Research industry patterns

Find what an industry actually does, then benchmark our situation against it. The deliverable is not a survey of the field - it is a placement of us within the field, and a verdict on every place we differ.

Pin the scope

Two things must be concrete before researching, or the report is unanchored:

• The context to research - a market plus a domain. "The software industry" is unresearchable; "trunk-based development for a small team" is.
• Our situation - what we currently do, or the decision we face, in real terms from this conversation.

If either side is fuzzy, pull it from the conversation first; ask once only if it cannot be recovered.

Research, don't recall - but stay tight

Go find current practice with real searches - pretraining recall is stale and carries no source. A genuine standard shows up when independent sources converge; one vendor's marketing or a single loud blog is a data point, not a norm. Date each finding, because a 2019 "best practice" may be today's anti-pattern.

Keep the pass tight by default: a handful of targeted searches, and reason from result snippets. Fetch a full page only to verify a specific load-bearing claim - full-page fetches are the main token cost, so spend them deliberately. Go deeper only when the user asks for depth or the decision is high-stakes enough to warrant it.

Benchmark against our situation

Structure the report around us, not the industry. For each pattern found, place us against it:

• Aligned - we already do this.
• Diverged - we do something different.
• Absent - the industry does it and we don't do it at all.

Convention or constraint

For every divergence, deliver the verdict this skill exists for: is the pattern a constraint or a convention?

• Constraint - it exists for a real reason, and violating it causes a specific failure. This is the headline; name the failure.
• Convention - everyone does it, but nothing breaks if we don't. Report it, then say it is safe to skip.

Do not cargo-cult: an industry standard we can consciously dismiss is a finding, not a to-do. A market norm may itself be a local convention - what Taiwan ecommerce assumes need not bind a global product - so judge the pattern's reason, never its popularity.

Done

Every researched pattern is placed against our situation, and every divergence carries a convention-or-constraint verdict with a dated source. Nothing is reported that we can neither act on nor consciously set aside.