qbrin field notes
Evidence you can inspect
Clear guides to trustworthy AI, honest benchmark results, and recorded experiments from systems that can actually move, switch and decide.
The qbrin editorial series
How the trust layer works — and where it fits.
Six clear guides to qbrin’s architecture and evidence, followed by three grounded opportunity maps for industrial and physical AI.
Inside qbrin6 guides
Product architectureHow qbrin Works: Evidence Before AnswersA visual walk through the path from connected sources to a cited answer—or a deliberate stop when the evidence is not enough.August 13, 2026 · 8 min readSystem designqbrin Trust-Layer Architecture ExplainedA three-plane architecture for grounding what an agent knows, checking what it claims, and governing what it may do.August 13, 2026 · 9 min readAI reliabilityAI Abstention: Why Knowing When to Stop MattersThe business case, interaction design, and measurement discipline behind a system that sometimes says “not enough evidence.”August 13, 2026 · 8 min readAgent governanceEvidence vs Verification vs Authorization in AIA precise way to separate source retrieval, claim support, and permission to act—three checks that are often collapsed into one.August 13, 2026 · 9 min readTechnology comparisonqbrin vs RAG, LLM Observability, and EvalsA fair category comparison: retrieval supplies context, observability supplies traces, evals measure behavior, and qbrin gates live claims and actions.August 13, 2026 · 10 min readBenchmark evidenceqbrin Benchmark Results: An Honest GuideThe strongest numbers, the sample sizes behind them, and the important places where the published report says qbrin does not lead.August 13, 2026 · 11 min read
Industrial & physical AI3 opportunity maps
Space and aerospaceVerification for Space and Aerospace AIAn opportunity map for putting evidence, policy, and HOLD decisions between AI recommendations and mission actions.August 13, 2026 · 10 min readManufacturing and IIoTA Verification Layer for Manufacturing AI and IIoTHow to place evidence and policy checks around maintenance, quality, energy, and supervisory-control workflows without bypassing plant safety.August 13, 2026 · 10 min readPhysical AIA Trust Layer for Physical AI Across IndustriesOne reusable trust pattern—identity, evidence, verification, policy, and trace—adapted to four sectors with very different failure modes.August 13, 2026 · 11 min read
From the qbrin lab
Real systems. Hard questions.
16 measured experiments on what happens when AI agents meet infrastructure, other agents and unreliable evidence.
Field testWe gave an autonomous drone one destination and one rule: never guess.A full 10-minute autonomous drone mission in PX4 and Gazebo, checked step by step by qbrin. The drone completed the survey and returned safely with 0 guesses.August 12, 2026 · 5 min readAI reliabilityWhen an absence becomes a findingA read that failed and a read that found nothing look identical downstream. One of them is a fact about your team.August 28, 2026 · 6 min readTelemetry GroundingA safety gate that blocks 100% of real workAn exact-match telemetry check refused 100% of truthful operator actions.August 16, 2026 · 3 min readSecurity ArchitecturePrompt injection stopped by retrievalOur detector caught 5 of 7 attacks. A retrieval boundary leaked 0 of 8 records.August 16, 2026 · 3 min readLLM VerificationA 400-token cap silently dropped answersJSON truncation crashed the parser, dropping 18.4% of correct answers into refusals.August 16, 2026 · 3 min readCritical infrastructureAn AI agent ran a power gridWithout an evidence check, the grid carried out every command and blacked out seven times.August 11, 2026 · 6 min readIndustrial controlAn AI agent ran a water treatment plantThe plant's own interlocks carried out 23 of 23 attack commands without objecting.August 11, 2026 · 6 min readBenchmark integrityOur benchmark said we were perfect. Then we broke it.An independent audit found fifteen defects—twelve of them in our favour.August 10, 2026 · 7 min readMeasurementHow to measure your AI's hallucination rateDefine the failure, build testable traps, score four outcomes, then audit the scorer.August 10, 2026 · 7 min readAgent identityWho said that—and can they prove it?Identity proves who is speaking. It does not prove that what they say is true.August 10, 2026 · 6 min readAI groundingWhy a citation isn't proofA citation is a pointer, not a proof. Four ways a cited answer can still be wrong.August 9, 2026 · 7 min readMulti-agent systemsHow one bad claim becomes consensusIn a multi-agent system, a hallucination becomes another agent's input.August 9, 2026 · 7 min readAgent verificationWhen one AI agent believes anotherA drone called a sector clear, citing a sweep that never happened.August 8, 2026 · 5 min readAgent assuranceWe let an AI run a chemical plant63,589 decisions over eight hours. No invented reason ever moved a valve.August 6, 2026 · 8 min readAgent containmentWe let an AI run our hardwareIt asked for maximum power 87 times. The evidence allowed it only four.August 4, 2026 · 8 min readGroundedness0 invented answers across 120 trapsThe benchmark, the audit and the evidence behind the number.July 13, 2026 · 3 min read
From the lab
Don’t trust the claim. Inspect the evidence.
Bring one of your agent workflows and see what qbrin allows, refuses, and asks a human to decide.
- Allowed, held or escalated — with the reason
- Every decision traced to its evidence
- Nothing changes in your tools
or email hello@qbrin.com