Where We Stand
The architecture is decided. These calls are locked and logged in the decision register; anything below can be revisited, but only with a written rationale that supersedes the entry.
- DECIDEDForecast-first, census-opportunistic. Capital forecasting is built from age distributions and install dates. Census data enriches over time — it is never a dependency.
- DECIDEDNo full census. Twenty-five years of work-order history across 36,000 homes is the actuarial table. Inventory only lumpy building-level assets — roofs, HVAC, boilers — at the project level, where replacement waves break reserves. Appliances are carried as a rate.
- DECIDEDPanel-QR anchor. One QR per unit on the electrical panel (~36k tags), opening the unit's full asset list. Not one tag per appliance (~180k).
- DECIDEDYardi: one-time export. Geography changes rarely; a live licensed interface is unjustified. The ask is the property code — the join key.
- DECIDEDSit above Tailorbird. Their inference engine is a data-acquisition feed at the inferred confidence tier. This platform is the decision layer that consumes it.
- DECIDEDRoofs first. Highest value, longest horizon, aerially assessable — no field-adoption machinery required to start.
- DECIDEDLease-term is the lens. Under MHPI ground leases, remaining useful life only matters relative to remaining lease term. Per-project SPE waterfalls mean cash cannot move freely between projects — allocation logic must respect that.
- DECIDEDKnowledge routing. Asset data is canonical in Supabase; decision reasoning is canonical in the vault; this site displays both and stores neither.
Next Steps
Seven open threads, roughly in dependency order. The first two are conversations, not code — discovery answers shape everything downstream.
Log the Tailorbird Answers
The first vendor meeting was July 24. Capture their answers to the eleven follow-up questions in the workbench — above all: can their output be exported via API as a structured feed? That single answer determines whether they slot in as a data source or stay a demo.
Run the Client Questionnaire
Twenty-two current-state questions across six sections, already in client-safe Word format. Sections B and C — existing CNA studies and per-project lease/reserve structure — feed the forecasting model directly. North star: what did the investment bankers ask for that was hard to produce?
Santo Unit Roster
Get the address- and unit-level roster for Santo Terrace from the data contact. This replaces synthetic units with real ones for the pilot and validates the four-level geography model against reality.
SanGIS Footprint Pull
OSM coverage of Santo is partial (88 buildings captured). Pull the authoritative SanGIS building footprints and parcels for complete coverage — Santo sits on city land, so the public record exists.
Boundary + Year-Built Layer
Draw the Santo community boundary and join county assessor year-built data to each footprint. Year-built is the seed for every age-based forecast where install dates are missing.
Stand Up Supabase, Wire It Live
Create the Supabase project, load the schema (including the decisions and open-questions registers), import Santo, and point the workbench at live reads instead of embedded static data.
Lease-Term vs. Asset-Life Chart
The signature analytic: remaining useful life plotted against remaining ground-lease term, per project. This is the chart an investment banker has not seen before — and the one that frames every extension decision.
What Exists Today
Everything below is built and versioned in the private repository.
Santo Pilot Workbench
Real footprint map, honest confidence-tiered asset view, and seven decision accordions with decided/open status.
Footprint Map
88 Santo buildings over Esri imagery; 75 residential-class, measured roof areas per building.
Synthetic Portfolio
203 communities, 6,409 buildings, 36,396 homes across 24 real installations — seeded and regenerable, every row marked inferred.
Database Schema
Generic asset model with four-level geography, confidence tiers, capex-vs-lease view, plus seeded decision and open-question registers.
Discovery Kit
Client questionnaire (Word, 22 questions), internal discovery brief, and the tiered data request.
Roadmap + Decision Log
Five-phase roadmap and four written decision notes, mirrored to the knowledge vault.