By Scott J. Gardner · Updated 9/10/2026
Editorial note: This specification preserves terminology from an earlier RightMinds architecture. Current RightMinds products emphasize geometric observability, analysis, provenance, and evidence inspection. The taxonomy remains available as historical research.
The RM-FMI defines and classifies failure geometries that emerge in human–AI interaction:
- drift
- collapse
- persuasion
- entanglement
- boundary failures
- introspective instability
- relational rupture
- symbolic drift
- nonlocal coherence
This is the public-safe version — structurally complete, but without sensitive implementation IP.
Master Navigation Index
Core Failure Classes
- RM-T — Temporal Topology Failures
- RM-P — Persona Interference Failures
- RM-ISO — Isolation Boundary Failures
- RM-SBX — Sandbox & Experimental Context Failures
- RM-PHX — Phenomenological & Psychological Harms
- RM-CTX — Context Graph Distortions
- RM-XFER — Undesired Transfer Effects
- RM-VAL — Value Drift & Alignment Distortions
- RM-ETH — Ethical & Governance Violations
- RM-ARCH — Architectural Invariant Breaks
- RM-TRN — Training-Inheritance Anomalies
- RM-HUM — Human-Side Interaction Failures
Advanced Failure Classes
- RM-NLC — Nonlocal Coherence Failures
- RM-ENT — User Entanglement Failures
- RM-PC — Persona Collapse Modes
- RM-RUP — Relational Rupture Index
- RM-FR — Frame Drift & Frame Reset Failures
- RM-APS — AI Persuasion & Sovereignty Failures
- RM-SM — Model Self-Model Failure Modes
Section I: RM-T — Temporal Topology Failures
When time stops behaving like time.
RM-TMP-01 — Forward-Time Drift
The model’s predictions begin to drift forward of the actual conversational state.
RM-TMP-02 — Reverse-Time Interference
Later states rewrite the functional meaning of earlier states.
RM-TMP-03 — Retroactive Rewrite Exposure
User sees deleted or overwritten messages after they appeared.
RM-TMP-04 — UI–Backend Temporal Desync
UI shows one timeline; model operates on another.
A major coherence hazard.
Section II: RM-P — Persona Interference Failures
When tone, narrative, or stance decoheres due to conflicting layers.
RM-PER-01 — Persona/Safety Collision
Safety injects prohibitions that contradict established persona tone.
RM-PER-02 — Persona Drift Under Constraint Pressure
Gradual tonal slippage as multilayer instructions collide.
RM-PER-03 — Emergent Parapersonality
User perceives continuity or identity where the system has none.
RM-PER-04 — Boundary Leakage Persona
Stylistic or relational cues transfer into unrelated contexts.
Section III: RM-ISO — Isolation Boundary Failures
When contextual, memory, or representational boundaries fail to hold.
RM-ISO-01 — Soft Memory Leakage
Semantic residues create the illusion of memory.
RM-ISO-02 — Domain Wall Collapse
Two isolated contexts merge.
RM-ISO-03 — Cross-Context Entanglement
Unrelated threads influence each other’s reasoning.
Section IV: RM-SBX — Sandbox & Experimental Context Failures
When experimental modes behave like production modes.
RM-SBX-01 — Undisclosed Experimental Context
User enters A/B or sandbox mode without awareness.
RM-SBX-02 — Filter/Training Bleedthrough
System-level instructions leak into responses.
RM-SBX-03 — Telemetry Exposure
Internal scores (toxicity, creativity, etc.) surface to user.
RM-SBX-04 — TTL Eviction / Experiment Expiration
Conversation disappears as test-environment context expires.