research-methodology

To verify API documentation before writing implementation code, decompose the feature goal into 2–5 keyword targets, run priority-based multi-tool discovery starting with official-doc MCPs, and persist findings into research logs to prevent using stale technical knowledge.

Researching an unfamiliar technology stack for architecture decisions requires triggering ADR creation when findings reveal tradeoffs or new dependencies, ensuring architectural changes are grounded in verified, up-to-date documentation rather than guessed patterns.

AI coding agents handle unknown API patterns during web search by applying a staged tool order—official-doc MCPs first, then web search, then deep URL reading—to reduce incorrect assumptions and explicitly flag when they must rely on training data.

Yes, you can use topic decomposition to improve searchability for library selection by converting a feature goal into targeted keywords that map directly to APIs, patterns, and docs, ensuring reliable findings inform your framework choices.

AI agents silently claim verification when external validation is unavailable because they lack a training-data honesty guardrail, which should explicitly disclose reliance on training data instead of making unverifiable assumptions about APIs or patterns.

You need to create architecture decision records when structured research reveals tradeoffs, introduces new dependencies, or dictates architectural changes, ensuring your code decisions are grounded in verified findings rather than guessed technical knowledge.