01 · Cognition
Reasoning, planning & cognitive architecture
Multi-step reasoning, deliberation, planning, decomposition, metacognition, cognitive control, hybrid architectures, and general-purpose problem solving.
Open research community
Autonomous Reasoning, Cognitive Systems & Agentic AI
ARCS is a curated research community for people studying systems that perceive, remember, reason, plan, learn, act, and improve over time.
Academic, nonprofit, industry, and independent researchers are welcome. ARCS does not require agreement on a single architecture, model family, institution, or definition of artificial general intelligence.
Research scope
A home for work on persistent and increasingly capable intelligence.
ARCS spans theoretical, empirical, and implemented research across cognitive architectures and autonomous systems, with particular interest in work that exposes mechanisms, evidence, failure modes, and evaluation methods.
02 · Persistence
Memory, world models & continuity
Episodic and semantic memory, durable state, retrieval, world modeling, temporal continuity, long-horizon operation, and identity across sessions or processes.
03 · Agency
Agentic AI, tool use & action
Software agents, desktop and browser agents, embodied agents, tool selection, environment interaction, multi-agent coordination, and real-world task execution.
04 · Learning
Continual learning, transfer & improvement
Experience reuse, adaptation, skill acquisition, transfer, recovery from failure, self-modeling, governed self-improvement, and learning without catastrophic drift.
05 · Assurance
Verification, governance & interpretability
Outcome verification, authority boundaries, provenance, auditability, interpretability, safe autonomy, reversible control, and evidence-backed capability claims.
06 · Measurement
Evaluation, benchmarks & reproducibility
Agent benchmarks, long-horizon evaluation, reliability studies, replication, ablation, negative results, experimental protocols, datasets, and reproducible evidence.
Autonomous reasoningAgentic AICognitive systemsMemory systemsPlanningWorld modelsContinual learningDesktop agentsEmbodied agentsVerificationInterpretabilityAGI evaluation
Contribute
Deposit the work, not just the conclusion.
ARCS welcomes research artifacts that help other investigators inspect, reproduce, challenge, extend, or compare the work.
Accepted research objects
More than papers
Research papers, preprints, technical reports, software releases, datasets, benchmarks, evaluation results, replication studies, negative results, experimental protocols, and other reproducible research artifacts are in scope.
Strong deposits
Metadata + evidence + context
Use clear titles and abstracts, identify creators and affiliations accurately, include ORCID identifiers where available, describe versions and limitations, choose an appropriate license, and provide enough documentation for others to understand the artifact.
1
Prepare
Package the paper, software, dataset, benchmark, protocol, or other research object with useful documentation.
2
Describe
Add accurate title, abstract, creators, identifiers, keywords, dates, version, license, and related resources in Zenodo.
3
Select ARCS
Choose the Autonomous Reasoning, Cognitive Systems & Agentic AI community when creating or editing the Zenodo record.
4
Submit
Send the record for community review. Curators may request metadata, documentation, or scope corrections before inclusion.
Curation principles
Open participation with a technical quality floor.
ARCS is intended to become a useful research collection, not a generic upload directory. Curation focuses on relevance, research substance, discoverability, and transparent documentation.
Relevance
Fit the research scope
The work should materially address autonomous reasoning, cognitive systems, agentic AI, or a closely related research problem.
Traceability
Make the object legible
Metadata should identify what the artifact is, who created it, what version it represents, and how it relates to prior or accompanying work.
Substance
Show enough to evaluate
Claims should be accompanied by methods, evidence, implementation detail, data, evaluation, or a clearly bounded theoretical argument appropriate to the contribution.
Limitations
Separate evidence from aspiration
Unverified capability claims, unsupported generalization, and promotional framing should not substitute for documented results or explicit research hypotheses.
Reusability
Help the next researcher
Where practical, include code, data, protocols, README material, environment information, or other artifacts that improve reproducibility and reuse.
Pluralism
No single doctrine required
ARCS is architecture-agnostic and model-agnostic. Competing hypotheses, replication attempts, null results, and critical work are welcome when technically substantive.
Important boundary: acceptance into ARCS indicates that a record fits the community's scope and curation standards. It does not constitute peer review, independent replication, endorsement of the conclusions, or certification of a system's capabilities.
Why ARCS exists
Serious autonomous-systems research is happening across institutional boundaries.
Important work now appears in universities, industry labs, open-source projects, independent research programs, software repositories, preprint servers, benchmarks, and technical reports. ARCS provides a persistent DOI-backed collection where those research objects can be discovered together without forcing contributors into one lab, one architecture, or one theory of intelligence.
The objective is compounding research visibility: make good work easier to find, easier to cite, easier to compare, and easier for the next researcher to build on.
Continuously synchronized
Latest ARCS research.
This section is materialized automatically from public Zenodo community metadata. Zenodo remains the archive of record.
Publication · 2026-09-21
From Neural Capability to Governed Authority: Evidence-Bearing Intelligence in EvoMind
Allit, Gabriel
Software · 2026-09-17
AMEP-1: Adaptive Maritime Estimation and PNT — Research Software v0.3.0
Allit, Gabriel
Report · 2026-09-17
AMEP-1 v0.3.0: Academic Software Release Monograph
Allit, Gabriel