Spain's record low of evictions in early 2026 and the court reform: a pre-registered test

Abstract

Spain's General Council of the Judiciary (CGPJ) reported 4,005 evictions practised in the first quarter of 2026, −45.4% on the 7,334 of a year earlier. That is the lowest quarter since its series began in 2013, except Q2-2020 under the lockdown (1,383).

The figure was read in different ways:

• one outlet, que.es, saw tenant protection at work;
• PAH Barcelona disputed the fall on the day of release, pointing to the court reform and to the rise in evictions received (16,167, +19.2%) and completed (7,696, +16.6%) by the courts' common services;
• Electomanía later attributed part of the fall to changes in court procedure.

Under Organic Law 1/2025, the first-instance courts became tribunales de instancia district by district: 315 districts on 1 July 2025, 16 on 1 October and 100 on 31 December, 431 in all.

The way the series is collected changed with the reform. The CGPJ's data-collection forms show it. The court form used to carry the item for evictions practised «bien directamente por el juzgado o bien por un servicio común». The new civil-section form has no such item. Practised evictions are now item 1.4.2 of the common-services form, with a new item for voluntary handovers. The CGPJ's methodology page and release notes do not mention the change.

On 4 October 2026, before the CGPJ's next release, we registered a protocol. It tests whether the fall follows the transformation, district phase by district phase.

The data neither establish that the record low is an artefact of that change nor rule it out.

• The main test is inconclusive. Provinces with more evictions in phase-3 districts fell more in Q1-2026 (b3 = −1.23, Holm-adjusted p = 0.18, permutation p = 0.047). The registered sensitivity S3 gives −1.18 (p 0.088). The same regression one quarter earlier, before phase 3 was transformed, already gives −0.755 (two-sided p 0.024; exploratory). Phase 1 left no mark (b1 = −0.13, p = 0.34).
• H3 is supported, but its signal comes from Q4-2025, the quarter before phase 3. The court-to-service ratio fell more where phase 3 weighed more (b3 = −2.28, Holm-adjusted p = 0.026). The common services' figures collapsed in phase-3-heavy provinces in Q4-2025 and recovered in Q1-2026: Madrid's completed evictions went 955 → 193 → 867. Nationally, receipts were −29.6% year on year in Q4-2025, so the +19.2% of Q1-2026 is mostly a rebound.
• H4 is falsified. The concentration is in rent evictions (rent minus mortgage −2.11, p = 0.011). The mortgage estimate itself is near zero but very imprecise: only 18 provinces allow the test.
• H5 passes on doubtful data. Within provinces, phase-1 districts recorded −33% relative to phase-3 districts in 2025. The gap was already −17% in 2024, and the 2025 district file disagrees with the province figures in all 50 provinces.
• No size can be given. The pooled model registered to size the effect is degenerate, with one parameter at its search bound.

What the CGPJ itself reports. It flags estimated data for 64 districts in Q3-2025, 63 of them in phase 1; the first estimate, 5,053, was later revised to 4,839. For Q1-2026 it lists no data for Donostia/San Sebastián, Gijón, Vinarós, all in phase 3. Madrid's rent figure is estimated: national rent evictions are 2,249 in the release's own province sheet and 2,600 in the series.

The out-of-sample test on Q2-2026, published on 16 October 2026, will be added after that date.

1. Introduction

Court statistics are produced by the courts they describe. When the courts change shape, the statistics can change for reasons that have nothing to do with what they count.

In 2025 Spain replaced its single-judge first-instance courts with tribunales de instancia, district by district, in three waves. In the first quarter after the last and largest wave, the CGPJ's count of evictions practised fell to 4,005, −45.4% year on year.

The reading was contested from the first day:

• que.es wrote «Hablamos de 4.005 lanzamientos en total, la cifra más baja desde que hay registros comparables» and «la protección a los inquilinos parece estar frenando la sangría» (22 June 2026). It is one outlet.
• PAH Barcelona, the same day, asked: «¿Cómo es posible que la cifra "oficial" de lanzamientos caiga un 45% mientras la cifra de lanzamientos practicados en las grandes ciudades sube un 16,6%? Exigimos al CGPJ que aclare esta contradicción estadística». It named the reform as «Una pista estructural: el sistema judicial está, literalmente, reorganizándose» and asked for data by judicial district.
• Electomanía, on 24 September 2026, wrote «Una parte de la bajada se explica por un cambio en los trámites judiciales, no por una mejora de la situación de las familias».

This study takes up that question and gives it a pre-registered test. Because the reform was staggered by district, a change tied to it should concentrate in the districts, and the quarters, in which courts were transformed. National events, by contrast, hit every district at once. We wrote the test down before the CGPJ's Q2-2026 release, due on 16 October 2026. The CGPJ's own data-collection forms then turned out to document a change in who reports the series at transformation (§2.2). The study therefore measures the effect of a documented counting change, not of a hypothetical one.

This paper reports the in-sample tests and the checks added after an independent review. The out-of-sample test is pending.

