Ground on live data with Work IQ

Tool: Cowork Work IQ + exported markdown Source status: Advanced grounding

Live work data is powerful. It is also the easiest place to lose the whole session. The goal is not to build a full Microsoft Graph integration. The goal is to ground your spec in evidence.

Do not spend the hack on Graph auth

A real Graph path means Azure AD app registration, permissions, admin consent questions, MSAL, token cache, and paging. Budget 30-60 minutes. That is most of the hack. Use one of the paths below instead.

Steps

1. Choose a realistic data path.

   Path	Use when	How
   A · Cowork Work IQ	You can use your own tenant data.	Ask Cowork to retrieve and summarize, then export the result.
   B · Snapshot, do not stream	You need repeatability.	Generate one markdown snapshot from ~30 days of calendar. Regenerate monthly.
   C · Persona pack	You need safe demo data.	Use the Marcus Webb persona pack at /Allfiles/persona-pack/.

2. Pull the evidence you need, not everything you can access.

   Review the last 30 days of my calendar. Summarize time allocation by category, recurring vs one-off meetings, self-organized vs invited meetings, accept/decline/tentative ratios, and obvious gaps between stated priorities and calendar behavior. Do not include private meeting details in the final output.

3. Extract the parts self-report cannot reach.

   Signal	Why it matters
   Time allocation by category	Shows what the person actually funds with attention.
   Response latency by sender	Gives an honest stakeholder ranking. People answer what they truly prioritize.
   Accept / decline / tentative ratios	Reveals conflict style and boundary strength.
   Self-organized vs invited ratio	Shows whether the person drives work or absorbs work.
   Recurring meeting load	Shows fixed capacity before new work arrives.

4. Write the result as a small markdown file.

   # revealed.md
   
   ## Window
   - Source: calendar snapshot
   - Range: last 30 days
   - Generated: 2026-07-29
   
   ## Measured behavior
   - 42% of scheduled time is customer-facing.
   - 31% is internal alignment.
   - 18% is focus time.
   - 9% is admin or travel buffer.
   
   ## Calendar vs stated intent
   - Says: protects mornings for strategy.
   - Calendar shows: mornings are the most fragmented block.
   
   ## Stakeholder response pattern
   - Fastest responses go to launch owners and direct manager.
   - Slowest responses go to broad FYI threads.
   
   ## Capacity rule implied by evidence
   - If a same-day executive ask collides with a peer promise, cut scope before slipping the peer promise.

5. Keep the snapshot slow-moving. Working style changes over weeks, not minutes.

   Regenerate revealed.md monthly, or after a role change. Do not stream live calendar data into every agent run unless the task truly needs today's schedule.

6. Strip personal details before demoing.

   • Replace names with roles.
   • Remove meeting titles that reveal confidential work.
   • Show percentages and patterns, not raw calendar entries.

You'll know it worked when... your spec contains at least one measured behavior that contradicts, sharpens, or limits what the person says about themselves.

Apply it to your scenario

Use Work IQ to create the evidence layer. Your agent can only reason from stated preferences unless you add observed behavior. revealed.md is where the scenario becomes grounded instead of aspirational.