Market Intel
San Francisco housing, measured. A weekly read on the market, by the numbers.
The city, classified by role
Every subdistrict is placed by its own single-family data (price, overbid intensity, and how house-heavy it is), not by its district as a whole. The top of the market splits in two: Luxury areas price to value and barely overbid, while Prestige areas list low and overbid hard.
Where 2026 stands so far
Every neighborhood, at a glance
The sparkline carries the shape of the last ten years in the width of a word; the endpoints carry the numbers. Read the level, not the wiggle; single-year counts are small.
The market, four ways
Houses vs condos, ten-year price paths, the cash fingerprint, and the top of the market. Pick a view.
{{ marketTabs }} {{ marketBody }} {{ nextCTA }}Wait for Fall to list?
Should a seller bring a listing on now, in the summer, or hold for after Labor Day? Ten and a half years of closings, grouped by the month each one hit the market, say the folklore is mostly wrong, and that the launch date matters more than the season.
{{ season }} {{ nextCTA }}
What POTM is
POTM stands for Pulse On Today’s Market. POTM Command is the engine behind Market Intel: it pulls San Francisco’s MLS closed-sale records into one consistent dataset that runs back a decade, refreshed each cycle so the figures stay current rather than months old. Every number traces to a specific closed sale, by neighborhood and housing type.
Governed means it runs on fixed rules, not one-off spreadsheets: one source of truth, single-family kept separate from condos and TICs, thin samples flagged, and nothing typed in by hand.
How it works
Each cycle it does the same four things:
- 1Pulls closed-sale records from the San Francisco MLS.
- 2Cleans and de-duplicates them, so a sale counts once and a closed sale stays closed.
- 3Organizes every sale by SFAR subdistrict, property type, and month, a decade deep.
- 4Computes the medians and shares you see here, sale-to-list, price, days on market, and cash, each carried with its sample size so a thin month reads as directional, not fact.
Where AI fits
Claudio is the team’s agentic analyst, built on Claude AI. Under the team’s direction, Claudio does the heavy lifting: cleaning and reconciling the records, building the maps, charts, and tables, and drafting the first read of what the data shows. The AI makes the work faster and more consistent. It does not decide what is true, and it does not speak for the team.
What Claudio does to build and run Market Intel:
- Builds and maintains the app: the house-market map, every chart, the sparkline tables, and the interactivity.
- Cleans ten years of governed MLS closed sales into one deduplicated dataset, refreshed each cycle.
- Focuses on the single-family home market, read subdistrict by subdistrict. Condos, townhomes, and TICs enter under The market (Homes vs Condos) and the all-residential lanes.
- Classifies each subdistrict into a market role from its own decade of behavior.
- Keeps measured facts and field observations apart, and labels which is which.
Where accountability sits
Level Up Group sets the questions, checks every number against its source, and adds the part no dataset carries: what the team is actually seeing in live deals. The line this site keeps between measured (what the sales records prove) and observed (what agents see in the field) is the team’s rule, and observed is never dressed up as measured.
Final judgment, interpretation, and accountability sit with Level Up Group. The AI is a tool; the standard of honesty and the responsibility for every claim are the team’s. If a number here is wrong, it is the team’s to correct, which is what the Feedback tab is for.
Methodology
The exact rules every number answers to.
Sources and order of authority
Figures derive from the San Francisco Multiple Listing Service, processed through POTM Command, a governed analytics system that applies automated ingestion, repair, data quality review, deduplication, and reporting readiness gates before any figure is published. Reports are only built when the system’s readiness check passes. Active, pending, and sold hero counts defer to InfoSparks (ShowingTime Plus), the San Francisco Association of Realtors’ statistics platform, when sources disagree.
This update’s data
Closed sales through 2026-07-21. After deduplication this update rests on 62,438 unique closed listings (2002 to present; analysis windows begin 2016). Removals this cycle: 7 sale-to-list outliers (data entry errors above 200% of list) and 4 quarantined events excluded by published data quality review decisions.
Definitions
| Closed sale | Status “Closed” or “Sold Off MLS”, deduplicated by listing number keeping the final sold state. |
| Sale vs. list | Close price as a percent of final list price. Medians reported. Values above 200% are excluded as data entry errors. |
| % over asking | Share of closed sales with sale vs. list strictly above 100%. |
| Financed / cash share | Of sales where buyer financing was actually reported. Unreported sales are excluded from the denominator, not assumed either way. This is stricter than feeds that count unreported sales as financed. |
| Days on market | Median days from list to contract as reported to the Multiple Listing Service. |
| Property types | Single Family, Condo / Townhouse (Condominium plus Townhouse), Tenancy in Common, Multi-Unit (Duplex, Triplex, Quadruplex, 5+ Units). |
| Luxury | $5M+ single family and $3M+ condominiums, roughly the top 5% of sales. |
Windows: compared to what?
- Window cascade: last 7, 30, and 90 days against the rolling 12 months, each shown beside the identical window one year earlier. Short windows are momentum clues; the rolling year is the benchmark.
- Trends: matched January to May windows each year since 2016, so seasonality never distorts the comparison.
- Districts: 2026 year to date against each district’s own January to May peak since 2016. Districts are only compared with themselves; years with fewer than seven sales do not set peaks.
- Year over year always means the same period one year prior, never the immediately preceding period.
Sample size discipline
Every figure carries its sample size or a reliability label drawn from POTM Command’s published thresholds: Strong samples support conclusions; Directional samples are momentum clues only; Anecdotal samples are disclosed but never argued from. Neighborhood cells under 15 closings are excluded from rankings entirely.
Presentation doctrine
Charts and tables follow two standards. From Edward Tufte: no chartjunk, no misleading axes, small multiples for fair comparison, and every visual must map to a decision a buyer or seller actually faces. From Patrick Carlisle’s rules for honest real estate statistics: context first (a number without a comparison is worthless), small sample discipline, medians over means, trends over snapshots, and no forecast theater. If a future update ever breaks one of these rules, this page will say so.
Reliability and revisions
All information is deemed reliable but is not guaranteed. Multiple Listing Service data changes after the fact: late-reported closings arrive, statuses are corrected, and quality review reclassifies records. Figures here are subject to change, correction, and revision between updates, and a number may differ slightly from the same number in an earlier or later update for that reason. No claim is made that any figure is 100% accurate, and no responsibility is assumed for errors or omissions. All data, including all measurements and calculations of area, is obtained from various sources and has not been, and will not be, verified by broker or MLS. All information should be independently reviewed and verified for accuracy.
What this is not
Not a forecast, not an appraisal, and not individual financial, legal, tax, or lending advice. Market data describes the past; your decision deserves a conversation about your specific block, property, and situation.
{{ nextCTA }}Found a bug, a bad number, or a gap?
This guide gets better with your eyes on it. Flag anything that looks wrong: a figure that does not match what you see in the field, a subdistrict misclassified, or a metric you wish existed. It routes straight to the Level Up Group team so it can be corrected or built.
{{ feedback }}