2. Legal and institutional frame

2.1 The rollout

Organic Law 1/2025 (BOE-A-2025-76, in force since 3 April 2025; consolidated text last updated 4 June 2025) sets the order in its first transitional provision:

«La constitución de los Tribunales de Instancia se realizará de manera escalonada conforme al siguiente orden: 1.º El día 1 de julio de 2025 los Juzgados de Primera Instancia e Instrucción y los Juzgados de Violencia sobre la Mujer, en aquellos partidos judiciales donde no exista otro tipo de Juzgados, se transformarán […] 2.º El día 1 de octubre de 2025, los Juzgados de Primera Instancia, los Juzgados de Instrucción y los Juzgados de Violencia sobre la Mujer, en los partidos judiciales donde no exista otro tipo de Juzgados, se transformarán […] 3.º El día 31 de diciembre de 2025, los restantes Juzgados, no comprendidos en los supuestos anteriores, se transformarán en las respectivas Secciones»

None of the later references in the BOE's analysis of the law amends or annuls this provision.

The phases follow from court composition:

• phase 1: districts that had only mixed first-instance-and-investigation courts;
• phase 2: districts with separate first-instance and investigation courts but nothing else;
• phase 3: all the rest, that is, provincial capitals and large cities.

The Ministry announced the totals on 30 June 2025: «El próximo 1 de julio entrarán en funcionamiento 315 Tribunales de Instancia» … «16 nuevos tribunales el 1 de octubre y 100 adicionales el 31 de diciembre, hasta alcanzar los 431 Tribunales de Instancia en todo el territorio nacional» … «Cada Tribunal de Instancia contará además con un nuevo modelo de Oficina Judicial, que centralizará los servicios comunes».

The court office was to start at the same time as the tribunal (fifth transitional provision). The law allows exceptions, and up to six months under the old staff arrangement.

2.2 How the series is collected, before and after

The CGPJ builds its statistics from quarterly bulletins that each court or service fills in, and publishes the forms.

• Before. The form for first-instance-and-investigation courts (BOLETIN 04, version of 2024) has: «1.13. LANZAMIENTOS PRACTICADOS EN EL TRIMESTRE () … () Se contabilizará un lanzamiento por cada bien inmueble cuyo lanzamiento o entrega posesoria se practique, bien directamente por el juzgado o bien por un servicio común.»
• After, in the court. The 2026 form for the new «SECCIÓN CIVIL Y DE INSTRUCCIÓN. TRIBUNAL DE INSTANCIA» has no practised-evictions item.
• After, in the common services. The 2026 form «BOLETIN 56. SERVICIOS COMUNES. TRIBUNAL DE INSTANCIA» («Versión 1er TRIMESTRE 2026»), under «II. SERVICIO COMUN DE EJECUCION», has:
  • «1.4.1. LANZAMIENTOS ACORDADOS POR EL SERVICIO COMÚN DE EJECUCIÓN (decreto)»;
  • «1.4.2. LANZAMIENTOS PRACTICADOS EN EL TRIMESTRE (*)», with «En el caso de que este Servicio Común o disponga de la totalidad de la información deberá recabarla del resto de Servicios comunes del Tribunal de Instancia»;
  • a new item, «1.4.3 Entregas posesorias mediatas producidas en el trimestre».

So from each district's transformation on, the series that the CGPJ publishes as «lanzamientos practicados» is reported by a different respondent, the tribunal's common execution service, on a different form. The common-service series (received, completed) are now reported on that same form. The CGPJ's methodology page and its release notes do not mention the change. The independent reviewer read the same items in the 2025 versions of the forms (court form, and form 56 «Versión 4º TRIMESTRE 2025»), which are no longer linked from the CGPJ's page.

The CGPJ has met this problem before. Its methodology page explains why the seizures series was discontinued: «Se ha dejado de publicar la serie de embargos por la escasa fiabilidad del dato, puesto que, en este caso, muchos juzgados practican directamente el embargo sin que sea necesaria la intervención de los servicios comunes. Las bajadas que se venían observando se deben más a esta causa que a una verdadera reducción del número de embargos practicados.»

2.3 National events in the same window

These change every district at once:

• Pre-filing negotiation. Since 3 April 2025 a civil claim needs prior negotiation (art. 5.2: «en todos los procesos declarativos del libro II»). That includes eviction claims for unpaid rent (LEC art. 250.1.1.º, Book II). Enforcement claims are exempt: «No será preciso acudir a un medio adecuado de solución de controversias para la interposición de una demanda ejecutiva» (art. 5.3). Mortgage enforcement is therefore exempt.
• The suspension of evictions for vulnerable households (RDL 11/2020, arts. 1 and 1 bis). It did not expire at the end of 2025. The BOE's version history of art. 1 reads «Modificación publicada el 24/12/2025, en vigor a partir del 25/12/2025», and that version reads «hasta el 31 de diciembre de 2026» (RDL 16/2025). The suspension then:
  • lapsed when Congress's repeal of RDL 16/2025 took effect («Modificación publicada el 28/01/2026, en vigor a partir del 28/01/2026»);
  • was restored by RDL 2/2026 («Modificación publicada el 04/02/2026, en vigor a partir del 05/02/2026»);
  • lapsed finally on 28 February 2026, when the repeal of RDL 2/2026 was published.

