Courts portal data: cases filed in 2024 and 2025, as of 13 November 2026
Eviction requests at local courts: how many are granted and how long they take
We counted every case with “evacuare” in its title filed in 2024 and 2025 at Romania's 179 local courts: 5,191. Of those decided so far, 53% were granted, in full or in part; of those the court either granted or rejected, 65%. Pick the kind of request and see what usually happened.
What happened to eviction requests
Eviction requests end differently depending on their kind. “Decided” means the case ended at first instance: the request was granted or rejected, or the case ended another way. Pick your kind of request.
59%of 2,698 decided cases (95% confidence interval: 57–61%)
- granted, in full or in part
- rejected
- ended another way: withdrawal, settlement, annulment, lapse, joinder or transfer to another court
Of every 100 decided cases: 59 granted, 29 rejected, 12 ended another way.
Counting only the cases the court granted or rejected, 67% were granted.
Another 194 cases of this kind have no ruling yet and are not in these percentages.
How long until a first ruling
We count from the day the request is filed to the court's first ruling. That can also be an order ending the case by withdrawal, settlement or transfer to another court. Cases with no ruling yet are counted as still pending, not as finished, so the time is not understated. The 180-day and one-year percentages are estimates, because newer cases could not be followed that long.
Of every 100 cases of this kind, how many had a first-instance ruling within:
Half of the cases had a ruling within 80 days.
A ruling at the local court does not mean anyone has moved in or out. If a party appeals, the appeal follows. If the ruling becomes final, a court bailiff carries out the eviction, and those steps are not in this data.
The full table
| Kind of request | Filed | Decided | Still without a ruling | Granted | Rejected | Ended another way | Granted, of those granted or rejected | Ruling within 90 days | Ruling within a year (estimate) |
|---|---|---|---|---|---|---|---|---|---|
| Special eviction procedure | 2,892 | 2,698 | 194 | 58.7% | 28.6% | 12.7% | 67.2% | 54.6% | 90.2% |
| Ordinary action | 2,197 | 1,864 | 333 | 47.5% | 28.9% | 23.7% | 62.2% | 34.4% | 75.3% |
| Interim order (ordonanță președințială) | 102 | 96 | <10 | 5.2% | 44.8% | 50.0% | 10.4% | 81.4% | 94.1% |
| All | 5,191 | 4,658 | 533 | 53.1% | 29.0% | 17.9% | 64.6% | 46.6% | 84.0% |
What the data does not show
- The data show what the first-instance court decided, not who left the home. The appeal and the bailiff come after that and are not in this data.
- A granted case does not mean someone lost their home: the portal does not show enforcement.
- “Granted in part” can mean, for example, that the court allowed some payment but not the eviction. The files do not say which part was granted.
- We do not know who the parties are (tenant, relative, former owner) or why the eviction was sought.
- The 533 cases without a ruling are counted as pending, not as rejected. Some of them are suspended cases.
- A case declined to another court can appear twice, under different numbers. We count it as ended another way.
- Individual courts are not compared: only 39 have at least 30 decided cases over two years, too few for a comparison.
- The portal keeps cases filed from October 2023 onwards, so a comparison with earlier years is not possible.
- We took cases whose title contains the word “evacuare”. Other eviction requests may be filed under a different title and are missing.
- These data have not been compared with official court statistics.
How we counted
Source
The public service of the Romanian courts portal (portalquery.just.ro): all 179 local courts, of which 175 had cases with “evacuare” in the subject, filed between 1 October 2023 and 13 November 2026. The figures use cases filed in 2024 and 2025 so each had time to be decided. The whole collection holds 8,605 first-instance cases: 2,586 filed in 2024 and 2,605 in 2025.
How we read the rulings
A program reads what the court decided in each case: granted, granted in part, rejected, withdrawal or settlement, or another ending (annulment, lapse, transfer, joinder). The program was checked by two AI models (Opus and Fable), each a fresh copy with no access to the program's answer. They received only the ruling text with names removed and a codebook. They are not lawyers and not human reviewers, and their agreement shows that they read alike, not that they are right.
The first check, on 370 cases, found a mistake. Of the 40 cases the program thought were pending, 22 were in fact finished, mostly rejected. That inflated the share granted and shortened the times. On decided cases, agreement was 94%. We corrected the rule, froze the program and checked again on 136 new cases it had never seen. On the 96 decided cases in the second sample, the program and the two models agreed in 97% of cases. Of the cases the program thought were pending, finished in fact: 1 of 39. The two models agreed with each other in 99% of cases.
Rules fixed beforehand and what happened to them
Before counting, we fixed written rules, with the SHA-256 checksum of the files (6c841589a5c7 for the project brief, 14452ce96321 for the 13 November 2026 addendum): at least 3,000 cases filed in 2025, and results for individual courts only if at least 60 local courts had at least 30 decided cases. The first rule said to drop the page if it was not met, and it was not: there were 2,605. The second was not met either: there were 34 courts when we decided and 39 after the program was corrected. The thresholds were chosen before any count, from four large courts, so they were a guess about volume, not a test of the result. We decided to publish anyway, with national totals and no comparison of courts. The check with the two models stayed mandatory and was passed only after the program was corrected, on the second sample. The consistency test between halves of the same court passed (coefficient 0.49 against a minimum of 0.4), but the results stay national.
Right of reply
The method and the table have not yet been sent to the Superior Council of Magistracy or the Ministry of Justice. If they reply, the reply appears here, with the date. No reply does not mean agreement.
Privacy
Party names were not kept, case numbers only as a code, and texts were cleaned of names, addresses and calendar dates. The page shows only totals, no person and no court.