The buildings still betting against the earthquake

Vol. I, No. 13 · San Francisco — Wednesday, July 29, 2026

BY GWEN OKAFOR, HOUSING DESK

I read the Rent Board’s annual reports the way other people read novels, so I know what a compliance list is supposed to say: it is supposed to say a rule was made and mostly obeyed, and to bury the exceptions in an appendix. San Francisco keeps such a list for its soft-story buildings — the wood-frame apartment houses, three stories and up, five units and more, built before 1978, that sit on a ground floor too open to hold them steady when the ground moves. After the 1989 quake collapsed exactly these buildings in the Marina, the city spent years arguing about whether it could order private owners to fix them, and in 2013 it decided it could. The list of who has and who hasn’t is now 4,945 buildings long. I have read it the way I read an eviction filing: the claimed outcome first, then the evidence for it.

The claimed outcome is a success, and by the numbers it is one. Of the 4,945 buildings, 4,678 — ninety-five percent — have finished the work and hold a certificate of final completion. That is a remarkable thing for a mandate to have accomplished against private money and a great deal of it: thousands of ordinary apartment buildings quietly bolted down over a decade, most of them without their tenants ever knowing the crew on the sidewalk was buying them a few more seconds. The program worked. It is the evidence for the other 262 that I want to lay out, because a life-safety rule is not judged by the buildings that obeyed it.

The 262 are not scattered evenly. The program ranked its buildings into four tiers by hazard and use, each with its own deadline, the first falling in September 2017 and the last in September 2020 — so every open case on the list is at least five years late. Tier I, the seven buildings with the most vulnerable occupants, the care homes and schools and assembly halls, is finished to the last address; the city put the frailest first and got them done. But run down the tiers and the compliance thins: Tier II is 2.9 percent unfinished, Tier III 4.5 percent, and Tier IV — the buildings with a store or restaurant on the ground floor, the ones that came due last and cost the most to fix because you cannot simply empty a shop the way you can a garage — is 9.2 percent non-compliant, better than three times the Tier II rate. The mandate got easier buildings done and left the mixed-use, tenant-and-commerce buildings for last, and last is where it stalled.

And where the holdouts stand is the part I would not let the file bury. Strip the numbers off the addresses and a map draws itself: the non-compliant buildings cluster along the Mission and Castro spine — Mission Street, 16th, 23rd, Folsom, Guerrero, Church, Castro — and along the old northeast hills, Lombard and Hyde and Larkin and Mason and Sacramento and Jackson, the dense pre-war stretch running up through Russian Hill, Nob Hill, the edge of Chinatown and down into the Tenderloin. Lombard Street alone carries six of them; Sacramento, Mission and 16th five apiece. These are not empty lots or absentee curiosities. They are full apartment buildings — five units and up, by the program’s own definition — in the most crowded, most rent-controlled, oldest-stock corners of the city, which is to say the corners where the most people sleep over the fewest bolts. The file tells me how many and where. It does not tell me why each owner has run half a decade past a life-safety order, and I will not invent a reason; the honest read is that money, contested work, and buildings too full to empty are slow, and a mandate is only as fast as its last holdout. My confidence is total on the count and the geography — it is a compliance ledger, refreshed last July, and it says what it says. On why these particular 262 are still soft, the file is silent, and so, for now, am I. The earthquake will not be.

The seismographs near San Francisco logged almost twice their usual traffic in 2025. It would be a fine year to feel vindicated about the retrofit, except that nearly half the shaking came from one town on the far side of the East Bay hills, doing again, harder, what it last did in 2015 — and came to nothing then, too.

BY SOL BREGMAN, CONSULTING ACTUARY

I am the one name on this masthead happiest when nothing happens, and 2025 was, on first glance, a year that happened. Within a hundred kilometers of San Francisco — a circle that takes in the Hayward, the Calaveras, and the northern reach of the San Andreas — the seismic network recorded 98 earthquakes of magnitude 2.5 or greater. Set that against the recent run of quiet years — 39 in 2020, 42 in 2022, 44 in 2023, 39 again in 2024 — and it is better than double the ordinary. A person inclined to read the almanac as an omen, with a soft-story deadline fresh in mind, could make something of that. My trade is to resist exactly that inclination, so I went to see where the 98 came from.

