Sharpen dashboard vs deep-dive views: 1 compass + N deep-dives, no overlap

[ranjiao]

Background

User flagged that dashboard and board views overlap. Looking at the spec carefully, the issue generalises across the whole catalog: dashboard was loosely defined as 'whole-project overview: phase summary + open P0/P1 + USER input queue + recent decisions + ...', with nothing forbidding it from embedding full content from each section.

This made the boundary fuzzy:

• Dashboard's 'open P0/P1' section → could be inline table → duplicates board view
• Dashboard's 'recent decisions' → could embed ADR bodies → duplicates decisions view
• Dashboard's 'audit drift' → could embed audit findings → duplicates architecture view

Fix

One design principle, applied uniformly:

• dashboard = compass. One summary-style view. Each section = 1-line summary + top-1-or-3 items (titles only) + jump-link to corresponding view. NEVER embeds full content.
• All other views = deep-dive on one domain. Full content for that domain.
• Reading times are different on purpose: dashboard ≈ 30 sec orientation; deep-dive ≈ minutes of work.

What's changed in pmo/reference/rendering.md

1. New 'Design principle' section before the catalog with the rule + a per-section spec table (what dashboard shows vs does NOT show, per BOARD / Phase / Decisions / Architecture / Runbooks / Incidents / User Input Q).

2. Catalog rewrite:

   • dashboard row: explicitly 'Compass page'; source-files annotated 'read headers/counts only — do NOT load full content'; mirrors PMO chat-standup dashboard's row set.
   • board, phase, architecture, decisions: each marked 'the deep-dive that dashboard's X section links to' for symmetric documentation.
   • handoff: explicitly noted as an exception (self-contained briefing IS the point).
   • Extension rule added: 'When adding a new view, confirm it is a deep-dive on one domain. Perry has exactly one compass.'

What's NOT changed

• No new views added
• No views removed (board is still its own view — it's the right deep-dive)
• SKILL.md untouched (rendering.md is loaded only on /pmo render, not in every standup)
• bin/perry-render-index no code changes (catalog enumeration is fine as-is; design rule is for the agent that generates HTML, not for the script that indexes)

Test plan

• Future /pmo render dashboard run produces an HTML that shows counts + top items + 'see board view' style links — not full tables embedded.
• Future /pmo render board still produces the full interactive task table.
• If both are rendered same day, opening dashboard and then clicking through to board should feel like 'overview → detail', not 'two pages with the same content.'

🤖 Generated with Claude Code