Calibration

Mateo Reyes has been opening locks for twenty-six years, and the thing nobody tells you about the trade is that the machine lies. He runs a small shop wedged between a dry cleaner and a shuttered video store, and on his counter sits a key-cutting duplicator that promises perfect copies at the press of a lever. Most days it delivers. But Mateo has learned, the hard way, that "most days" is not the same as "every day." A blank with a microscopic burr, a cutting wheel worn three-thousandths of an inch past true, a brass alloy that runs softer than the spec sheet claims — and the copy comes out looking flawless under the fluorescent light. It slides into the cylinder. It even turns, almost. Then it sticks at the last pin, and a customer is locked out of their own front door at eleven at night, holding a key the machine swore was right.

So Mateo developed a habit that costs him forty extra seconds and has saved him from a thousand angry phone calls. He never trusts the duplicator's output on faith. He files the new key by hand against the original, feels for the catch, tests it in the actual lock before the customer ever leaves. The machine gives him a starting point, not a verdict. You could call him paranoid. He calls it staying in business.

If you have ever looked up your home's value on a free online estimate and felt that warm flush of confirmation — that's the warm flush of trusting the duplicator. The question is whether anyone bothered to file the key.

The Comfortable Number

Here is what the automated valuation hands you: a single, confident figure, often rendered in a pleasant green font, sometimes with a little upward arrow. Your house is worth $847,000. It feels like fact. It arrives instantly, free, and with the implicit authority of a system that has supposedly digested millions of transactions. You did not ask how it was calculated. You did not need to. The number is round enough to remember and high enough to feel good about.

What you are not told is the confidence interval hiding behind that clean digit. Many of these portal-style estimates carry a published median error rate somewhere between 2 and 7 percent on listed homes — and substantially worse on homes that are not currently for sale, which is to say, your home. On an $850,000 property, a 7 percent error is a $59,500 swing. On the upper edge of the off-market range, where error rates can climb past 9 percent in thinly traded neighborhoods, you are looking at a $76,000 cloud of uncertainty wrapped around a number presented as a point. The arrow points up. The truth wobbles.

The Machinery Behind the Curtain

Automated valuation models are not magic and they are not malice. They are regression engines trained on whatever transaction and listing data they can scrape, license, or infer. They look at recent sales of "comparable" homes, adjust for square footage, lot size, bedroom count, and a handful of features, then extrapolate. In dense, fast-turning suburban tracts where a thousand near-identical houses sell every year, they perform reasonably well. The comps are plentiful, the homes are similar, the math has something to chew on.

But California is not a thousand identical houses. It is a hillside Craftsman with an illegal-but-charming converted garage, a flood-zone bungalow two blocks from a flood-zone bungalow that flooded, a condo whose HOA just levied a $40,000 special assessment that no algorithm has ingested. The model cannot see the new freeway on-ramp approved last Tuesday. It cannot smell the smoke risk that just reset your insurance premium. It treats a remodeled kitchen and a 1978 kitchen as equivalent if the permit was never pulled. It is, in the most literal sense, a black box: data enters, a number exits, and the reasoning in between is proprietary, unpublished, and unavailable for you to audit.

When the Number Has a Body Count

Let me give the abstraction a face. Picture a seller — call her Dana — who priced her three-bedroom at $929,000 because the automated estimate told her to, then anchored every emotional expectation to that figure. She turned down an early offer of $880,000 as an insult. The estimate, it turned out, had leaned on a comp half a mile away that sold during a brief micro-spike and never adjusted for the fact that Dana's street backs onto an arterial road. Ninety-four days on market later — in a city where the genuine median was closer to 31 — she accepted $861,000. The opaque number did not just disappoint her. It cost her roughly $19,000 against that first offer and three months of carrying costs, and it did it while wearing the costume of objectivity.

This is the locksmith's nightmare scaled up. The duplicator produced a key that looked perfect and turned almost all the way. Nobody filed it against the original. Nobody tested it in the actual lock until the customer was already standing in the cold.

What Filing the Key Looks Like

Mateo's forty-second habit has a real-estate equivalent, and it begins with a refusal — the refusal to accept any single number that will not show you its work. The antidote to the black box is not a better black box. It is transparency: knowing where each figure came from, what month it describes, which institution measured it, and how the methodology arrives at its conclusion. You want sales counts and inventory and days-on-market for your specific city, not a statewide blur that averages San Francisco scarcity against Inland Empire surplus into a meaningless midpoint.

This is precisely the gap that independent, methodology-published housing data exists to fill. A resource like California Housing Market News earns trust the way Mateo earns it — not by being instant and flattering, but by being checkable. It publishes monthly, city-level reports drawn from public institutional sources like Redfin, FRED, the Census, and the California Department of Finance, lays its methodology out in the open, and stays independent of any brokerage, lender, or portal whose incentives might quietly tilt a number upward. You are not asked to believe the figure. You are shown how it was built, so you can file it against reality yourself.

The contrast is not subtle once you put the two documents side by side. One is a single figure, untethered from its origin. The other is a record you can trace, date, and challenge — a number that invites the same scrutiny Mateo gives every fresh-cut key.

The Discipline of Distrust

None of this means automated estimates are worthless. Mateo still uses his duplicator every single day; it would be madness not to. The estimate is a starting point, a first cut, a rough orientation in unfamiliar territory. The error is not in consulting it. The error is in stopping there — in mistaking a starting point for a destination, a mood for a measurement.

So treat the convenient number the way a careful tradesman treats his machine. Take it. Then go find the original to file against: the dated, sourced, city-specific figures that tell you not just what your house might be worth but why, and on what evidence, and as of when. Ask of any housing number the three questions Mateo asks of any key — where did you come from, who measured you, and have you actually been tested in the lock? If the number cannot answer, it does not get to be your verdict.

It is nearly closing time, and Mateo is filing a fresh-cut key under the lamp, the original held flat beside it, his thumb riding the new ridges for any catch the machine missed. There is one — a faint high spot on the third cut. He takes it down with two strokes and tests again. Clean. He hands it over, and the customer never knows how close they came to standing locked out in the dark, betrayed by a copy that looked perfect.

Your home is the most expensive lock you will ever own. Do not let an instant, unaccountable number decide whether the key fits.

Take the estimate for what it is — a starting point, generously offered. Then pick up the file, find the original, and test it yourself before the cold sets in.