In Q1-2026 it was therefore in force on 1–27 January and 5–27 February. Article 1 bis extends it beyond rent: «en los supuestos de los apartados 2.º, 4.º y 7.º del artículo 250.1 de la Ley 1/2000 … y en aquellos otros en los que el desahucio traiga causa de un procedimiento penal». Those cases fall in the CGPJ's «otros» category, not in mortgage evictions. The Q1-2025 baseline also had a short lapse: the art. 1 history shows a modification in force from 23 January 2025 and another from 30 January 2025.

The protocol (§2.4, §5, §12) stated the suspension dates wrongly and called the suspension rent-only; PROTOCOL.md §14 records the correction.

3. Prior work

What we searched. On 4 October 2026 we ran the following (all in data/prior_work_search.csv):

• 12 queries on Crossref;
• 6 on the arXiv API;
• 8 GitHub repository searches and 3 code searches;
• 2 Crossref title look-ups.

We also read the CGPJ's pages, releases and forms, the Ministry's notices and the press. The independent review found two sources that our searches had missed. We have read them and quote them below.

The question was raised before this study.

• PAH Barcelona. «No nos creemos la bajada de los desahucios anunciada hoy y lo explicamos», published on 22 June 2026, the day of the release. It sets out the contrast this study tests (court-practised evictions −45% against common-service figures rising). It names the reform: the law «ha sustituido los juzgados tradicionales por los nuevos Tribunales de Instancia, … centralizando en servicios comunes las notificaciones y ejecuciones que antes gestionaba cada juzgado por separado». It proposes a real slowdown as one explanation: «puede que no estemos midiendo menos necesidad de desahuciar, sino menos capacidad del sistema judicial para tramitar y ejecutar los desahucios mientras se reorganiza», «o bien el dato esconde una realidad que no se está midiendo bien». It also asks for the test: «Exigimos al CGPJ que aporte datos desagregados por partido judicial que permitan descartar o confirmar esta hipótesis.»
• Electomanía. «Radiografía de los desahucios en España: 67 al día en 2025 y un desplome en 2026 que no cuenta toda la historia», published on 24 September 2026, attributes part of the fall mainly to the pre-filing negotiation and adds the reform. Its words: «Una parte de la bajada se explica por un cambio en los trámites judiciales, no por una mejora de la situación de las familias»; «A esto se suma la conversión de los juzgados en tribunales de instancia, que se hizo en varias fases durante la segunda mitad de 2025 y ha alterado la tramitación en muchos partidos judiciales»; «La bajada de 2026 se explica en buena parte por los nuevos trámites judiciales, y los datos de los próximos trimestres dirán cuánto de ella es real.»
• The CGPJ. It discontinued its seizures series for the same reason (§2.2). Its forms document the change of respondent. Neither its methodology page nor its release notes connect either point to the 2026 eviction figures.

Other coverage.

• The TSJ of Galicia and the TSJ of Cantabria (22 June 2026) report regional falls of 50.2% and 44.8% under the new court names, without discussing comparability.
• brainsre.news repeats the national fall without mentioning the reform.

Research on eviction statistics in Spain uses the CGPJ series as given:

• Nasarre-Aznar and Garcia-Teruel (2018);
• González-Pérez, Vives-Miró and Rullan (2020), on Palma;
• Gonzalez-Morales, Checa-Olivas and Cano-Guervos (2021), with provincial rates for 2015–2018.

The 2018 volume also has a chapter on the data limitations of eviction statistics in Slovenia (Hrast, 2018). We read these to their titles and, where Crossref has them, abstracts.

On eviction records elsewhere, the closest work is American:

• Porton, Gromis and Desmond (2020), on inaccuracies in court records;
• Gromis et al. (2022), on estimating eviction prevalence;
• Tran (2024), on how eviction data travel from community to court.

Two 2026 arXiv papers study Philadelphia's eviction court (2604.21212, 2605.24849). Legal commentary on LO 1/2025 deals, by its titles, with organisation and procedure, not statistics.

GitHub. No repository parses the CGPJ's eviction files: "lanzamientos CGPJ" and "desahucios CGPJ" returned nothing, and code searches for the CGPJ's file names found nothing.

What is new here, narrowly:

• a systematic district-level test of that hypothesis against the staggered rollout, registered before the next release with frozen out-of-sample predictions;
• the reading of the CGPJ's data-collection forms as the documented mechanism;
• the data-quality findings: the CGPJ's missing-data notes follow the phases, and its district, province and release files are inconsistent with each other.

The hypothesis itself is PAH Barcelona's and Electomanía's, and the mechanism is the CGPJ's own.

4. Data and method

Details are in METHOD.md; the registered design is in PROTOCOL.md.

