Diligence Question Builder

How the assembly rules work — and why they are published

The Diligence Question Builder answers one narrow question: given a sector, an Atlas exposure, and an entity type, which of the questions already published on this site should a serious reader put in front of a founder, a manager, or a filing first? The answer is produced by deterministic rules over a tagged question bank, not by judgment calls made out of view. The bank holds every question from the thirteen Learn diligence checklists plus the diligence questions from the sector research guides — one hundred and fifty-five questions in total — and each one carries tags naming the selections it applies to. A base layer applies to every research target: scoping questions, team questions, and the regulatory and export-control prompts that matter regardless of what is being examined. On top of that base, the entity-type layer swaps in the questions that match the chosen wrapper, the sector layer adds the questions from the matching sector checklist and sector guide, and the exposure layer adds the questions tagged to the selected Atlas capital priority. When the sector or exposure sits in the published dual-use implication set — defense and robotics sectors, defense-industrial and semiconductor exposures — the dual-use questions join the brief as well. Duplicates are removed by id, and the result is ordered by diligence dimension, then by the bank's own order.

The entity-type layer is where the builder earns its keep, because the most common diligence error on this site's subject matter is asking startup questions of things that are not startups. Select a private startup and the brief includes cap-table and follow-on financing questions: liquidation preferences, stacked notes, option pools, and what happens in a bridge or down round. Switch the same selection to a fund and those questions disappear — a fund has no cap table to diligence in that sense — and the manager questions arrive instead: sourcing edge, reserve strategy, realized DPI versus paper TVPI, fee and carry terms, and how the syndicate or SPV handles conflicts, follow-on, and failure. Switch to a public company and the brief pivots again, toward filings, segment linkage, liquidity, and valuation assumptions, because startup-style product claims are the wrong lens for a listed business. Each entity type also carries an explicit evidence-standard note in the brief header, using the same phrasing as the Company Comparison Matrix, so the standard travels with the questions.

It is equally important to say what this tool does not do. It is not legal advice: the regulatory and export-control group identifies issues — CFIUS, ITAR, EAR, sanctions, licensing, data movement — that belong in front of qualified counsel, and nothing the builder emits substitutes for that review. It is not exhaustive: a real diligence process adds deal-specific questions that no generic bank can anticipate, and a brief of tailored prompts is a starting agenda, not a completed program. And it is not a scoring tool: the builder never grades answers, ranks companies, weighs one question against another, or converts a completed checklist into a recommendation. A selection that returns more questions is not riskier or more attractive than one that returns fewer; the count reflects how much of the published bank is tagged to that slice of the research space, nothing more.

The relationship to the Learn checklists is deliberately one-directional. The checklists are the canonical record: each card pairs its questions with the evidence to request, the positive signs, the warning signs, and a caution about what the checklist cannot prove. The builder reads from that same source — the texts are verbatim, and every question in a brief footers the checklist it came from with a link to the canonical anchor — but it never writes back. If the two ever disagree, the checklist page wins, and the build fails rather than shipping a divergence. That design keeps the app honest: it can rearrange the site's published diligence thinking around a reader's specific selection, but it cannot quietly become a second, unmaintained copy of it.

Read the output the way a careful analyst would read any prepared agenda. The questions are prompts for evidence, not conclusions; several of them exist precisely to surface what is still unverified. The exposure layer connects the brief to the Dependency Atlas, whose priorities are re-scored as source data changes, so the brief should be read against the as-of date shown in the header. And when a question produces an interesting answer, the next step is never inside this tool — it is the company profile, the sector page, the Atlas investment case, the primary source, and where appropriate, professional advice.

What this app does not do

• It does not score, rank, or grade companies, funds, or answers — it only assembles questions.
• It does not produce an exhaustive diligence program; a real process adds deal-specific questions no generic bank can anticipate.
• It does not provide legal, tax, accounting, or regulatory advice — the regulatory and export-control questions identify issues to raise with qualified counsel.
• It does not replace the Learn diligence checklists, which remain the canonical record with evidence lists, positive signs, warning signs, and cautions.

Those boundaries are the point of the tool. A question set is useful because it is honest about being a question set: it tells the reader what to ask and where the canonical context lives, and it refuses to pretend that asking is the same as knowing. Questions past that line require direct diligence and, where appropriate, legal, tax, regulatory, or technical advice.