How to Read Comps Like an SF Agent (With Our Comps Tool)

A comp is a recent closed sale similar enough to the home you want that its price tells you something about yours. In San Francisco that is harder than it sounds, because two houses on the same block can differ by a view, a legal unit and a brick foundation. Here is the method agents use, the adjustments that matter here, and a walkthrough of the comps tool, which shows every comp behind its estimate.

The three tests: similar, nearby, recent

Every candidate comp has to pass three tests, and the art is in how much you relax each one when the data is thin.

• Similar. Same property type first (condo, single-family, and ideally not a TIC when you are buying a condo), then size within about 20%, same bedroom count or one off, similar era and condition.
• Nearby. Same building or block is best. In SF, "nearby" means the same micro-neighborhood and the same side of any hill, highway or commercial strip, not a radius on a map.
• Recent. The last six months carry the most weight. Anything older needs a time adjustment, and anything over 18 months is context, not evidence.

When you cannot get three sales that pass all three, loosen one test at a time and write down what you gave up.

Why SF comps are hard

Most of the country values homes in subdivisions of similar houses. SF values them in a city of one-offs:

• Micro-neighborhoods. Price per square foot can change by a third across one boulevard or one ridge line.
• Views. A Bay, bridge or downtown view carries a premium no dataset records, and it can vanish when the lot next door builds up.
• Parking. Deeded garage parking versus leased versus none moves condo values materially.
• Lot slope and foundation. Two Edwardians with the same footprint can carry very different structural risk.
• Unwarranted space. A "bonus room" without permits may be counted in one listing's square footage and not another's. The 3R report gives the legal unit count and permit history; the disclosure package explains how to read it.

So in SF you weigh comps by how well you know them, not just by how close they are. A sale your agent walked through is worth two you only saw online.

Adjustments that matter

There is no published SF adjustment schedule; agents and appraisers use paired sales and judgment, so think in percentages rather than fixed dollars.

Make the big adjustments first (tenant, unwarranted unit, parking, view), then the small ones. If a comp needs more than three large adjustments, it is not a comp.

Time adjustment, pendings and off-market sales

In a market where the median house price rose to a record $2.15 million in March 2026, up roughly 18% to 20% in a year, a comp from last spring is stale without a time adjustment. The tool handles this with a monthly citywide price index per property type: each comp's price is multiplied by today's index over the index in its sale month, smoothed over three months.

Two sources of evidence never reach public records on time: pending sales, whose contract prices your agent can often learn from the listing agent, and off-market closings, which surface only when they record. Ask your agent to add both to any comp set and to say which of the tool's comps they would untick.

Walkthrough: the comps tool

The tool's estimate is a predicted sale price built from closed sales, with every input visible. For any SF condo or single-family address:

1. Search. Enter the address and open the property page. You see an estimated sale price today, a range, a confidence label and the comparable sales table.
2. Read the comp set. Candidates are sales of the same type within 0.35 miles and 18 months, widened in steps to 3 miles and 36 months only when there are too few. Each gets a similarity score that multiplies distance (with a bonus for a nearby home on the same street), recency (half-life about seven months), size ratio, bedroom match, year built and how sure we are of the price. Every sale inside the search circle is ticked by default and weighted by that score, and tags such as "similar size" and "recent" tell you why.
3. Toggle. Untick comps that fail your tests, or drag the search circle to change which sales count. The estimate recalculates in your browser. When at least three ticked comps have square footage, the estimate is the similarity-weighted median adjusted $/sqft times the subject's square footage. For a home for sale, it also uses the asking price and how far over asking the comps sold. The range is sized from testing so that about 8 in 10 sales land inside it.
4. Read the confidence. High means six or more ticked comps in a tight range, medium means three or more, low means you are extrapolating. Treat a low-confidence estimate as a prompt for more data, not as a value.
5. Check the history and the lost offers. The page shows the home's own sale history and, for signed-in members, the losing offers other buyers reported on it, without names.

Two honest limits. Closed sales take a few weeks to show up, and older sale prices are inferred from public records, so the time index does real work. And the tool knows nothing about condition, views, renovations or HOA health. The methodology page spells all of this out.

When to override the estimate

Override when you know something the data does not:

• Unique homes. A view house on a double lot, a live-work loft, a penthouse. Use the tool's $/sqft as a floor and build up from the one or two genuinely similar sales.
• Thin data. Fewer than three solid comps, or a low confidence label. Widen the time window before the geography, and lean on your agent's pendings.
• Fast-moving segments. When single-family homes are averaging 20% or more over asking and the index lags, the newest pendings matter more than any closed comp.
• Known defects. A large Section 1 pest figure, a soft-story building, an unwarranted unit. Price the fix explicitly.

Comps vs appraisal

Your lender's appraiser will look at the same closed sales with narrower rules: typically at least three closed comps, a preference for the most recent and nearest similar sales, and documented adjustments on a standard form. Pendings and your agent's knowledge count for less, which is why a value that leans on momentum can beat the appraisal; that is the mechanism behind appraisal gaps. Many jumbo lenders require a second appraisal above a loan-size threshold that varies by lender, often on exactly the houses that go furthest over asking; ask yours where it sits. If your value depends on a comp the appraiser will not accept, know that before you waive the appraisal contingency.

From comps to an offer range and a walk-away price

Turn the work into three numbers:

1. Comps-based value. The tool's estimate after your toggles, adjusted by hand for condition, view, parking and tenants.
2. Offer range. The value plus or minus the uncertainty in your comp set. With high confidence, a few percent either way; with thin data, wider.
3. Walk-away price. The most you would pay and still be glad you bought, set with your partner and your budget, written down before offer day.

Share all three with your agent, with the comps you used, and ask which they would change and why. A good agent moves your number with evidence, not with the count of other offers. Keep the write-up; comps also drive the property tax appeal you may file if the market softens.