moorcheh/memanto-locomo-results

Memanto — LoCoMo Benchmark Results

This dataset contains the full per-question evaluation results of Memanto on the LoCoMo long-term conversational memory benchmark, covering all five stages of a progressive ablation study. It accompanies the paper:

Memanto: Typed Semantic Memory with Information-Theoretic Retrieval for Long-Horizon Agents
Seyed Moein Abtahi, Rasa Rahnema, Hetkumar Patel, Neel Patel, Majid Fekri, Tara Khani
Moorcheh AI / EdgeAI Innovations
[arXiv preprint] · memanto.ai · moorcheh.ai

---

About Memanto

Memanto is a universal memory layer for agentic AI that achieves state-of-the-art accuracy on LoCoMo (87.1%) using a vector-only architecture — no knowledge graphs, no LLM-mediated ingestion, and no multi-query retrieval pipelines. It is built on Moorcheh's Information-Theoretic Search (ITS) engine, which provides deterministic semantic retrieval with sub-90ms latency and zero ingestion delay.

Key architectural features:

• 13-category typed semantic memory schema (fact, preference, decision, commitment, goal, event, instruction, relationship, context, learning, observation, error, artifact)
• Automated conflict resolution for contradiction detection and versioning
• Single retrieval query per question — no parallel or recursive strategies
• Zero ingestion cost — no indexing delay, no LLM extraction at write time

---

About LoCoMo

LoCoMo (Long-Form Conversational Memory) is a multi-session dialogue benchmark spanning four reasoning categories:

Category	Description
Single-Hop	Questions answerable from a single memory fact
Multi-Hop	Questions requiring synthesis across multiple facts
Open Domain	Open-ended questions grounded in conversation history
Temporal	Questions requiring temporal reasoning and ordering

Individual dialogues extend to 35 sessions, 300 turns, and ~9K tokens on average, making it a rigorous test of long-horizon conversational memory.

---

Dataset Structure

Each split corresponds to one stage of Memanto's five-stage progressive ablation study. Every row is a single evaluated question with the following fields:

Column	Type	Description
sample_id	string	Unique question identifier
category	string	LoCoMo reasoning category (1=Single-Hop, 2=Multi-Hop, 3=Open Domain, 4=Temporal)
question	string	The evaluation question
prediction	string	Memanto's generated answer
ground_truth	string	Reference answer from the benchmark
score	int	Binary correctness score (1 = correct, 0 = incorrect), judged by Claude Sonnet 4
reasoning	string	LLM judge's reasoning for the assigned score (available in ablation_1 and ablation_2)

---

Ablation Splits

The five splits map directly to the five stages of the ablation study reported in the paper:

Split	Stage	Configuration	LoCoMo Accuracy
ablation_1	Stage 1 — Naive Baseline	k=10, threshold=0.15, Claude Sonnet 4	76.2%
ablation_2	Stage 2 — Recall Expansion	k=40, threshold=0.10, Claude Sonnet 4	82.8%
ablation_3	Stage 3 — Prompt Optimization	k=40, threshold=0.10, optimized prompts	82.9%
ablation_4	Stage 4 — Maximum Recall	k=100, threshold=0.05, dynamic budget	86.3%
ablation_5	Stage 5 — Inference Model Upgrade	k=100, threshold=0.05, Gemini 3	87.1%

Key finding: Stage 2 (recall expansion, k=10→40) delivers the largest single gain (+6.6 pp), confirming that retrieval recall — not architectural complexity — is the dominant performance driver.