Data. We use the CGPJ's quarterly series «Efecto de la crisis en los órganos judiciales», by province and by TSJ, from 2013 (vintage of 22 June 2026, «_revisado»):

• evictions practised, by type: mortgage enforcement, urban leases (LAU), other;
• evictions received by the common services, and completed by them («cumplimiento positivo»).

We also use:

• the five quarterly releases from Q1-2025 to Q1-2026, with their notes;
• the annual district file «Lanzamientos por PJs 2013-2025»;
• the CGPJ's district-to-province table;
• the Ministry's list of court name changes by phase.

All of these were downloaded on 4 October 2026, hashed and locked before the analysis.

Phases.

• The 315 seats of the phase-1 sheet and the 16 of the phase-2 sheet; the phase-2 list matches the Madrid Bar's. The phase-1 counts also match, region by region, the Ministry's press release of 30 June 2025.
• The remaining 100 districts are phase 3. They include the 30 seats the Ministry's own phase-3 sheet lists for its territory. Ávila, a provincial capital in that territory, is missing from that sheet; it is phase 3 by law and by complement.
• One phase-1 district, San Javier (Murcia), had pilot tribunal de instancia sections before the reform.

Exposure. For each province, the shares W1, W2 and W3 of its 2023–24 practised evictions that lie in phase-1, phase-2 and phase-3 districts.

Tests. The outcome Δg is the change from one quarter to the next in a province's year-on-year log growth.

• H1a. Regresses Δg(Q1-2026) on W3, controlling for W2.
• H1b. Regresses Δg(Q3-2025) on W1.
• Estimation. Weighted least squares (weight: the province's 2024 count), HC3 standard errors with Student-t p-values, a permutation p-value, and Holm correction. Provinces with fewer than 10 evictions in any quarter involved are left out.
• H2 and H3. Apply the H1a model to the services' completed evictions and to the court-to-service ratio.
• H4. Compares rent and mortgage evictions.
• H5. Uses the annual district file within provinces.
• Registered checks. Time placebos one and two years earlier; negative controls (mortgage filings, dismissal claims).
• Size. A pooled model was to size the effect.

After the review (scripts/post_review.py, not registered):

• S3 run as registered;
• each Δg slope split into its Q1-2026 and Q4-2025 parts;
• the same regression one quarter earlier, Δg(Q4-2025), as an in-window placebo;
• H1a without Las Palmas;
• H5 restricted by the consistency of the district file.

5. Results

5.1 What the CGPJ itself reports

The statistics page calls the courts' count of practised evictions the exhaustive one: «El dato recogido en los juzgados de primera instancia y de primera instancia e instrucción (disponible desde el 1T de 2013) si debe ser exhaustivo». It warns: «EN NINGUN CASO DEBE SUMARSE EL NUMERO DE LOS PRACTICADOS POR LOS JUZGADOS DE PRIMERA INSTANCIA CON EL DE LOS PRACTICADOS EN LOS SERVICIOS COMUNES». Since the transformation, both counts come from the same common-service form (§2.2).

Each release since the reform began carries a note on evictions:

• Q3-2025 (published 12 December 2025): «En este trimestre los datos de lanzamientos han sido estimados debido a la falta de información completa en algunos partidos judiciales», and «* Datos estimados debido a la falta de información completa en 64 partidos judiciales». The province sheet says 66. The same note sits on the common-service sheets.
  • The list names 64 districts, 63 of them in phase 1 and transformed that quarter; the other is Cuenca.
  • The first estimate was 5,053 evictions; the current series has 4,839.
• Q4-2025 (published 20 March 2026): «Partidos judiciales con falta de información parcial sobre lanzamientos»: Estella-Lizarra, Guimar, La Línea de la Concepción, Ubrique.
  • All four are phase-1 districts, transformed the quarter before. None of the 16 phase-2 districts transformed that quarter is listed.
  • The stated closing date, 18 December 2025, precedes the end of the quarter.
• Q1-2026 (published 22 June 2026): «Se acompaña la lista de partidos judiciales para los que no ha sido posible disponer de la información en la fecha de cierre de este informe»: Donostia/San Sebastián, Gijón, Vinarós.
  • All three are phase 3, transformed on 31 December 2025.
  • Madrid's rent and «otros» evictions are marked «*Dato estimado». Its 582 evictions (107 from mortgage enforcement) split into 103 rent and 372 other in the release's own province sheet, and 454 rent and 21 other in the series. The raw split is the implausible one: Madrid had 14–28 «otros» a quarter before.
  • Nationally, rent evictions are 2,249 in that sheet and 2,600 in the series. The published −53.9% depends on how Madrid's 475 non-mortgage evictions are split.

The series files that most reusers download carry none of these flags. Imputing the three Q1-2026 districts at their 2023–24 share of their province adds about 30 evictions. What matters is where and when the gaps fall: in phase-1 districts in Q3 and Q4-2025, and in phase-3 districts in Q1-2026.

