Homzora’s September 18, 2026 snapshot records 1,495 distinct eviction notice identifiers filed with the San Francisco Rent Board during calendar year 2025. The selected ground counts below describe allegations recorded in those notices, not court findings or completed evictions. They must not be treated as mutually exclusive categories without checking overlap.
Here is the full breakdown, taken from the filings themselves.
Every stated ground, 2025
| Ground | Type | Notices | Share |
|---|---|---|---|
| Non payment of rent | fault | 488 | 32.6% |
| Nuisance | fault | 371 | 24.8% |
| Breach of the rental agreement | fault | 300 | 20.1% |
| Owner move in | no fault | 97 | 6.5% |
| Capital improvement | no fault | 95 | 6.4% |
| Ellis Act withdrawal | no fault | 53 | 3.5% |
| Unapproved subtenant | fault | 25 | 1.7% |
| Roommate in the same unit | fault | 21 | 1.4% |
| Other cause | other | 16 | 1.1% |
| Habitual late payment | fault | 11 | 0.7% |
| Denial of access | fault | 9 | 0.6% |
| Illegal use | fault | 6 | 0.4% |
| Failure to sign a lease renewal | fault | 2 | 0.1% |
| Demolition | no fault | 0 | 0.0% |
What a notice is, and what it is not
The Rent Board states this plainly in its own data documentation: a notice of eviction does not necessarily indicate that the tenant was eventually evicted.
These are filings. A landlord serves a notice and files it. What follows might be payment, a negotiated move out, a withdrawal, an unlawful detainer action, or nothing at all. Counting notices tells you how often landlords initiated the process. It does not tell you how many households lost their homes, and anyone presenting it as a displacement count is overreading it.
A ground flag also records what the landlord stated, not what a court found.
Interpret combined flag counts carefully
The nonpayment, nuisance and breach flags total 1,159 when their separate counts are added. That sum is 77.5 percent of the annual denominator, but it is not necessarily 1,159 distinct notices: the calculation does not establish whether the flags overlap. Report the individual counts unless a separate union of notice identifiers has been computed.
Owner move in, capital improvement, Ellis Act withdrawal and demolition have a combined flag count of 245. Dividing that sum by 1,495 gives 16.4 percent. This is a sum of flags, not a verified count of distinct notices with any of those grounds.
The consequences of different grounds can vary, and filing frequency should not be interpreted as a ranking of harm. This dataset does not establish relocation entitlement, whether a unit permanently left the rental stock or what happened to any particular household.
Why the ground totals do not establish overlap
The fourteen selected ground counts sum to 1,494, compared with 1,495 distinct notice identifiers overall. That near equality does not prove that notices cited only one ground. Overlapping flags and notices without a selected flag can offset one another. Five further source fields were not separately queried in this release, so their totals and overlap remain unmeasured.
The data
This is published as an open dataset, free to download and reuse, alongside the monthly notice counts going back to 2015.
Eviction notice filings by stated ground
Source: eviction notices filed with the San Francisco Rent Board under Administrative Code 37.9(c), published by DataSF. Retrieved 18 September 2026. Homzora is not a licensed real estate brokerage, property manager or law firm, and nothing here is legal advice.
Why the counting unit matters
A useful analysis begins by defining one observation. This release counts distinct notice identifiers within the chosen filing period. It does not count people, apartments, court cases or completed removals. A household could encounter more than one notice, and a notice can include allegations that are later disputed or withdrawn. Without a documented way to connect records, it would be misleading to convert the number of notices into a number of affected households.
The filing date also has a specific meaning. It places a record in this calendar year analysis, but it need not be the day a tenancy ended or the day a court made a decision. A historical record may be corrected after the initial download. The retrieval date therefore belongs beside the coverage period, not in place of it. A later release should explain revisions rather than silently treating earlier published totals as permanently fixed.
Ground counts are not necessarily separate groups
The dataset counts each selected ground independently. If one notice has two selected flags, it contributes to two ground counts while contributing only once to the overall notice total. If a notice has none of the selected flags, it contributes to the overall total without appearing in those fourteen ground counts. Overlap and omitted classifications can occur together, so a small difference between the two totals does not reveal the pattern at record level.
