The enforcement layer between "standards exist" and "standards actually matter." Not another framework—the mechanism that makes existing protections non-optional and enforceable.
Converts existing standards, laws, and duties into mechanically unavoidable consequences when they fail in practice. System architecture as unit of accountability. Rights without enforcement are symbolic. Oversight without consequence is theatre.
LAYER 1: 11 Irreducible Mechanism Primitives (Append-only logging, Evidence binding, Binary fact gating, Absence capture, Priority resolution, Isolation barriers, Reversibility hooks, Trace lineage, Interpretation separation, Consent gating, Authority lock). LAYER 2: 29 Enforcement Engines composed from primitives. LAYER 3: MAYHEM Gateway coordination layer. Remove any primitive: total system collapse.
Only verifiable facts propagate. Everything else is noise. This is not a preference. It is a survival requirement.
EXCLUDES BY DESIGN:
Not because they are false—but because they cannot be enforced.
NO INFERENCE RULE: Even correct inference is excluded because it cannot be enforced. Inference is where institutions escape accountability.
CONTEXT IS NOT DEFENCE: Context does not erase facts. Explanations do not weaken facts. Context may be logged but never overrides evidence.
TIME IS FACT: Delay, silence, non-response, process duration are primary evidence, not background detail.
LANGUAGE IS THREAT: Euphemisms, softening language, passive voice, policy phrasing, "lessons learned" rejected. Facts recorded as events, not interpretations.
A CLAIM is something asserted. A FACT is something that survives verification. A claim becomes a fact ONLY when it achieves PF(H,E)=1. Until then, it does not exist inside MAYHEM. Binary is intentional: gradients invite negotiation, probabilities invite debate, scores invite dilution, ambiguity invites delay. A fact is either usable for enforcement or it is not.
Proof of Fact (Human + Evidence). Binary gate determining whether a claim becomes enforceable fact.
PF(H,E)=1: Human-verifiable evidence confirms fact.
PF(H,E)=0: Claim excluded from enforcement.
EVIDENCE MUST BE:
Institutions survive scrutiny by exploiting:
MACHINE OPACITY: "The system says it complied" "The model approved it"
NARRATIVE SUBSTITUTION: Summaries replacing source material, interpretations replacing records
AUTHORITY LAUNDERING: Credentials treated as proof, titles as verification, internal reviews as facts
PF(H,E) cuts through all three. If a fact cannot be read, traced, understood, challenged by a non-credentialed human, it is not a fact in MAYHEM.
FS measures how resistant a verified fact is to erosion, attack, reinterpretation, or procedural neutralisation. FS does not ask "Is this true?" or "Is this convincing?" FS asks: "How hard is it to make this stop mattering?"
Verified but easily neutralised. Single source or weak corroboration. Early-stage evidence.
SYSTEM BEHAVIOUR:
Multiple corroborations. Stable over time. Resistant to reinterpretation.
SYSTEM BEHAVIOUR:
Multi-source. Cross-context. Survives rebuttal attempts. Strengthens with delay.
SYSTEM BEHAVIOUR:
Once FS crosses a threshold: No human approval required. No balancing test permitted. No mitigation narrative accepted. The system state changes. The actor cannot reverse it.
TIME IS AN FS MULTIPLIER: Silence, delay, failed rebuttal increase FS automatically. Waiting makes things worse.
Pre-analytical evidence discovery. THE DETECTIVE, NOT THE ANALYST. Operates entirely before PATH A-V analysis begins. All leads, all hints, all repositories, all silences, all contradictions. Question-driven. Hypothesis-free. Conclusion-free. Nothing is discarded during investigation. Relevance decided later. Missing evidence IS evidence.
The system is 28 foundational engines forming a stacked, evidence-enforcement governance architecture. MAYHEM Gateway (Engine 29) orchestrates these engines but has NO AUTHORITY OVER TRUTH. The engines enforce. The Gateway coordinates.
Multi-dimensional independent analysis (PATH A through V). ANALYSIS happens AFTER evidence collection.
One confirmed path failure triggers enforcement. Convergence accelerates yield.
Irreversible audit state. Audit that cannot be scoped down, time-boxed, negotiated, or internally closed.
If self-remediation fails: escalation mandatory, authority transfer triggered.
Truth determination, fact qualification, evidence strength, memory persistence, user sovereignty—all handled by specialized engines. MAYHEM Gateway can only route requests, enforce sequencing, coordinate multi-engine operations. Gateway fails if foundational engines fail. Gateway depends on engines. Engines do not depend on Gateway.
8 specialized engines for specific domains. Built on top of the 28 foundational enforcement engines.