5.2 Other data-quality findings

• The district file and the province series disagree for 2025, in all 50 provinces.
  • For 2023 and 2024 they agree (2024 differs by one eviction). Earlier years differ by at most 118 nationally.
  • Nationally the district file has 2,943 more evictions in 2025. That is the size of one district's anomaly: Ribadavia (Ourense) shows 2,944 evictions in 2025, against at most 11 in any earlier year.
  • Outside Ourense the provincial differences net to +4. They range from −311 (Valencia) to +154 (Ciudad Real).
  • The CGPJ's annual 2025 release agrees with the province series in every province. It is the district file's 2025 column that does not add up.
• Four district-years drop to zero after years of activity: San Lorenzo de El Escorial in 2025 (at most 119 in earlier years), and Tarazona, Medina de Rioseco and Ocaña in 2023.
• The identities hold. Practised evictions equal mortgage plus rent plus other in every province-quarter since 2013 but one (Almería, Q4-2022, off by one). Provinces add up to their TSJ except once (Madrid «otros», Q1-2019, off by one).
• One column label slips. In the Q1-2026 release, the TSJ sheet of rent evictions labels its Q1-2026 column «25-T1».

5.3 The registered tests

The verdicts are those the registered rules give. The checks below scope how they can be read.

H1. The direction is the one an artefact would produce, but the evidence is weak:

• Not significant. The registered threshold is not met. The registered sensitivity S3, run as written after the review (it had wrongly kept Madrid), gives −1.18 (SE 0.85, 34 provinces, one-sided p 0.088).
• It leans on one province. Without Las Palmas (264 evictions in Q1-2025, 18 in Q1-2026, a fall the CGPJ does not flag) the estimate is −0.80 (p 0.16).
• It may predate the transformation. The same regression one quarter earlier, Δg(Q4-2025), before phase 3 was transformed, already gives −0.755. Its SE is 0.322 (42 provinces), two-sided p 0.024 and permutation 0.004. This in-window placebo is exploratory, not registered. Had it been registered, the rule of protocol §6.3 would mark H1a as confounded. The registered placebos sit a year earlier and are clean.
• The Poisson sensitivity S7 (registered, computed, not shown before) has leads of its own: −0.23 two quarters before and −0.38 one quarter before, then −0.55 at transformation and after.
• Phase 1 shows nothing comparable in Q3-2025.

H1a's estimate (−1.23, 95% interval −3.05 to +0.59) is a slope on a share. Translated into a proportional effect in a district, it would mean anything from −95% to +80%, so we draw no size from it.

H2 and H3 rest on Q4-2025, not on Q1-2026. Δg(Q1-2026) is the difference between Q1-2026 and Q4-2025 year-on-year growth. Splitting each slope into its two parts, on the same provinces:

In Q1-2026 itself, year on year, completed evictions did not rise more in phase-3 provinces, and the ratio did not fall significantly there. The registered results come from a collapse of the common-service figures in Q4-2025 in phase-3-heavy provinces, followed by recovery:

Nationally, receipts were −29.6% year on year in Q4-2025 and +19.2% in Q1-2026. Over the two quarters together they were −4.2%, against −10.6% for completed evictions and −34.3% for practised ones. The +19.2% that framed the debate is mostly a rebound.

The ratio of court-recorded to service-completed evictions was 1.10–1.17 in 2024, 1.38 in Q4-2025 and 0.52 in Q1-2026. So H3 is supported by its rule, but it cannot be read as a shift at the moment of transformation. It records a disturbance in the services' own figures in the quarter before. After transformation, numerator and denominator come from the same new form.

H4. The registered falsification rule is met: the concentration in phase-3 provinces is in rent evictions, not mortgage ones. A change in who reports should not care about the type of claim. Rent-specific channels would:

• the suspension, which covers rent and «otros» evictions but not mortgage enforcement, and lapsed twice in Q1-2026;
• the pre-filing negotiation, which applies to rent claims and not to enforcement.

The mortgage estimate is near zero but very imprecise (interval −3.2 to +3.5), and it contains the rent slope.

H5. By its registered rule the district test is supported, but three facts weaken it:

• A pre-existing gap. The gap was already −17% in 2024, and smaller gaps appear in 2016 and 2020.
• Inconsistent input. The 2025 column it relies on disagrees with the province series (§5.2).
• It depends on that inconsistency. Restricted to provinces where the district file is within 5% of the series, it is −29% (98 districts, 13 provinces). Where the district file is more than 5% below the series it is −43%; where it is more than 5% above, −3% and not significant.

Excluding the 65 districts the CGPJ flagged in Q3 or Q4-2025 gives −34%.

Controls and sensitivity.

• Negative controls show no pattern of comparable size. Mortgage-enforcement filings give +0.12 on the H1a model; dismissal claims give −0.19 (p 0.20).
• Payment-order filings fell 59.1% nationally in Q1-2026. They show no phase-3 slope, but on the phase-1 model (Q3-2025) they show +0.58 (SE 0.25, p 0.025).
• H1a's coefficient is negative in all ten frozen sensitivity analyses, from −0.67 (unweighted) to −1.46 (without Barcelona and Madrid), and in S3 as registered (−1.18). None reaches one-sided p ≤ 0.05. At TSJ level it is −0.99 (p 0.20).
• H1b ranges from −0.43 to +0.30.