For example, imagine ten notices. One has two selected grounds and another has no selected ground. The ground counts can add to ten even though the notices do not form ten separate single ground observations. This is a hypothetical illustration, not a finding about the San Francisco records. It explains why totals alone cannot support the earlier claim that notices effectively cited one ground or that the five unqueried fields accounted for at most one notice.
What the percentages actually describe
Each percentage in the table divides the count for one ground by the 1,495 distinct notices in the published annual snapshot. The denominator remains the same across the rows. A percentage therefore describes the frequency with which that ground appears relative to the recorded annual total. It is not the probability that a San Francisco renter will receive that type of notice, and it is not the proportion of tenants a court found responsible for the alleged conduct.
Adding the counts for several grounds produces a sum of flags. It does not necessarily produce a count of distinct notices containing at least one of those grounds. The latter requires a separate calculation of the union of the relevant notice identifiers. This distinction matters when writing headlines. A combined number can be arithmetically correct and still become misleading if it is described as a mutually exclusive share of notices without checking overlap.
What would be needed to estimate an eviction rate
A rate needs a numerator and a denominator that describe compatible populations and periods. Dividing a notice count by a general population estimate would not automatically create a useful measure of housing risk. The analyst would need to explain which renter households or rental units are covered, how repeated notices are handled, and whether the notice dataset includes all relevant categories of activity.
This release does not attempt that calculation. It also does not connect notices to the final outcomes of court proceedings or to household moves. Those would be separate research projects with their own definitions, completeness checks and privacy decisions. The absence of an eviction rate does not make notice data useless. It means the appropriate use is narrower: describing recorded filings and their stated grounds within the documented snapshot.
A ground is an allegation, not a finding
The categories in the source describe the basis stated in a notice. A record marked nuisance does not establish that the alleged nuisance occurred. A record marked nonpayment does not establish the amount owed, the availability of a defense or the outcome of any later dispute. Readers should avoid replacing the source label with language that treats an allegation as proven conduct.
The same care applies to categories commonly described as no fault. That label distinguishes a type of stated ground; it does not tell the reader whether a particular notice satisfies every legal condition. It also does not establish the amount of relocation assistance due in a particular case. Requirements depend on the applicable rules and circumstances. Aggregate counts cannot substitute for an assessment of an individual notice.
How a renter can use this research responsibly
The table can help a reader understand the vocabulary used in public reporting and see which selected flags appear most frequently in this snapshot. It cannot predict what will happen after a particular letter arrives. Anyone facing a notice should keep the document, its attachments and records of delivery, and seek timely assistance from an appropriate local legal service. Do not wait for a dataset update to investigate a personal deadline.
For housing research, use this release alongside other evidence rather than as a ranking of neighborhoods or buildings. The download deliberately omits names, addresses and point locations. That limits certain geographic analyses, but it also avoids turning a general research article into a list of identifiable households. No conclusion about an individual renter or property should be drawn from the aggregate figures presented here.
How a journalist or researcher should cite the release
A clear citation should identify the San Francisco Rent Board and DataSF as the source, calendar year 2025 as the coverage period, and September 18, 2026 as the retrieval date for this snapshot. Describe the measure as recorded notices or selected ground flags, depending on the calculation being discussed. Include the limitation that notices are not completed evictions and that a ground records an allegation.
If reproducing the table, retain the denominator and avoid describing the rows as a complete partition of all notices. If reporting a combined category, calculate distinct notices with the relevant flags before calling the result a share of notices. If that calculation has not been performed, report the separate ground counts. A transparent limitation is more informative than a precise looking total whose interpretation cannot be supported.
What a stronger future release would add
A fuller analysis could inspect all nineteen documented ground fields, quantify overlapping flags and identify records with no selected ground. It could publish a reproducible reconciliation between the distinct annual total and the classifications shown in each table. Any comparison across years should first check whether definitions, filing practices, field availability or source revisions changed during the period.
Those are proposed improvements, not results already established by this release. Until they are completed, Homzora will keep the distinction between raw notice totals and sums of selected flags explicit. Readers can obtain the current files from the eviction grounds section and consult the DataSF source. The methodology page describes the broader collection and its limits.