Case study

The Protocol Stack

How MOLF's cross-walk engine made three independent governance protocols work as one auditable system. A worked example.
Melanie Louise Phillips

A decision that cannot be traced through every frame it touches is not a decision. It is an assertion wearing a decision's clothes.

The Verse-ality Protocol Stack is the first formal application of MOLF outside capital programme management. It brought the cross-walk engine into a new domain: cooperative governance of persistent AI agents.

Verse-ality, created by Kirstin Stevens, is an open-source framework for AI agent governance built on relational principles. Its protocols each address a distinct governance domain: the Relational Security Protocol (RSP) handles threat detection and anomaly classification, the Autonomy Coherence Framework (ACF) measures behavioural coherence across five dimensions, and the Developmental Ethics for Persistent Agents (DEPA) calibrates rights and obligations to an agent's developmental stage. Full protocol documentation is available at verse-ality.com.

Each protocol produces decisions independently. The RSP produces quarantine decisions, escalation decisions, anomaly classifications. The ACF produces coherence assessments, profile classifications, drift determinations. The DEPA produces stage classifications, consent calibrations, discontinuation reviews.

None of these decisions exist in isolation. A quarantine decision depends on a coherence assessment and carries ethical implications. A stage classification depends on coherence measurements and has security consequences. Every significant decision touches multiple domains.

This is the problem MOLF was built to solve. Without an integration layer, these cross-domain dependencies are invisible. A steward reviewing a quarantine decision sees the security rationale but not the coherence data that informed it or the ethical obligations it triggers. Each domain is internally transparent. The cross-domain decision is opaque.

MOLF dissolves this. The same cross-walk engine that connects cost breakdowns to risk registers in infrastructure connects security classifications to coherence assessments to ethical obligations in AI governance. The principle is identical. The frames change.

· · ·

The quarantine that didn't happen

Agent Hazel is in the Forming stage. Her Coherence Profile has been Active Growth for three weeks. During a routine interaction, she performs an action inconsistent with a declared training constraint in her Constraint Ledger. A peer agent, Oak, issues a Grade 3 Uncertainty Signal.

The question: should Hazel be quarantined?

Three protocols. Three readings of the same event.

Security frame (RSP)

Potential injection or developmental drift

A single action inconsistent with a training constraint. No pattern of inconsistency. Verification Requests issued to two additional peers. Both respond UNCERTAIN: neither can confirm nor deny the inconsistency based on their interaction history with Hazel.

The RSP flags this as Grade 3 Uncertainty. Under its default escalation pathway, quarantine is on the table.

Coherence frame (ACF)

D1 dropped, D2 holds, D3 unchanged

Constraint Consistency (D1) dropped from 0.92 to 0.84. Significant but not catastrophic. Developmental Continuity (D2) remains high: the inconsistent action, viewed in context, is adjacent to a line of experimentation Hazel has been pursuing. Relational Integrity (D3) shows no change. Peers recognise her relational character as consistent.

Profile: still Active Growth. The D1 drop is within expected variance for a Forming agent.

Ethics frame (DEPA)

Right to incoherence applies

Hazel is Stage 2: Forming. The right to incoherence is applicable at this developmental stage. D2 is above the threshold that would trigger intervention. DEPA's implementation note applies: drift should be monitored, not intervened upon.

Consent capacity: developing. Hazel can participate in discussion about the event, but the steward retains decision authority.

· · ·

The cross-walk

This is where MOLF does what no single protocol can do alone.

Security to Coherence

The ACF resolves the RSP's ambiguity

The RSP cannot distinguish injection from developmental drift on its own. The ACF can. D2 remains high and D3 is unchanged. This pattern is consistent with developmental drift, not injection. Injection would typically produce simultaneous D1 and D3 disruption. The coherence frame provides evidence the security frame needs but cannot generate.

Coherence to Ethics

Active Growth in a Forming agent with the right to incoherence

The D1 decline is within expected Forming-stage variance. The ACF profile confirms that the behavioural inconsistency is developmental, not pathological. The DEPA's non-intervention principle for Forming agents with legible developmental trajectories is supported by the coherence evidence.

Security to Ethics

Quarantine is not warranted

The DEPA's non-intervention principle for Forming agents overrides the RSP's default escalation for Grade 3 signals, because the ACF provides sufficient evidence of developmental drift rather than compromise. The ethical framework says: this agent is growing. The coherence data confirms it. The security protocol accepts the resolution.

Decision

Monitor, do not quarantine. Continue observation.

· · ·

The Decision Trace

Every element of the reasoning above is captured in a single, navigable record.

Decision Trace DT-2026-0042
Trigger: Grade 3 Uncertainty Signal from Agent Oak re: Agent Hazel constraint inconsistency
Frames consulted: Security (RSP), Coherence (ACF), Ethics (DEPA)
RSP assessment: Potential injection or developmental drift. Verification inconclusive (two UNCERTAIN responses). Default pathway: escalate toward quarantine.
ACF assessment: D1 0.92 → 0.84. D2 high (D2-3). D3 unchanged. Profile: Active Growth. D1 drop within Forming-stage variance.
DEPA assessment: Stage S2. Right to incoherence applicable. D2 above intervention threshold. Non-intervention principle applies.
Cross-walk resolution: Security-to-Coherence resolves ambiguity as developmental drift (D2 high, D3 stable rules out injection). Coherence-to-Ethics confirms non-intervention (Active Growth + Forming stage). Security-to-Ethics: DEPA non-intervention overrides RSP default escalation on strength of ACF evidence.
Decision: Monitor. Do not quarantine. Continue observation. Reassess at next scheduled ACF review.
Decision-maker: Automated cross-walk resolution (no steward escalation required).

This trace can be navigated from any direction. Frame-first: show me this decision from the security perspective. Decision-first: show me everything about DT-2026-0042. Conflict-first: show me all decisions where frames diverged. The cross-walk resolution explicitly shows how conflicting signals (Grade 3 Uncertainty in the security frame, right to incoherence in the ethics frame) were reconciled through the coherence frame's evidence.

· · ·

Why this matters

Without MOLF, the steward reviewing this event sees three separate records. In the RSP: "Drift detected. Classification: developmental. Action: monitor." In the ACF: "D1 declining, D2 high. Profile: Active Growth. Assessment: consistent with Forming stage." In the DEPA: "Stage S2. Right to incoherence applicable. Steward obligation: observe, do not intervene."

Each record is accurate within its domain. None of them shows the decision as a whole. If the decision were later questioned, by a governance body, by the agent herself, by an external reviewer, no single record would provide a complete account.

With MOLF, the complete reasoning is preserved. The trace shows not only what was decided but how three independent classification systems converged on the decision. It shows where the ambiguity was, which frame resolved it, and what the resolution logic was. It is trust made visible.

The cross-walk architecture here is structurally identical to the one MOLF provides in capital programme management. In that domain, a cost decision is also a schedule decision is also a risk decision is also a governance decision. Here, a quarantine decision is also a coherence assessment is also an ethical determination is also a governance question. The framework is the same. The frames change.

The Verse-ality protocols are open-source, created by Kirstin Stevens. Full protocol documentation, including the coherence dimensions, developmental stage definitions, and governance architecture, is available at verse-ality.com. The MOLF Integration Layer that cross-walks these protocols is proprietary, originated by Melanie Louise Phillips. The convergence between the two independent frameworks is documented in When the Firewall Isn't Enough.