Survey & Building Library

Template rules:

• Clone template slides, never rebuild from scratch
• Only swap text content inside existing XML elements
• Never touch formatting (fonts, colors, borders, cell sizes)
• Replace image relationships to swap photos/floorplans
• CBRE branding must remain intact

PPTX generation: Use python-pptx locally. Template at skills/survey/references/template.pptx. Images from Supabase Storage (direct access — no proxy needed on Mac Mini).

2. Query Building Data

Trigger: "what do we have on [address]" or "pull buildings in [submarket] with [criteria]"

Query Notion/Supabase by:

• Address (fuzzy ILIKE match)
• Submarket
• Building class
• RSF range / floorplate size
• Amenities (from notes field)
• Any combination

Return: building details + all available spaces + image status.

3. Add Buildings

Trigger: JD provides building info or drops a PPT archive

Write to BOTH Notion and Supabase in same operation:

1. Create Notion page in Building Library with all fields
2. Insert row in Supabase buildings table
3. Upload building photo to Supabase Storage
4. Report what was added

4. Process Landlord Flyers

Trigger: JD drops a PDF flyer

1. Extract images via PyMuPDF: floorplan (largest), interior photos, exterior if present
2. Extract text: floor, SF, asking rent, timing, lease type, comments
3. Upload images to Supabase Storage ([building-slug]/[floor-slug]/)
4. Update BOTH Notion Spaces entry AND Supabase spaces table
5. Flyer images ALWAYS override CoStar images (better quality)

5. Process CoStar Exports

Trigger: JD drops a CoStar PDF

1. Parse all buildings and availability
2. Cross-reference against existing data — flag new vs existing
3. For new buildings: create in both Notion + Supabase
4. Update availability data (SF, rent, timing)
5. Only store CoStar images if no flyer images exist
6. Cross-reference pipeline — flag if any CoStar building is in a prospect's target submarket

6. Cross-Reference (runs after every building/space update)

After ANY building or space data change:

• Check pipeline for prospects in same submarket
• Check pipeline for prospects looking at same building
• If match found: surface to JD with draft outreach angle
• This is the market-intel → outreach flywheel applied to the building library

Data Rules

• Notion = editing interface. Norman syncs to Supabase on every write.
• Flyer images > CoStar images. Never overwrite better quality.
• Address normalization: "440 Park Ave S" = "440 Park Avenue South". Resolve before insert.
• Flag spaces not updated in 60+ days as potentially stale.
• Images persist in Supabase Storage permanently.

Quality Checks (before delivering any survey)

1. All addresses correct and consistently formatted
2. Building SF, stories, year built match database
3. No template placeholder text remaining (search for < and > markers)
4. All images load
5. CBRE branding intact on every slide
6. Asking rent includes annum calculation where SF and rent both available

Architecture Reference

See references/architecture-v1.2.md for the full original architecture including Make.com scenarios, edge functions, and Vercel proxy details (legacy — mostly replaced by direct access).