About Swarm-It Discovery
Automated research paper discovery powered by RSCT certification and κ-gate quality scoring.
What We Do
Swarm-It Discovery automatically scans arXiv and other sources for new research papers, matches them against curated topics, and certifies their quality using RSCT (Representation-Solver Compatibility Theory).
Unlike traditional paper aggregators that rely on keywords or citations, we use semantic similarity and multi-agent certification to find papers that are genuinely relevant to cutting-edge AI research topics.
RSCT Certification
Three Independent Certification Axes
RSCT certifies representations on three provably independent axes. A system can be perfectly stable and completely wrong, or highly compatible but information-blocked. Single quality scores collapse these distinct failure modes—RSCT distinguishes them.
RSN Decomposition
Every paper is decomposed into three components that sum to 1.0 (simplex constraint):
Relevance (R)
The portion of the paper that directly addresses research goals and topics. High R means the paper is on-target.
Superfluous (S)
Content that occupies context but neither helps nor corrupts inference. Unlike noise, superfluous content is tolerable—it wastes space but doesn't mislead.
Noise (N)
Adversarial content that actively corrupts inference—not merely irrelevant, but misleading. When N ≥ 0.5, no solver can recover correct inference (Fano bound). Low N is critical.
Simplex Constraint: R + S + N = 1.0 (always)
Signal Purity (α)
The critical measure is purity: α = R/(R+N). This ratio determines the hard ceiling on solver performance—no amount of model capacity can overcome low purity (Fano's inequality).
When α → 0, correct inference becomes information-theoretically impossible. This is why Gate 1 blocks inputs with N ≥ 0.5.
κ-gate Quality Scoring
The kappa (κ) score represents overall quality and compatibility. This is a single number (0-1) that summarizes how well the paper passes our quality gates.
Quality Threshold: Only papers with κ ≥ 0.7 are published. This ensures all reviews meet a minimum quality standard.
Turbulence (σ)
The third certification axis measures representational stability. Even a compatible encoding can fail if it produces unstable dynamics.
Stable (σ ≤ 0.5)
Solver trajectories converge reliably. Safe to execute.
Turbulent (σ > 0.5)
Risk of trajectory divergence. Higher κ required via the Oobleck principle: κ_req(σ) = κ_base + λσ
The 4-Gate Gatekeeper
RSCT enforces quality through a sequential 4-gate system. Gate ordering is a security guarantee, not an efficiency choice—reordering creates vulnerabilities.
Multi-Agent Certification
RSCT certification uses a swarm of specialized agents to evaluate each paper from multiple perspectives:
Fetches and validates input
Matches against topics
Generates reviews
Each agent certifies its output before passing to the next stage. This creates a certified pipeline where quality is validated at every step.
The Pipeline
🔍Daily Scan
Automatically fetch new papers from arXiv, bioRxiv, and other sources. The scanner agent validates inputs and checks for duplicates.
🎯Semantic Matching
Convert paper abstracts to embeddings and compute cosine similarity against curated research topics. Papers above the threshold (typically 0.5+) proceed to certification.
✅RSCT Certification
Multi-agent swarm analyzes each paper, computing R/S/N decomposition and κ-gate scores. Only papers with κ ≥ 0.7 are published.
📄Review Generation
LLM-powered analysis generates comprehensive reviews with key insights, technical contributions, and relevance to AI safety and multi-agent systems.
🚀Publication
Reviews are published to this site with full RSCT metrics, tags, and difficulty ratings. Users can search, filter, and explore by topic.
Technology Stack
Frontend
- • Gatsby 5 - Static site generation
- • React 18 - UI components
- • TypeScript - Type safety
- • Tailwind CSS - Styling with dark mode
- • GraphQL - Data layer
Backend Pipeline
- • Python 3.11+ - Pipeline orchestration
- • OpenAI API - Embeddings & analysis
- • Swarm-It Sidecar - RSCT certification
- • arXiv API - Paper discovery
- • AWS Lambda - Scheduled execution
Patent Notice: The RSCT methodology and κ-gate scoring system are patent-pending technology developed by Next Shift Consulting. See patent documentation for details.
Start Exploring
Browse RSCT-certified papers or explore research topics.