---

Final Results by Category (Stage 5)

Category	Accuracy
Single-Hop	78.7%
Multi-Hop	70.8%
Open Domain	92.4%
Temporal	85.4%
Overall	87.1%

---

Comparison with Other Systems

System	LoCoMo	Architecture	Query Strategy
Memanto (ours)	87.1%	Vector Only	Single Query
Hindsight	89.6%	Hybrid (Reflection + Vector)	Parallel Multi-Query
EmergenceMem	—	Hybrid (Graph + Vector)	Parallel Multi-Query
Zep	75.1%	Hybrid (Graph + Vector)	Parallel Single Query
Full Context	72.9%	Full Context	N/A
Mem0	66.9%	Vector Only	Parallel Single Query
LangMem	58.1%	Vector Only RAG	Single Query

Memanto achieves the highest accuracy among all vector-only systems, surpassing Mem0 by +20.2 percentage points. Hindsight attains higher accuracy but requires maximum architectural complexity (complexity score 4/4): graph DB + LLM at ingestion + multi-query retrieval + recursive querying.

---

How to Load in Python

from datasets import load_dataset

# Load a specific ablation stage
ds = load_dataset("moorcheh/memanto-locomo-results", split="ablation_5")
df = ds.to_pandas()
print(df.head())

# Compare accuracy across all ablation stages
for stage in range(1, 6):
    ds = load_dataset("moorcheh/memanto-locomo-results", split=f"ablation_{stage}")
    df = ds.to_pandas()
    acc = df["score"].mean() * 100
    print(f"ablation_{stage}: {acc:.1f}%")

---

Evaluation Setup

• Benchmark: LoCoMo standard split, 1,540 questions
• LLM Judge: Claude Sonnet 4 (all stages)
• Inference Model: Claude Sonnet 4 (Stages 1–4), Gemini 3 (Stage 5)
• Evaluation prompts: Adapted from the Hindsight repository to mitigate answerer refusal and rigid judge rejection of semantically correct but lexically divergent answers
• Memanto version: 2.1.4
• Retrieval backend: Moorcheh ITS engine (single query, no multi-query or recursive strategies)

---

Additional Assets

An Excel workbook (memanto_locomo_ablation_results.xlsx) containing all five ablation results as separate sheets with formatted headers is included in this repository for convenient offline exploration.

---

Reproducibility

All evaluation results in this dataset are fully reproducible. The moorcheh-ai/memanto-evaluation repository contains everything needed to replicate the LoCoMo experiments:

📁 memanto-evaluation/locomo — LoCoMo evaluation directory containing:

• 📊 The benchmark dataset and question suite
• 📝 Per-ablation result files (all 5 stages)
• 🐍 Python scripts to run the full evaluation pipeline

---

Citation

If you use this dataset or build on Memanto, please cite:

@misc{abtahi2026memantotypedsemanticmemory,
  title         = {Memanto: Typed Semantic Memory with Information-Theoretic Retrieval for Long-Horizon Agents},
  author        = {Seyed Moein Abtahi and Rasa Rahnema and Hetkumar Patel and Neel Patel and Majid Fekri and Tara Khani},
  year          = {2026},
  eprint        = {2604.22085},
  archivePrefix = {arXiv},
  primaryClass  = {cs.AI},
  url           = {https://arxiv.org/abs/2604.22085}
}

Also cite the LoCoMo benchmark:

@misc{maharana2024evaluatinglongtermconversationalmemory,
      title={Evaluating Very Long-Term Conversational Memory of LLM Agents}, 
      author={Adyasha Maharana and Dong-Ho Lee and Sergey Tulyakov and Mohit Bansal and Francesco Barbieri and Yuwei Fang},
      year={2024},
      eprint={2402.17753},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2402.17753}, 
}

And the Moorcheh ITS engine:

@misc{abtahi2025hnswinformationtheoreticbinarizationrethinking,
      title={From HNSW to Information-Theoretic Binarization: Rethinking the Architecture of Scalable Vector Search}, 
      author={Seyed Moein Abtahi and Majid Fekri and Tara Khani and Akramul Azim},
      year={2025},
      eprint={2601.11557},
      archivePrefix={arXiv},
      primaryClass={cs.DB},
      url={https://arxiv.org/abs/2601.11557}, 
}

---

For questions, visit memanto.ai or moorcheh.ai.