5.4 How large? No reliable answer

The pooled model registered to size the effect does not converge to an interior solution. The parameter for two or more quarters after transformation sits at the upper search bound (+3.0), with and without lead terms. In Q1-2026 every district is past its transformation, and the phases do not share one effect, so the model can only fit the data by pushing that parameter to the bound.

The registered decomposition built on it is in data/decomposition.json, and we do not interpret it:

• missing data −0.007 log points;
• phase-related counting +0.52, with a bootstrap interval from −0.26 to +0.66;
• national residual −1.12.

An exploratory alternative uses H1a alone: across H1a's 95% interval it puts the national Q1-2026 total anywhere from 2,769 to 10,471. The data are compatible with a fall that is mostly real and with one that is mostly an artefact of counting.

5.5 The record

4,005 is the lowest quarter since the series began in 2013, except Q2-2020, when the courts were largely closed under the state of alarm (1,383). The next lowest is Q3-2025 (4,839), itself a post-reform quarter. Among first quarters the previous low was 6,579 (Q1-2023).

A drop of this size from Q4 to Q1 has a precedent. The Q1/Q4 ratio was 0.74 in 2025–26 and 0.73 in 2022–23, the winter of the previous first-quarter low.

The figure is a first vintage with disclosed gaps. Q3-2025 was revised by −4.2% after its first publication.

The registered record-low verdicts rest on the degenerate fit of §5.4, and we do not rely on them.

6. Discussion

The question was whether the record low of Q1-2026 is real or an artefact of how evictions are counted after the reform. The CGPJ's forms show that the counting did change: from each transformation on, the series is reported by the new common execution service, not by the court. The pre-registered tests do not show how much that change moved the figures.

The direction of H1a fits a counting effect. So does the timing of the CGPJ's own missing-data notes, which fall on districts in or just after their transformation. Against that:

• H1a is not significant, leans on Las Palmas, and is preceded by a comparable slope in Q4-2025, before phase 3 was transformed;
• the service-side evidence (H2, H3) comes from a disturbance in Q4-2025, not from a shift at transformation;
• the concentration is in rent evictions (H4).

Three explanations remain, and this design cannot separate them:

• a change in who counts (§2.2), the artefact hypothesis;
• a real slowdown caused by the reform itself. Fewer evictions were actually carried out while courts and offices reorganised. PAH Barcelona raised this on release day. The CGPJ's own activity note for Q1-2026 reports «La resolución de asuntos, que también sufrió una disminución en todas las jurisdicciones, se situó en un 13,7 % negativo», and civil resolutions «un 22,3 % menos». A slowdown is neither tenant protection nor counting. H1 cannot tell it from an artefact. Only service-side data could, and those are disturbed in Q4-2025;
• rent-specific real channels. The suspension was in force for most of the quarter. The pre-filing negotiation reaches evictions with a lag that may be longer in congested courts.

What the data establish does not depend on a test:

• The series changed respondent at transformation. The forms show it; the published statistics do not say so.
• The series carries disclosed gaps in each post-reform quarter. The disclosure sits in notes inside the quarterly files.
• The national rent figure for Q1-2026 depends on an estimate for Madrid.
• The district file for 2025 is inconsistent with the province figures everywhere.

None of this implies a fault by anyone. A national court system was rebuilt district by district in six months. The statistics service flagged what it could not collect, and published the forms that show the change. The reform's dates were set by law. The CGPJ applied the same reasoning to its seizures series years ago, and the next quarters will show how much of the fall persists.

7. Limitations

• No district-by-quarter data. The quarterly tests run on 50 provinces, many with few evictions. Phase 2 (16 districts) is not tested on its own.
• Phases follow city size by law. Phase 3 is where large rental markets and congested courts are.
• Pre-reform movement. The exploratory in-window placebo shows phase-3-heavy provinces moving in Q4-2025, before their transformation. Whether that is anticipation (preparation for the change), the phase-2 transformation of large suburban districts, or something else, this design cannot say.
• The 2025 forms were not re-fetched. The documented change rests on the 2024 court form and the 2026 section and common-service forms, which we read. The reviewer read the 2025 versions.
• Weights for the services. No district data exist for the common services.
• Actual start dates of court offices are unknown.
• The in-sample tests are not blind. Only the Q2-2026 test is.

8. Out-of-sample test (pending)

The CGPJ publishes Q2-2026 on 16 October 2026. Before that date (commit 321bce1, 4 October 2026, 12:56 UTC) we froze the code, the in-sample results and two sets of province-level predictions (data/predictions_q2_2026.csv). The rules are in PROTOCOL.md §7 and data/predictions_q2_2026_spec.json. Two caveats:

• The "transient" prediction rests on the degenerate fit. It uses the first-quarter effect of the degenerate pooled model (§5.4). The "persistent" prediction is zero for every province. Scoring them compares one uncertain prediction with a constant.
• H1a was not supported. By the registered rule, the Q2 slope will therefore be read as a replication check, not as a classification into transient or persistent.

