Pith. sign in

REVIEW 9 cited by

The Experience Compression Spectrum unifies memory, skills, and rules in LLM agents along increasing compression levels and identifies the absence of adaptive cross-level compression as the missing diagonal.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

arxiv 2604.15877 v2 pith:2L52BJPW submitted 2026-04-17 cs.AI cs.CLcs.MA

Experience Compression Spectrum: Unifying Memory, Skills, and Rules in LLM Agents

classification cs.AI cs.CLcs.MA
keywords compressionmemoryagentexperiencerulesskillsspectrumsystems
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

LLM agents accumulate long histories of interactions. Instead of keeping every detail, systems can compress that history in different ways. Episodic memory keeps fairly detailed records and compresses only modestly. Procedural skills turn repeated patterns into reusable procedures that compress more. Declarative rules extract general principles that compress the most. The paper places these three kinds of knowledge on one line ordered by compression ratio and shows that existing systems each sit at one fixed point on that line. No current system can move fluidly between levels when needed. The authors also note that the memory community and the skill-discovery community almost never cite each other even though they solve overlapping problems. Evaluation benchmarks are tied to whichever compression level a system uses, so it is hard to compare across levels. Transfer to new tasks improves as compression increases, but the knowledge becomes less specific. The paper ends by listing open problems for building agents that can manage the full range of compression.

Core claim

Mapping 20+ systems onto this spectrum reveals that every system operates at a fixed, predetermined compression level -- none supports adaptive cross-level compression, a gap we term the missing diagonal.

Load-bearing premise

That the compression ratios (5-20x, 50-500x, 1000x+) assigned to memory, skills, and rules are comparable across heterogeneous systems and that the low cross-community citation rate directly implies independent solving of shared sub-problems without solution exchange.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 2 invented entities

The paper rests on the domain assumption that compression can be measured uniformly across memory, skill, and rule systems and on the ad-hoc definition of the three compression bands; no free parameters are fitted and no new physical entities are postulated.

axioms (2)
  • domain assumption Memory, skills, and rules can be ordered along a single axis of increasing compression
    Invoked when the spectrum is proposed and when systems are mapped onto it.
  • domain assumption Low cross-citation rate indicates independent solution of shared sub-problems
    Used to interpret the 1,136-reference analysis.
invented entities (2)
  • Experience Compression Spectrum no independent evidence
    purpose: Unifying axis for memory, skills, and rules
    Newly proposed organizing framework
  • missing diagonal no independent evidence
    purpose: Label for the absence of adaptive cross-level compression
    New term for the identified gap

pith-pipeline@v0.9.0 · 5541 in / 1594 out tokens · 19632 ms · 2026-05-10T08:41:12.222339+00:00 · methodology

0 comments
read the original abstract

As LLM agents scale to long-horizon, multi-session deployments, efficiently managing accumulated experience becomes a critical bottleneck. Agent memory systems and agent skill discovery both address this challenge, extracting reusable knowledge from interaction traces, yet a citation analysis of 1{,}136 references across 22 primary papers reveals a cross-community citation rate below 1\%. We propose the \emph{Experience Compression Spectrum}, a unifying framework that positions memory, skills, and rules as points along a single axis of increasing compression (5--20$\times$ for episodic memory, 50--500$\times$ for procedural skills, 1{,}000$\times$+ for declarative rules), directly reducing context consumption, retrieval latency, and compute overhead. Mapping 20+ systems onto this spectrum reveals that every system operates at a fixed, predetermined compression level: none supports adaptive cross-level compression, a gap we term the \emph{missing diagonal}. We further show that specialization alone is insufficient (both communities independently solve shared sub-problems without exchanging solutions), that evaluation methods are tightly coupled to compression levels, that transferability increases with compression at the cost of specificity, and that knowledge lifecycle management remains largely neglected. We articulate open problems and design principles for scalable, full-spectrum agent learning systems.

Figures

Figures reproduced from arXiv: 2604.15877 by Bing Zhu, Guanghui Wang, Peiyang He, Wei Qiu, Xing Zhang, Yanwei Cui, Ziyuan Li.

Figure 1
Figure 1. Figure 1: The Experience Compression Spectrum. Existing agent learning systems map onto a single axis from raw traces to abstract rules. Memory systems cluster at Level 1, skill systems at Level 2, with Level 3 largely empty. A small number of cross-level systems (dashed) bridge Levels 1–2 but none support adaptive level selection. Compression ratios are approximate. that implies agent systems should perform upward … view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 9 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Co-Evolving Skill Generation and Policy Optimization

    cs.CL 2026-06 unverdicted novelty 7.0

    Framework estimates context-dependent marginal utility of candidate skills via reward gaps in matched base vs. skill-augmented rollouts to filter skills and co-train policy as generator.

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

    cs.DB 2026-06 unverdicted novelty 7.0

    TOKI types four common contradiction-resolution heuristics as bitemporal operators on a dual-row schema, supplies soundness theorems, and shows via a verdict matrix that it alone avoids three write-time anomalies whil...

  3. Library Drift: Diagnosing and Fixing a Silent Failure Mode in Self-Evolving LLM Skill Libraries

    cs.AI 2026-05 unverdicted novelty 7.0

    Identifies library drift as a failure mode in self-evolving LLM skill libraries and shows a governance recipe improves pass@1 from 0.258 to 0.584 on MBPP+ hard-100.

  4. Library Drift: Diagnosing and Fixing a Silent Failure Mode in Self-Evolving LLM Skill Libraries

    cs.AI 2026-05 unverdicted novelty 7.0

    The paper diagnoses library drift in self-evolving LLM skill libraries and demonstrates a governance recipe raising pass@1 from 0.258 to 0.584 on MBPP+ hard-100.

  5. Ratchet: A Minimal Hygiene Recipe for Self-Evolving LLM Agents

    cs.AI 2026-05 conditional novelty 6.0

    Ratchet provides a minimal hygiene recipe for self-managing skill libraries in frozen LLM agents, delivering +0.328 rolling-mean pass@1 gain on MBPP+ hard-100 and +0.22 peak lift on SWE-bench Verified.

  6. Dynamic Skill Lifecycle Management for Agentic Reinforcement Learning

    cs.LG 2026-05 unverdicted novelty 6.0

    SLIM dynamically optimizes active external skills in agentic RL via leave-one-skill-out marginal contribution estimates and three lifecycle operations, outperforming baselines by 7.1% on ALFWorld and SearchQA while sh...

  7. Trace2Policy: From Expert Behavior Traces to Self-Evolving Decision Agents

    cs.AI 2026-06 unverdicted novelty 5.0

    Trace2Policy's EISR iteratively refines expert-derived rules into compiled Python code reaching 79.6% accuracy on skewed compliance tasks, outperforming one-shot LLM distillation and a deployed LLM baseline.

  8. SkillsVote: Lifecycle Governance of Agent Skills from Collection, Recommendation to Evolution

    cs.CL 2026-05 unverdicted novelty 5.0

    SkillsVote is a governance system for agent skills that profiles corpora, recommends via search, and gates updates on successful reusable outcomes, yielding benchmark gains without model changes.

  9. Dynamic Skill Lifecycle Management for Agentic Reinforcement Learning

    cs.LG 2026-05 unverdicted novelty 5.0

    SLIM dynamically optimizes the active external skill set in agentic RL via leave-one-skill-out marginal contribution estimates and lifecycle operations, delivering a 7.1% average gain over baselines on ALFWorld and Se...