They came, most of them, from San Ramon. A single swarm on the Calaveras fault, centered under that Contra Costa suburb some thirty miles east of here, threw 45 of the year’s 98 quakes — 24 of them in December alone. Take San Ramon out of the tally and San Francisco’s year falls to 53, a thoroughly unremarkable number. The “busy year,” almost half of it, was one fault clearing its throat well outside the city, felt here as a rumor if at all.

San Ramon has done this before, which is the whole point of an almanac. In 2015 the same patch of the Calaveras produced a swarm famous enough to make the papers — 36 quakes in that year by the same count — and geologists watched it closely for the thing swarms are feared to be: the small talk before a large one. It was not. It faded. A swarm is not a foreshock sequence; it is a cluster of small ruptures relieving a little stress and, far more often than not, signifying nothing you can bank on. 2025’s swarm was the larger of the two, 45 to 36, and I would caution anyone against reading that as escalation, because it is exactly the reading 2015 punished. Meanwhile the genuinely consequential number stayed small: the largest quake anywhere in the circle all year was a magnitude 4.29 near Berkeley in September, briefly felt, harmless; nothing crossed 4.3, and not one event reached magnitude 5. For scale, the network has logged 169 quakes of magnitude 4 or greater within this circle since 1900, and the truly busy modern year on record — 1990, with 122 — was busy because something had actually happened the autumn before, out at Loma Prieta. That is what a meaningful year looks like: a mainshock and its long tail. 2025 had the tail of a fault that never produced the shock. My confidence on the counts is complete; my read is the actuary’s standing bet, that a loud year and a dangerous one are different animals. One honest caveat cuts the same way I do: the modern network hears more small quakes than the old one did, so part of any long rise is better microphones, not busier faults. The retrofit on the front page is a wager against the rare, real event. The almanac’s job is to remind you it remained, in 2025, rare.

San Francisco adopted Vision Zero in 2014 with a single deadline attached: no one killed on its streets by 2024. The traffic file is in for the target year. In 2024, forty-three people were killed — the deadliest year in two decades of records.

BY INES CALLOWAY, TRANSIT DESK — ABOARD THE 22 FILLMORE, NORTHBOUND

A deadline is a promise with a date on it, and this one had both. When the city signed on to Vision Zero in 2014 — the pledge, borrowed from Sweden, that not one traffic death is acceptable and the toll can be driven to nothing — it wrote down the year it would arrive at zero. The year was 2024. That year is now closed in the crash file, so we can do the plain thing and check the promise against the count. In 2024, forty-three people were killed in traffic on San Francisco streets. It is not only that the city missed zero. It is that 2024 was the worst year in the entire record, 2005 to today — deadlier than 2007’s 42, deadlier than the 39 of 2022, the highest single-year toll the file has ever held. The target year was the deadliest year.

Across the eleven years the pledge was in force, 2014 through 2024, 328 people were killed on these streets. Twenty of 2024’s fatal wrecks were a pedestrian struck — about half the year’s dead were on foot — and another eight were a single car into a fixed object, the wreck of one driver and no one else. They did not concentrate at one notorious corner, either; the deaths spread across the map, four in West of Twin Peaks, four in Bayview–Hunters Point, three apiece in the Mission, Golden Gate Park, and the Tenderloin. This was not a bad intersection. It was a bad year, everywhere.

Here honesty requires a hand on the brake. The file shows 2025 falling to 25 killed, and this year, through June, at 14 — numbers that look like relief. But this is a police-reported crash file that backfills for months as reports are entered, so the freshest years are the least complete and will very likely rise; and 25 sits comfortably inside the ordinary band this record has always wobbled through, 20 to 43, while 2024 is the outlier at the top of it. The safe reading is not that the city turned a corner in 2025. It is that 2024 was a spike, 2025 is a partial reversion we cannot yet trust, and the ten-year promise ended, in the year it came due, at the highest toll on the books. I hold that with high confidence on 2024, which is a closed and complete year, and deliberately low confidence on where 2025 finally lands. Vision Zero is not a lie; it is an aim, and an honorable one, and the injured count — better than 3,500 hurt every year — is the wider war it is really fighting. But a deadline is a promise with a date on it. The date came. The number was forty-three.