# Cap Deep-Research Depth at One or Two Levels ## Claim Limit each AI deep-research branch to one or two levels of narrowing depth so exploration converges on something shippable instead of an open-ended rabbit hole. <!-- atomicity: 1 idea ✓ --> ## Scope & conditions - Holds when the goal is delivery on a bounded timeline — a real product problem that needs a shipped answer. - Relax the cap when the goal is genuine open-ended discovery with no deadline pressure; unlimited depth is appropriate there. - Applies specifically to the narrowing pattern (broad report → pick a segment → re-run breadth-first on that segment), not to the initial breadth-first pass itself, which should stay wide. ## Evidence Ross Morsali's deep-research workflow for WordPress faceted search started broad (survey scalable-systems techniques across Amazon, Netflix, Elasticsearch, Algolia), then narrowed one level (bitmap storage specifically), then narrowed again (10 ways to implement bitmaps) before selecting a testable candidate. Without this discipline, the same research instinct carried him into sonar/radar pattern-matching, Rust adoption, and eventually applying for a GPU grant to train his own LLM — each a legitimate but scope-expanding tangent from the original problem. ## Links - [[3. Resources/Permanent Notes/Benchmark-Gated Adoption Lets You Ship Techniques You Don't Fully Understand]] — because applies-in; the benchmark gate is what makes a depth-capped research branch actually shippable rather than just stopped. - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — because applies-in; a scope-discipline pattern specific to AI-assisted research workflows. ## Tests / falsifiers - Compare time-to-ship on research efforts that respected the one-or-two-level cap versus ones that didn't, holding problem complexity roughly constant — if uncapped research doesn't measurably slow delivery, the claim is weaker than the source's framing suggests. - Check whether a hard depth cap ever caused a maker to miss a genuinely better solution one level further down — a real cost the source doesn't examine. ## Maintenance - Last tended: 2026-08-16 - Next review: 2026-11-16 ## Source - [[3. Resources/Readwise/Articles/How to Use AI Deep Research to Improve WordPress Plugins]] — Steve Burge interviewing Ross Morsali, The PublishPress Podcast, on scoping deep-research exploration to stay shippable.