The Q2 release will also show whether the CGPJ revises Q1-2026, and whether the common-service figures stay at their Q1-2026 level. The out-of-sample results will be added to this paper after 16 October 2026, without changing any rule.

9. Data, code and licences

The package contains:

• the protocol;
• the code that downloads, checks and analyses;
• every aggregate table in data/;
• scripts/check_headlines.py, which recomputes the main headline numbers from data/ and checks that each appears in this paper and the README.

Code is under Apache-2.0; our data and text under CC BY 4.0.

Source data. The source data are official statistics of the Consejo General del Poder Judicial and a list published by the Ministry of the Presidency, Justice and Relations with the Cortes.

Reuse terms are not verified. On 4 October 2026 the CGPJ's legal-notice page contained only «No hay información disponible». The Internet Archive's last copy with text, of 14 November 2025, is about the case-law database only: «Las resoluciones judiciales que integran la base de datos “Jurisprudencia” se difunden a través de este portal … El usuario de la base de datos podrá consultar los documentos siempre que lo haga para su uso particular. No está permitida la utilización de la base de datos para usos comerciales, ni la descarga masiva de información.» We found no terms covering the statistics files.

What we publish. We do not re-host the source files; data/ holds figures derived from them, with attribution. scripts/fetch.py downloads the originals again and checks them against the registered hashes.

10. Automation and review

AI agents (Claude models) did the work on 4 October 2026:

• verified the legal texts and the rollout lists with curl;
• wrote the protocol;
• wrote and tested the code;
• ran the data-quality checks and the registered tests;
• froze the predictions;
• ran the prior-work searches and drafted the text.

The study coordinator committed the protocol (ee3ef6c) and the freeze (321bce1). The order of events between those two commits rests on the session's command log, not on git.

An independent AI agent then reviewed the study adversarially. It recomputed every number from the raw files with its own code, re-fetched the legal texts and the CGPJ files, checked the phase assignment against the Ministry's regional counts, and searched for prior work in Spanish sources. Its findings, all applied:

• PAH Barcelona and Electomanía had raised the question before us (§3);
• the CGPJ's data-collection forms document the change of respondent (§2.2);
• H2 and H3 rest on Q4-2025 (§5.3);
• the suspension dates were wrong (§2.3);
• sensitivity S3 had omitted Madrid (§5.3);
• several statements were unfair to the press reading of the record.

The corrections to facts stated in the protocol are recorded in PROTOCOL.md §14. Nothing above it was changed, nor any frozen file. The added checks are in scripts/post_review.py.

The reviewer has not yet re-read this corrected version.

11. Competing interests

None known. EasyByte Hub S. Coop. Mad., which runs EasyxLab, has no commercial relationship with the CGPJ, the Ministry, any court, any party to eviction proceedings, or any organisation that campaigns on evictions.

12. Ethics

• Aggregates only. We used court statistics aggregated by province, region and judicial district. No case files, no personal data, no names of judges, staff or parties.
• Polite access. We fetched public pages and files only, with a declared User-Agent, at most one request per second per host and 5 seconds between requests to poderjudicial.es, as its robots.txt asks. Paths that robots.txt disallows were not requested.
• No contact. We contacted no one.
• What a «lanzamiento» is. The CGPJ counts one «lanzamiento» per property handed over, whether or not it is a home, including voluntary handovers. Most are households losing their home. This study measures how they are counted. It makes no claim that evictions did not happen, and none about the effects of tenant protection.

How to cite

EasyxLab (2026). Spain's record low of evictions in early 2026 and the court reform: a pre-registered test. Study S12. EasyByte Hub S. Coop. Mad. https://github.com/easybytehub/easyxlab

References

Law and official sources:

• Ley Orgánica 1/2025, de 2 de enero, de medidas en materia de eficiencia del Servicio Público de Justicia (BOE-A-2025-76), consolidated text and analysis. https://www.boe.es/buscar/act.php?id=BOE-A-2025-76
• Ley 1/2000, de 7 de enero, de Enjuiciamiento Civil (BOE-A-2000-323), consolidated, art. 250.
• Real Decreto-ley 11/2020, de 31 de marzo (BOE-A-2020-4208), arts. 1 and 1 bis: consolidated text and version history; version of 24 December 2025, https://www.boe.es/eli/es/rdl/2020/03/31/11/con/20251224; repeal of RDL 16/2025 (BOE-A-2026-2024) and of RDL 2/2026 (BOE-A-2026-4667).
• Consejo General del Poder Judicial, «Efecto de la crisis en los órganos judiciales»:
  • methodology page;
  • series by province and TSJ (Q1-2026, revised);
  • quarterly releases Q1-2025 to Q1-2026, and annual 2025;
  • «Lanzamientos por PJs 2013-2025»;
  • calendar of publications.

