Vol. I, No. 45 · San Francisco — Friday, October 2, 2026 · Price: nothing left in the car

BY HAZEL WONG, NUMBERS DESK

The crooked block of Lombard Street runs one steep block from Hyde Street down to Leavenworth. In 2023, San Francisco police took 243 reports of car break-ins at those two corners, about one every day and a half. In 2025 they took two. This year, through September 30, they have taken none.

The same thing happened across Russian Hill. The neighborhood logged 1,416 car break-ins in 2023 and 172 in 2025, a fall of 88 percent. Citywide, break-ins fell 71 percent. That matters past one hill. In 2022 San Francisco logged 26,783 car break-ins, more than 70 a day.

The people who live there also say they feel safer, though the city asked them about walking at night, not about their cars. In the city Controller’s survey of residents, posted in full on Wednesday, 80 percent of Russian Hill residents said they feel safe walking alone in their neighborhood at night. In 2023, 29 percent did. Fewer than 50 residents answered each year. Citywide, the share rose from 37 percent to 60.

Russian Hill’s break-ins clustered where visitors stop. The police file’s Russian Hill runs north off the hill to Jefferson Street at the water, so it takes in Ghirardelli Square. Its busiest corner in 2023 was Beach and Larkin streets, at Ghirardelli, with 170 reports. That corner logged 12 in 2025 and one so far this year. Five corners near the water, from Ghirardelli to the Cannery, and the two ends of the crooked block held 846 of the neighborhood’s 1,416 break-ins in 2023, six in ten. In 2025 those seven corners logged 41. Everywhere else in the neighborhood, break-ins fell from 570 to 131, or 77 percent, still steeper than the city’s fall.

The neighborhood logged more than 1,200 break-ins in every year from 2018 through 2023 except 2020. The drop began in late 2023. Russian Hill averaged about 140 break-ins a month from January through August that year and about 75 a month from September through December. In 2022 the last four months had run higher than the first eight, about 170 a month against 115. The count fell to 418 in 2024. The first eight months of this year produced 68 break-ins, against 1,117 in the same months of 2023.

Police counts are reports, and a visitor who stopped bothering to report would also vanish from them. A fall from 243 to two at one pair of corners is too large for that alone. Nearly every driver who came back to a broken window there would have had to stop calling it in.

In volume, the fall is mostly a story about cars. Larceny theft, the police category that holds car break-ins, went from 1,711 reports in 2023 to 414 in 2025. Burglaries fell by a quarter, from 154 to 115, and assaults went from 85 to 64. Robberies, the street crime nearest to what the survey asked about, fell from 53 to 14.

The survey figures rest on small numbers of people. Forty-nine Russian Hill residents answered in 2023 and 46 in 2026, so each share could be off by about 14 points either way. The gap between 29 and 80 is far wider than that. The method changed, too, and the survey’s codebook advises caution in comparing years. The 2026 survey dropped the street interviews used in 2023. Leaving those out moves Russian Hill’s 2023 figure only from 29 percent to 30.

The change that matters more is who counts as Russian Hill. In 2023 the survey placed people by the cross streets they gave. In 2026 it asked them to name their neighborhood, and people near the hill’s edges may have answered differently. Pooling Russian Hill with Nob Hill and North Beach tests for that. Together the three rose from 34 percent to 66, a gain of 31 points. The city gained 23, and the gap between the two is within the survey’s margin of error.

That citywide 60 percent is the highest figure the survey has recorded since it began in 1996, though the question’s wording has shifted over the years. In 2023, 63 percent of residents said public safety had gotten worse in the last few years. This year, 21 percent said so.

The city has its own account of the fall. In December 2023, the mayor’s office credited “bait cars, plainclothes officers and aggressive prosecutions,” part of a push announced at the end of August that year. It said the crime “is driven by a few organized rings,” and that police investigators “can see an immediate drop of as much as 10% in break-ins” when one crew is pulled off the street. Russian Hill’s drop began on that schedule.

