Developer builds context integrity layer to catch stale, unverifiable AI agent inputs

A developer has identified a largely overlooked reliability gap in AI agent pipelines: context passed to models is rarely checked for freshness, attribution, or internal coherence before the agent acts on it. Unlike retrieval scoring, which only measures similarity to a query, these three properties determine whether context is actually trustworthy and safe to act upon. To address this, the developer built an evaluation engine that scores context across all three dimensions using a decay-adjusted model, with results written to a signed ledger verifiable offline via Ed25519 keys. The tool operates over the Model Context Protocol and is available as a hosted endpoint with a live demo at freshcontext.dev, with its specification published under an MIT licence. Features such as a safety harness and telemetry dashboard are noted as still in development, and the developer is seeking feedback from those building agents, retrieval, or memory systems.