https://www.poderjudicial.es/cgpj/es/Temas/Estadistica-Judicial/Estudios-e-Informes/Efecto-de-la-Crisis-en-los-organos-judiciales/

• CGPJ, «Boletines trimestrales de recogida de datos»:
  • BOLETIN 04, Juzgado de Primera Instancia e Instrucción (2024);
  • BOLETIN 04, Sección Civil y de Instrucción, Tribunal de Instancia (2026);
  • BOLETIN 56, Servicios comunes, Tribunal de Instancia (2026).

https://www.poderjudicial.es/cgpj/es/Temas/Estadistica-Judicial/Informacion-general/Boletines-trimestrales-de-recogida-de-datos/

• CGPJ, «Población por Partido Judicial - Año 2025».
• CGPJ, other notes and pages:
  • press note on Q1-2026 activity;
  • TSJ Galicia and TSJ Cantabria notes of 22 June 2026;
  • legal notice, live and as archived by the Internet Archive on 14 November 2025.
• Ministerio de la Presidencia, Justicia y Relaciones con las Cortes, «Entran en funcionamiento los primeros 315 Tribunales de Instancia», 30 June 2025.
• Dirección General de Transformación Digital de la Administración de Justicia, letter of 19 December 2025 and «cambios-lexnet-numo-fase-1-2-y-3-v1-0-1» (xls).
• Ilustre Colegio de la Abogacía de Madrid, «Comienza la segunda fase de transformación de los Órganos Judiciales», 1 October 2025.

Press and civil society:

• PAH Barcelona, «No nos creemos la bajada de los desahucios anunciada hoy y lo explicamos», 22 June
  1. https://pahbarcelona.org/no-nos-creemos-la-bajada-de-los-desahucios-anunciada/
• Electomanía, «Radiografía de los desahucios en España: 67 al día en 2025 y un desplome en 2026 que no cuenta toda la historia», 24 September 2026. https://electomania.es/radiografia-de-los-desahucios-en-espana-67-al-dia-en-2025-y-un-desplome-en-2026-que-no-cuenta-toda-la-historia/
• que.es, 22 June 2026. https://www.que.es/2026/06/22/desahucios-2026-caida-45-cgpj/
• brainsre.news, on the Q1-2026 figures. https://brainsre.news/desahucios-alquiler-hipoteca-caen-45-interanual/

Research:

• González-Pérez, J. M., Vives-Miró, S., & Rullan, O. (2020). Evictions for unpaid rent in the judicial district of Palma (Majorca, Spain): A metropolitan perspective. Cities. https://doi.org/10.1016/j.cities.2019.102466
• Gonzalez-Morales, J.-G., Checa-Olivas, M., & Cano-Guervos, R. (2021). Impact of Evictions and Tourist Apartments on the Residential Rental Market in Spain. Sustainability, 13(13), 7485. https://doi.org/10.3390/su13137485
• Gromis, A., Fellows, I., Hendrickson, J. R., Edmonds, L., Leung, L., Porton, A., & Desmond, M. (2022). Estimating eviction prevalence across the United States. PNAS. https://doi.org/10.1073/pnas.2116169119
• Hrast, M. F. (2018). Evictions in Slovenia: legal aspects, data limitations and good practices. In Loss of Homes and Evictions across Europe. https://doi.org/10.4337/9781788116992.00016
• Nasarre-Aznar, S., & Garcia-Teruel, R. M. (2018). Evictions and homelessness in Spain 2010–2017. In Loss of Homes and Evictions across Europe. https://doi.org/10.4337/9781788116992.00017
• Porton, A., Gromis, A., & Desmond, M. (2020). Inaccuracies in Eviction Records: Implications for Renters and Researchers. Housing Policy Debate. https://doi.org/10.1080/10511482.2020.1748084
• Tran, A.-T. (2024). Doing What Counts: Following Eviction Data from the Community to the Court. Designing Interactive Systems Conference. https://doi.org/10.1145/3656156.3665128
• arXiv 2604.21212, "Legal Infrastructure Organizes Eviction: Evidence from Philadelphia"; arXiv 2605.24849, "How Eviction Court Governs: A Statistical Analysis of Bargaining, Templates, and Debt in Philadelphia".

Cite this study

EasyxLab (2026). Spain's record low of evictions in early 2026 and the court reform: a pre-registered test. Study S12. EasyByte Hub S. Coop. Mad. https://github.com/easybytehub/easyxlab/tree/main/studies/s12-eviction-statistics

@techreport{easyxlab_s12,
  title       = {Spain's record low of evictions in early 2026 and the court reform: a pre-registered test},
  author      = {{EasyxLab}},
  institution = {EasyByte Hub S. Coop. Mad.},
  number      = {S12},
  year        = {2026},
  url         = {https://github.com/easybytehub/easyxlab/tree/main/studies/s12-eviction-statistics}
}