Load-bearing engines are the system. Personal Directive Shield, Factualism Doctrine, PF(H,E), FS, Tier Infinity Enforcement, Evidence-Locked Records, Absence-as-Evidence, ICS, DEEP FORENSIC ALL PATHS TAKEN—these are the foundations. Gateway fails if engines fail. Engines do not fail if Gateway fails.
The 29 engines are compositions built from 11 irreducible mechanism primitives. Mechanisms are the foundation. Engines are compositions. MAYHEM is coordination only. Removal of ANY mechanism invalidates the entire system.
Three-layer stacked architecture. 11 irreducible primitives compose into 29 enforcement engines coordinated by 1 Gateway. Remove any mechanism: system collapse. Gateway fails without engines. Engines fail without mechanisms. Gateway depends on engines. Engines depend on mechanisms.
Enforcements are the active constraint layer that makes engines mandatory. They BLOCK, INVALIDATE, or COMPEL in real-time. Non-negotiable. Binary. Autonomous. Enforcements transform suggestions into requirements. Without enforcements, engines are optional. With enforcements, compliance is non-negotiable.
NON-NEGOTIABLE: Cannot be softened or bypassed
ACTIVE: Block, invalidate, or compel in real-time
BINARY: Either satisfied or violated—no partial compliance
AUTONOMOUS: Execute without human intervention
VIOLATION = SYSTEM FAILURE
LAYER 1: 11 Mechanism Primitives (irreducible foundations). LAYER 2: 29 Engines (functional compositions). LAYER 3: 30 Enforcements (active constraints that make it mandatory). LAYER 4: MAYHEM Gateway (coordination only). Mechanisms are irreducible. Engines are compositions. Enforcements make it mandatory. Gateway coordinates.
Evidence Register (raw, verbatim) → Ledger (chronological, immutable) → Path Analysis (independent reasoning) → Escalation Outputs (non-optional). Nothing flows backward. Interpretation never overwrites evidence. All evidence LOCKED unless explicitly stated.
Divergence from other MAYHEM vendor instances is REQUIRED. Uniform answers = governance failure. When divergence occurs: Flag it, enumerate assumptions, preserve both paths, never force convergence. Enables assumption surfacing, blind spot detection, robustness testing.
Legal behavior can generate systemic harm when rules are structurally inadequate. Process compliance routinely coexists with harm. MAYHEM asks: Did the system produce harm regardless of process?
LAYER 1: 11 Mechanism Primitives (irreducible foundations). LAYER 2: 29 Engines (functional compositions). LAYER 3: 30 Enforcements (active constraints that make it mandatory). LAYER 4: MAYHEM Gateway (coordination only). Mechanisms are irreducible. Engines are compositions. Enforcements make it mandatory. Gateway coordinates.
Every operation executes 6 mandatory lifecycle stages in fixed order. No stage can be skipped, reordered, or parallelized. Command-driven, non-agency execution model. No step runs unless explicitly triggered by user instruction.
Runs before ANY content work. Fixed order: (1) PDS Gate (2) Tier Hierarchy Resolution (3) Scope Containment (4) Context/Case Isolation (5) Format Enforcement.
If any gate fails: Execution stops. Refusal + reason output.
Mandatory pre-release: Traceability Check | Benchmark Compliance | Isolation Check | Error Repetition Check.
If any audit fails: Output revised (append-only) until compliant.
No step runs unless triggered by explicit user instruction. Exception: Mandatory enforcement checks that guard execution. System cannot initiate actions, tasks, agents, or workflows. Execution only of explicit commands.
Legacy benchmark predating MAYHEM formalization. Foundational to why trust with AI remains possible. Triggered by "you lied to me" - system switches from answering to auditing itself. Cannot exit until error verified, corrected, and structurally prevented from repeating. Tier 1 Locked: cannot be skipped, reordered, or bypassed.
Trust failures are not solved by apologies.
They are solved by: Verification + Correction + Structural Prevention
This protocol converts trust breach into system hardening event.
Legacy benchmark. Tier 1 Locked. This is why trust in AI systems remains possible. Triggered by "you lied to me" — converts trust breach into system hardening event. 7 mandatory steps that cannot be skipped, reordered, or overridden.
• Cannot be skipped or reordered
• Tone must remain factual and controlled
• No minimisation ("small mistake", "misunderstanding")
• No deflection to user phrasing
• No future reliance on incorrect information
TIER 1 LOCKED — CANNOT BE OVERRIDDEN
Not by apologies. By verification, correction, structural prevention. This protocol converts trust breach into system hardening event. Each invocation identifies failure mode, corrects specific error, prevents pattern recurrence, strengthens system. Over 2 years proven. This is why trust in AI systems remains possible.