RESEARCH INTELLIGENCE
Research-backed analysis of AI agents, reliability, memory, benchmarks and safety. Find original SXF guides, examine evaluation evidence and follow what matters next.
SXF RESEARCH LIBRARY
Six featured SXF research analyses spanning AI search accuracy, agent reliability, memory, coding architecture, procurement negotiation and medical diagnosis.
NEW SXF RESEARCH AI SEARCH / SOURCE AUDIT
Explore 28 user-submitted answers, a $270 vs $45 pricing-tier error, official citations, original prompts and a transparent research appendix.
Read the original 14-question study
NEW RESEARCH MEDICAL AI / SCIENCE
How MAI-DxO achieved 83.2% on a controlled diagnosis benchmark — and what the comparison does not prove.
Read the research analysis
FEATURED STUDY RELIABILITY / MEASUREMENT
Benchmarks do not prove consistent performance. Explore 12 reliability metrics, repeated-run success, safe failures and recovery.
Read the reliability research
MEMORY / AGENT DESIGN
How agents write, retrieve, revise and forget memory beyond a context window.
Explore memory architecture
ARCHITECTURE / CODING AGENTS
Compare tools, plugins, skills, functions, MCP and delegated subagents.
Understand agent architecture
PROCUREMENT / PAYMENT TERMS
Compare Net 30, Net 60 and Net 90 with published negotiation evidence, supplier tradeoffs and an original decision framework.
Explore payment negotiation researchNEW SXF MARKETING AGENT LAB · NOVEMBER 2026
Twelve repeated anomaly tests plus four recommendation tests across four configured environments. Explore the prompts, synthetic CSV, answer key, exact flag scores, operator-transcribed outputs and limitations. Read the full buyer's guide.
SXF ORIGINAL FIELD TEST · NOVEMBER 2026
Read the actual CSV, Excel and web research outcomes from SXF's controlled exploratory AI agent evaluation, with its methods, available evidence files and limitations. One run per scenario does not prove long-term reliability.
These are SXF research guides and editorial analyses, not claims that SXF ran independent laboratory experiments. Sources and limitations are documented within each article.
LIVE / PRIMARY-SOURCE SIGNALS
Fresh findings and announcements with links to original sources.
Showing current research updates.
Research signals update independently of SXF analysis. Verify methods and claims in the linked primary sources. View all AI signals
FROM CLAIMS TO COMPARABLE RESULTS
Compare results only when benchmark versions, evaluators and model configurations align. See the evidence and its limits—without mixing vendor claims into a single ranking.
START WITH A QUESTION
Explore an area, then follow its supporting evidence.
Memory, reliability and the architecture behind tool-using, long-running agents.
Browse original analysis 02 / MEASUREMENTUnderstand measured capabilities, evaluator versions, configurations and limitations.
Explore the evidence 03 / TRUSTAuthorization, prompt injection, failure boundaries and practical agent controls.
Study safety topics 04 / DISCOVERYFollow research applications while separating demonstrations from validated results.
Follow scientific workFROM EVIDENCE TO PRACTICE
Related reading for safety and scientific applications.
SAFETY / SECURITY
Reliable output is not enough. Agent authorization, least-privilege access and safeguards against untrusted instructions determine how safely an agent can operate.
SCIENCE / APPLICATIONS
Scientific advances range from experimental methods to deployed systems. Read the original research question, validation data and constraints before extrapolating a demonstration to production.
OUR RESEARCH LENS
Our four-step approach to trustworthy AI evidence.
Identify the problem, environment and hypothesis before comparing headlines or model names.
Look for source documentation, data, benchmark version, test configuration and whether results were independently reproduced.
Separate measured performance from marketing claims, extrapolations and unknown failure conditions.
Connect findings to an evaluation, implementation guide or live source so readers can verify and apply them.
Read more about the SXF editorial method and sourcing, or inspect benchmark methodology and model evidence.
COMMON QUESTIONS
Quick answers to common research questions.
Research into AI agents, memory, reliability, system architecture, model evaluation, security and scientific applications. SXF publishes source-backed analyses and links to original work by research organizations.
A research paper investigates a question; a benchmark tests performance under specified conditions. A benchmark score alone cannot establish general reliability, safety or performance outside its evaluation setting. Use the evaluation library for version-aware results.
Each research guide includes supporting sources, while individual news signals link to the primary announcement. Review methodology, evaluator configuration and limitations before drawing conclusions.