Staging Co · Lead Engine

SC2, explained plainly

A calm, no-jargon walkthrough — what this thing is, what's decided, and where you start building the board. Written for Jonah.

First, the reassuring part: I read the notes from all your other open SC2 chats, and every one of them finished its work and is safe to close. Nothing is half-done or secretly broken. The swarm did its jobs and wrote everything down. Your only job now is to understand it — which is what this page is for.

1What SC2 actually is

SC2 is a machine that answers one question for Brandon's team every morning:

"Which houses in San Antonio should we call today to sell them staging?"

That's the whole point. Every bit of jargon, every chat, every report is in service of that one sentence.

The machine has four steps. The last one is the part you actually want to build.

1

Find the houses

Pull every home for sale in San Antonio off Zillow into our own database.

2

Look at the photos

Have an AI look at each home's listing photos and answer: is it empty, sparsely furnished, or already staged? An empty house sitting unsold is a hot staging lead.

3

Score them

Combine "how empty is it" with "how long has it sat unsold" into a single 0–100 score per home.

4

Show the list your UI

Put the scored homes on a board, ranked best-to-worst, so Brandon's team can filter, sort, and start calling.

Step 4 is the UI you want to build. Steps 1–3 are the kitchen that feeds it — and here's the good news: the past month of confusing chatter was almost entirely about steps 1–3, and those questions are now settled. The kitchen works. What's left is mostly the part you want to do: the board.

2The words people kept throwing around

Here's the jargon, decoded. You never need to memorize these — they're here so nothing on this page is a mystery.

A "lead"
A house that's a good sales opportunity — e.g. empty and sitting on the market, so the seller might pay to stage it.
Scraper a.k.a. "API" / "Apify actor"
A little program that reads a website (Zillow) and pulls the house data out into spreadsheet form. Apify is the service that hosts these programs. An actor is just one such program.
The bake-off
A month-long contest testing ~35 scrapers to find which give the most house data for the least money. Winner for bulk data: a cheap one nicknamed igolaizola — about $12 to grab all of San Antonio once.
Vision a.k.a. "the vision sweep"
Using AI to look at each home's photos and decide empty / sparse / staged. The "sweep" was the contest to pick the best, cheapest AI for it. Winner: Claude Haiku, running on the subscription you already pay for — so roughly $0 extra.
"rep-6" a.k.a. sampling gates
A listing has ~23 photos. Reading all of them per house is slow. We proved that reading 6 well-chosen photos (one per room type) gives the same answer as reading all 23 (97% of the time). "Lock rep-6" just means "read 6 smart photos per house, not all 23." That's all.
FSBO "For Sale By Owner"
A house the owner sells without an agent. You suspected we were missing these — we checked hard (five independent sources) and the honest truth is the by-owner market is genuinely tiny (~166–210 homes across the whole metro). We weren't missing coverage.
The funnel
The story of why Zillow shows ~13,000 homes but our board shows only ~4,500. See the next section — it directly answers your suspicion.
Delta / diff engine
Instead of re-scraping all of San Antonio every day, a cheap daily layer that only asks "what changed since yesterday?" — new listings, price cuts, homes that sold. Designed, cost-measured (~$99/mo to keep the whole board fresh), not built yet — you decide when.
Cadence dial
Your knob, per data lane, for how often we refresh (daily / every 3 days / weekly / off). Stored in the database, not the code, so it's adjustable without a developer.
Seats / orchestrator / handoffs
The project was run by many parallel Claude chats. One ("orchestrator") kept the master ledger; the others ("seats") each did one job and reported back. That's why you have a pile of open chats. You don't need to track any of it — that's the orchestrator's job.

3The funnel — why the board looks small

You suspected the board was too small because of a de-duplication bug. We measured it. The dedupe is innocent. Here's where the 13,201 harvested homes actually went:

Plain English: the board is small because of a price filter you never picked, plus a data load that was never re-run — not because of a bug. The two biggest cuts (the price filter and the never-loaded homes) are both things you'd undo.

Your product direction: widest net. Show every home, let Brandon's team filter. The only thing we drop is raw land (you can't stage a vacant lot). Rebuild the board that way and it grows from ~4,500 to ~11,000+.

4What's decided vs. waiting on you

Decided — you can just nod

Waiting on you your call

5So — where do you start on the UI?

The UI's job, in one line: show the widest-net list of San Antonio homes, let Brandon's team filter (price, type, empty/staged, days-on-market) and sort, and mark who they've contacted.

Step 0 — look at the seed prototype together →
leads.leverand.dev (sample data, not real yet). Ask yourself: "what do I want this to feel like for Brandon?"

The very first buildable step is usually the same: get the real, widest-net data into a simple filterable table on screen. Everything else — score tweaks, the HubSpot push, saved views — layers on top of that one table.

When you're ready, we spin up a clean "SC2-UI build" mission so the UI work has its own home, separate from all the bake-off chatter.