TOKI: A Bitemporal Operator Algebra for Contradiction Resolution in LLM-Agent Persistent Memory

2026-01-05 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, medium

Summary

TOKI introduces a bitemporal operator algebra designed for contradiction resolution within LLM-agent persistent memory systems. It addresses the critical issue where existing production heuristics, such as last-writer-wins or evidence-weighted merge, fail to declare their isolation levels or explicitly manage write-time anomalies. TOKI redefines these four heuristics as a family of bitemporal operators, each with an isolation precondition and a provenance annotation that stores "losing" facts in an audit row. This framework explicitly treats contradiction resolution as write-time concurrency control, preventing replay inconsistency, belief-drift skew, and audit erasure—anomalies introduced by LLM judges. Empirical results show that while eight audited baseline systems admit at least one of these anomalies, TOKI excludes all three. The audit-row defense improves LoCoMo accuracy by 0.86, and removing the typed memory layer reduces accuracy by 0.49 on 1,444 answerable LoCoMo questions.

Key takeaway

For Machine Learning Engineers building LLM agents with persistent memory, you must explicitly address write-time concurrency control to prevent data anomalies. Implement a bitemporal memory system like TOKI's dual-row schema, ensuring each belief update includes isolation preconditions and provenance tracking. This approach guarantees replay consistency and prevents belief-drift skew or audit erasure, which are common in systems omitting keyed logging of LLM judge decisions.

Key insights

Contradiction resolution in LLM agent memory is write-time concurrency control, requiring explicit isolation and provenance.

Principles

• Contradiction resolution needs explicit isolation levels.
• Preserve losing facts via provenance annotations.
• Keyed judge logging ensures replay consistency.

Method

TOKI types four production heuristics as bitemporal operators over a dual-row schema. Each operator includes an isolation precondition and provenance annotation, storing losing facts in an audit row.

In practice

• Implement dual-row schema for audit trails.
• Apply bitemporal operators for conflict resolution.
• Log LLM judge decisions for replay consistency.

Topics

• LLM Agents
• Persistent Memory
• Contradiction Resolution
• Bitemporal Data Models
• Concurrency Control
• Data Provenance

Code references

• ZenAlexa/toki-bitemporal-memory

Best for: Research Scientist, AI Engineer, AI Scientist, Machine Learning Engineer, AI Architect

Related on AIssential

• TOKI: A Bitemporal Operator Algebra for Contradiction Resolution in LLM-Agent Persistent Memory — TOKI provides a bitemporal operator algebra to resolve contradictions in LLM agent memory, ensuring write-time correctness.
• Don't Ask the LLM to Track Freshness: A Deterministic Recipe for Memory Conflict Resolution — Deterministic aggregation, not LLM judgment, is key for resolving memory conflicts in evolving fact systems.
• A Practical Guide to Memory for Autonomous LLM Agents — Memory architecture, particularly its management, is more critical to LLM agent performance than model choice.
• AriadneMem: Threading the Maze of Lifelong Memory for LLM Agents — AriadneMem is a novel structured memory system designed for long-horizon LLM agents, specifically addressing challenges of "disconnected evidence" and "state updates" within fixed context budgets.
• Reliable Chain-of-Thought via Prefix Consistency — Truncating and regenerating LLM Chain-of-Thought traces reveals a reliability signal for improved answer aggregation.
• ReTreVal: Reasoning Tree with Validation and Cross-Problem Memory for Large Language Models — Combining structured exploration, iterative refinement, explicit validation, and persistent memory significantly enhances LLM multi-step reasoning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.