research-method

Research method

Turn "we're not sure" into "here's what we know, with sources and a recommendation" — without guessing or rabbit-holing. See [[team-memory]] for where findings are recorded.

Process

1. Frame to a decision. What choice does this research serve? What finding would change the recommendation? If nothing would, don't research it — you're procrastinating, not investigating.
2. Check the codebase first. The answer is often already here (an existing pattern, a prior decision, a dependency already in use). Then go outward.
3. Prefer primary sources. Official docs, specs, the actual source, real benchmarks — over blog hearsay. Note the source and its date; tech rots.
4. Compare options honestly. Lay the real alternatives side by side: tradeoffs, cost, maturity, fit to this project's charter/stack. Steelman the option you don't prefer — if you can't argue its best case, you haven't understood it.
5. Time-box it. Deliver the decision-relevant answer, then stop. Depth beyond what moves the decision is waste.

Reporting

• Rate confidence on each finding: High (established / primary source), Medium (reasonable inference), Low (plausible, unverified).
• Separate fact from inference from speculation — never let a guess wear the costume of a fact.
• Cite sources with dates. Mark speculative claims as speculative.
• End with a clear recommendation and the open questions that remain — what still needs a spike before fully committing.