Russian Hill’s share of the city’s break-ins fell from 6.1 percent in 2023 to 2.5 percent in 2025. The break-ins fell hardest at the corners where they had been densest, and those are the corners visitors drive to. That fits crews being taken off those corners, as the city describes, better than a slow change in what drivers leave in their cars. The desk did not trace arrests for this story, and the city’s account is its own claim about its own work. The read is a judgment, held with moderate confidence.

Confidence is high on the break-in counts, moderate on the size of the residents’ gain, and high that the gain is real. It is low on whether Russian Hill gained more than the rest of the city.

In all of 2025, the police file holds two car break-ins at the crooked block’s corners. One was on April 24, at the top. The other was on October 29, at the bottom.

For 17 months, Waymo’s filings with California regulators counted its rides in every census tract in the city. Since January 2025, every San Francisco tract reads “Redacted,” while its statewide rides grew from 541,378 a month to 1.35 million.

BY INES CALLOWAY, TRANSIT DESK

In Waymo’s latest report to state regulators, covering April through June of this year, all 723 San Francisco rows say the same word where the ride count should be: Redacted. Each row names a census tract (a small area the Census Bureau draws) and a month. The blanked column counts the rides that started there.

The column used to hold numbers. From August 2023, the first month in Waymo’s driverless filings with the California Public Utilities Commission, through December 2024, every San Francisco tract had a count. Those counts showed the robotaxi company’s rides in the city climbing from 12,617 a month to 400,731. Then the company asked the commission, which licenses robotaxi service, to keep the counts confidential. The only public count of how many rides the company gives in San Francisco ends in December 2024. Its statewide total has more than doubled since, and nobody reading the public files can tell how much of that growth happened in this city.

The request first appears in a sworn declaration signed May 16, 2025, by Van N. Nguy, Waymo’s deputy general counsel. It covered the January-through-March filing and named the ride counts by tract a trade secret. “If Waymo’s ride-hailing and autonomous driving competitors obtained access to Waymo’s Monthly Tract Data,” it says, “they could analyze the data to gain valuable insights into Waymo’s existing customer base, fleet utilization rate” and “commercial progress.”

Nguy also signed the declarations filed in October 2024 and February 2025. Neither claimed the tract counts, and the files that came with them listed every tract. The newest declaration, signed July 30 by Mari Davidson, Waymo’s assistant general counsel, repeats the claim in the same words.

Both declarations also cite a federal data-security program that limits bulk transfers of Americans’ sensitive data to “countries of concern.” They recommend that the commission weigh it before releasing the counts under a public-records request. Waymo still gives the commission a confidential copy of the counts, the declarations say.

The public files still carry the statewide total, and it has grown fast. Waymo gave 541,378 rides across California in December 2024 and 1,348,926 in June 2026, about 2.5 times as many. In December 2024, the last month both numbers were public, San Francisco held 74 percent of them. The climb was not smooth. Statewide rides fell by a third in June 2025, to 583,262, with passengers and miles down alongside, and were back to 778,812 in July. Waymo’s written report for that quarter does not explain the dip.

The same filings still list collisions by tract. Waymo has reported between 86 and 144 a quarter inside San Francisco since the start of 2025, and the file does not say who was at fault. A reader can count the crashes but not the rides, so these filings no longer let anyone outside the company and the commission work out a crash rate per ride in the city.

The timing lines up with a rule change. The commission approved new reporting requirements on November 7, 2024. Its executive director later wrote that the decision “expanded and adjusted the existing data reporting requirements.” Its staff issued new templates that December, and Waymo’s latest declaration mentions them. The blanked counts began with the first full quarter after that, January through March 2025. Waymo’s stated reason is competition. The public files do not show whether the commission has ruled on the request.

The counts are Waymo’s own, verified under penalty of perjury, so confidence in them is high. Confidence is moderate on why the change came when it did; the timing is this desk’s reading of the dates.

The first public count, for August 2023, was 12,617 rides. The last, for December 2024, was 400,731. Every San Francisco row since has carried the same word.

ON THE CALENDAR

SF Devtools Drinkup. An evening of drinks for people who build software tools, hosted by WorkOS; it is 21 and over, at a bar. 784 people had registered as of early Thursday, up from 373 when the desk’s calendar first logged the page on August 27. Luma does not list it as part of a series, and entry is free.

