory: The maximum rate at which a controller can issue corrections is bounded, preventing the controller from reacting faster than the system can respond.

Governance application: Demerzel should not issue more than N directives per repo per cycle. Consumer repos should not send more than M compliance reports per period. This prevents rapid-fire directive-response loops.

Implementation:

• Maximum directives per repo per PDCA cycle: 3
• Minimum interval between directives to same repo: 1 cycle
• Compliance report batching: aggregate into single report per cycle
• Knowledge package delivery: max 2 per repo per cycle

VSM mapping: Rate limiting matches System 2’s coordination cadence to System 1’s operational cadence. If the governance loop runs faster than operations can respond, directives pile up and oscillation amplifies.

5. Exponential Backoff (Adaptive Cooldown)

Control theory: After repeated failed corrections, the controller increases wait time exponentially before retrying, preventing resource exhaustion and giving the system time to stabilize.

Governance application: If a directive is issued and compliance is not achieved after one cycle, wait 2 cycles before re-issuing. If still non-compliant, wait 4 cycles. This prevents Demerzel from hammering a repo that may need structural changes (not just quick fixes).

Implementation:

• First non-compliance: re-issue directive next cycle
• Second non-compliance: wait 2 cycles, escalate severity
• Third non-compliance: wait 4 cycles, escalate to human
• Fourth non-compliance: halt automated directives, require human intervention
• Reset backoff to 0 on successful compliance

VSM mapping: Exponential backoff is a variety attenuator on the S3-to-S1 channel. It prevents the control system from overwhelming operations with repeated corrections that aren’t working.

Coordination Transparency Framework

The 2026 Springer paper “Coordination transparency: governing distributed agency in AI systems” provides academic validation for this approach through four components:

Component 1: Interaction Logging

Record every Galactic Protocol message with sender, receiver, timestamp, and content hash. Demerzel already partially supports this via Article 7 (Auditability), but the logs need to capture interaction patterns, not just individual messages.

Component 2: Live Coordination Monitoring

Track quantitative metrics that detect oscillation:

• Convergence index: Are compliance scores trending toward targets or oscillating?
• Oscillation index: Frequency of directive-counterdirective pairs within a window
• Policy-similarity drift: Are repos diverging in their governance compliance profiles?
• Cascade interdiction count: How often do dampening mechanisms prevent unnecessary actions?

Component 3: Intervention Hooks

Provide stop/pause/reroute capabilities at the coordination layer:

• Circuit breakers: If oscillation index exceeds threshold, halt cross-repo directives until human review
• Rate limiters: Enforce maximum directive frequency (see mechanism 4)
• Approval gates: High-impact directives require human confirmation

Component 4: Boundary Conditions

Constrain interaction topologies:

• Repos cannot directly trigger directives to each other (all coordination routes through Demerzel)
• Maximum directive chain depth (prevents A→B→C→A loops)
• Sandboxing: experimental governance changes apply to one repo before propagation

Open-Loop vs. Closed-Loop Comparison

Aspect	Current (Open-Loop)	With Dampening (Closed-Loop)
Directives	Fire-and-forget	Target state + error correction
State changes	All propagated	Filtered by deadband + hysteresis
Update frequency	Unlimited	Rate-limited per cycle
Repeated failures	Same directive re-issued	Exponential backoff + escalation
Oscillation detection	None	Convergence/oscillation indices
Intervention	Manual only	Circuit breakers + approval gates

Implications for Demerzel

1. Enhance Galactic Protocol — Add target_state to directives and measured_state to compliance reports, enabling closed-loop correction (negative feedback).
2. Define dampening parameters — Specify deadband widths, hysteresis gaps, rate limits, and backoff schedules as configurable governance parameters, not hardcoded values.
3. Add oscillation monitoring — Track convergence and oscillation indices across PDCA cycles. Store in state/coordination/oscillation-metrics.json.
4. Implement circuit breakers — If oscillation index exceeds threshold, halt automated directives and escalate to human. This is the governance equivalent of a fuse.
5. Preserve algedonic bypass — Dampening mechanisms must NOT apply to algedonic channel signals (CYB-001 Gap D, now resolved via policies/algedonic-channel-policy.yaml). Emergency bypass always overrides coordination dampening.

Relationship to CYB-001

This course directly addresses Gap B from CYB-001: “System 2 contracts are static, not actively dampening.” The five mechanisms transform the Galactic Protocol from a passive interface specification (open-loop) into an active anti-oscillation coordinator (closed-loop), fulfilling System 2’s core function in Beer’s VSM.

The algedonic channel (CYB-001 Gap D) was resolved separately via policies/algedonic-channel-policy.yaml. Dampening and algedonic bypass are complementary: dampening slows normal coordination to prevent oscillation; the algedonic channel bypasses all dampening for genuine emergencies.

Sources

• Beer, S. (1972). Brain of the Firm. Allen Lane.
• Beer, S. (1979). The Heart of Enterprise. John Wiley.
• Beer, S. (1985). Diagnosing the System for Organizations. John Wiley.
• Coordination transparency: governing distributed agency in AI systems. (2026). AI & Society, Springer. https://link.springer.com/article/10.1007/s00146-026-02853-w
• Gorelkin, M. (2025). “Stafford Beer’s VSM for Building Enterprise Agentic Systems.” Medium. https://medium.com/@magorelkin/stafford-beers-viable-system-model-for-building-enterprise-agentic-systems-81982d6f59c0
• Fearne, D. (2025). “Applying Stafford Beer’s VSM to Create The Autonomous AI Organisation.” Medium. https://medium.com/@fearney/applying-stafford-beers-viable-system-model-to-create-the-autonomous-ai-organisation-aaaed39b37e2
• IBM Research. (2025). “Agentic AI Needs a Systems Theory.”
• NI. (2025). “PID Theory Explained.” https://www.ni.com/en/shop/labview/pid-theory-explained.html
• GeeksforGeeks. (2025). “Feedback Loops in Distributed Systems.” https://www.geeksforgeeks.org/system-design/feedback-loops-in-distributed-systems/

Cross-References

• Prerequisite: state/streeling/courses/cybernetics/en/CYB-001-vsm-ai-governance-mapping.md
• Protocol: contracts/galactic-protocol.md
• Algedonic policy: policies/algedonic-channel-policy.yaml
• Department: state/streeling/departments/cybernetics.department.json
• Grammar: grammars/sci-cybernetics.ebnf
• Policy: policies/seldon-plan-policy.yaml