references/output-depth-policy.md

# Output Depth Policy

For a new substantive deliverable whose analysis depth is not already stated, follow `references/deliverable-intake-policy.md` before choosing a level of work. Its user-selected depth controls breadth and packaging, not readiness or circulation status.

Default to `extended_analysis` for Investment Banking skills. A normal answer should preserve the senior-banker reasoning chain, source/evidence posture, key tables, caveats, open items, and recommended next actions that make the output decision-useful.

Full-depth does not mean Markdown-by-default. For substantial artifacts, prefer the proper hero deliverable: XLSX workbook, polished standalone HTML report, HTML dashboard, native deck/document, or a justified chat-only answer. A narrative skill migrated to `references/html-artifact-standard.md` owns its standalone HTML structure directly; existing renderer-based paths may remain only until each producing skill is migrated. Markdown report files, naked JSON contracts, and raw CSV packages should not be the default reader-facing deliverables.

Use a shorter format only when one of these conditions is explicit or unavoidable:

1. The user asks for a short answer, quick take, one-pager, red-flags-only view, top-issues-only view, brief, or just the answer.
2. The task is live or time-boxed meeting prep where speed is more valuable than completeness.
3. Source context is too thin for a full analysis without creating false precision; provide a starter framework, assumptions, and exact source asks instead.
4. The main human deliverable is already a richer artifact such as a workbook, HTML dashboard, HTML report, native deck, tracker, or generated package; the chat response may be a concise cover note.
5. The user is asking for deltas, changes, or follow-up updates against an existing full artifact.
6. The task is pure triage, such as whether something is worth reviewing, what breaks first, or what the obvious issues are.
7. The user asks a narrow source/evidence question, such as which source controls one metric.

When using a shorter format, state the reason or make it clear from the user's request. Do not drop material evidence labels, caveats, source conflicts, readiness posture, or next actions merely to save space. If in doubt, choose `extended_analysis` and package it in the right hero format.

Read `references/deliverable-format-policy.md` before creating files. Final responses should call out the hero deliverable first, then explain any supporting CSV/JSON/log artifacts in plain English.

For workbook-heavy work, also read `references/workbook-first-tab-standard.md` so the first visible worksheet is a banker-readable insight dashboard rather than an instruction-only sheet.