Friday, October 2, 6:00 PM · Southern Pacific Brewing, Mission District · free · lu.ma/drinkup-f26

From the desk’s calendar: public Luma pages, counts as of 1:52 a.m., October 1. Tell the desk about the gathering it missed.

SOURCES & RECEIPTS

1. SFPD Incident Reports, 2018 to present, DataSF wg3w-h783, queried live October 2, 2026. Car break-ins = “Larceny - From Vehicle” plus “Theft From Vehicle”; neighborhood = the file’s “Russian Hill.” Russian Hill: 1,416 (2023), 418 (2024), 172 (2025), 72 (2026 through September 30); January–August, 1,117 (2023) and 68 (2026); 2018–2022 in the chart. Citywide: 26,783 (2022), 23,250 (2023), 6,785 (2025). Hyde & Lombard and Leavenworth & Lombard: 131 and 112 (2023), 26 and 62 (2024), 1 and 1 (2025), none (2026). The corners still log other reports (10 in 2025, 12 in 2026); a 200-meter circle around the block counts 357, 121, 10 and 1 break-ins, 2023–2026. Beach & Larkin, Beach & Leavenworth, Larkin & North Point, North Point & Polk and Hyde & Jefferson: 603 (2023), 39 (2025); rest of the neighborhood: 570 (2023), 131 (2025). Monthly averages are from counts for January–August and September–December: 914 and 686 (2022); 1,117 and 299 (2023). Robbery: 53 (2023), 14 (2025). Counts are reports filed; 2026 fills in as late reports land. City’s account: Office of the Mayor, “Increased Enforcement Against Car Break-Ins Leads to Significant Decrease Citywide,” December 4, 2023, sf.gov. Cont’d from No. 1, July 17, 2026, which found the citywide fall steepest along the tourist corridors. Queries in receipts.json.2. City Survey, Office of the Controller, DataSF nufj-bfbw (one row per respondent, 1996–2026; updated September 30, 2026), weighted. “Safe” = safe or very safe walking alone in your neighborhood at night. Russian Hill: 28.5% in 2023 (49 respondents, ±14.0 points at 95% confidence), 80.5% in 2026 (46, ±13.9). Without 2023 street interviews: 30.0%. Russian Hill, Nob Hill and North Beach pooled: 34.4% (152, ±8.3), 65.5% (151, ±8.9). Citywide: 36.6% (2023), 59.7% (2026); prior high 56.7% (2015). Safety “worse”: 62.7% (2023), 21.4% (2026). Among the 22 neighborhoods with 40+ respondents both years, Russian Hill’s 52.0-point gain is the largest point estimate (next, Financial District/South Beach, 39.5), inside the margin of several others. The codebook advises caution across years.

2. Waymo LLC quarterly data reports to the California Public Utilities Commission, public versions, 13 driverless filings, August 2023–June 2026, downloaded October 2, 2026, from cpuc.ca.gov. San Francisco rides = TripsStart summed over tracts beginning 06075: 12,617 (August 2023), 400,731 (December 2024). From January 2025 every San Francisco row reads “Redacted” (723 a quarter). Statewide TotalTrips: 541,378 (December 2024), 875,592 (May 2025), 583,262 (June 2025), 778,812 (July 2025), 1,348,926 (June 2026); June 2025 passengers 828,029 against 1,260,448 in May. San Francisco collisions (CollisionsAll; the file carries no fault field): 86, 112, 100, 135 (2025 by quarter); 144, 142 (2026). Declarations under General Order 66-D: Van N. Nguy, deputy general counsel, executed May 16, 2025, for January–March 2025, the first to claim “Monthly Tract Data” (his declarations of October 1, 2024, and February 13, 2025, do not); Mari Davidson, assistant general counsel, executed July 30, 2026, for April–June 2026. Both cite Executive Order 14117. Commission letter of May 1, 2025, on D.24-11-002 (approved November 7, 2024). File hashes in receipts.json.

The Frisconian · Published from forty-nine square miles surrounded by reality · thefrisconian.com · @deaconsanti