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.

α
Signal Purity
Content level
κ
Compatibility
Solver level
σ
Turbulence
Dynamics level

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.

🥇
Exceptional
Gold tier certification
κ ≥ 0.9
🥈
High Quality
Silver tier certification
κ ≥ 0.8
🥉
Certified
Bronze tier certification
κ ≥ 0.7

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.

🛑
Gate 1: Integrity Guard
If N ≥ 0.5 → REJECT (noise saturation)
⏸️
Gate 2: Consensus Gate
If consensus coherence c < 0.4 → BLOCK
🔄
Gate 3: Admissibility (Oobleck)
If κ < κ_req(σ) → RE_ENCODE
🔧
Gate 4: Grounding Repair
If κ_L < 0.3 → REPAIR (low-level health)

Multi-Agent Certification

RSCT certification uses a swarm of specialized agents to evaluate each paper from multiple perspectives:

🔍
Scanner Agent

Fetches and validates input

🧠
Analyzer Agent

Matches against topics

📝
Publisher Agent

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

1

🔍Daily Scan

Automatically fetch new papers from arXiv, bioRxiv, and other sources. The scanner agent validates inputs and checks for duplicates.

2

🎯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.

3

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.

4

📄Review Generation

LLM-powered analysis generates comprehensive reviews with key insights, technical contributions, and relevance to AI safety and multi-agent systems.

5

🚀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.