---
license: cc-by-sa-4.0
language:
- pl
pretty_name: Polish Legal Apprenticeship Entrance Exams (adw/radc, notarial, bailiff) 2007–2025
size_categories:
- 1K<n<10K
task_categories:
- question-answering
- multiple-choice
tags:
- legal
- polish
- mcq
- law
- benchmark
- eval
configs:
- config_name: default
  data_files:
  - split: test
    path: aplikacje_2007_2025.jsonl
---

# Polish Legal Apprenticeship Entrance Exams — adwokacka/radcowska, notarialna, komornicza (2007–2025)

Native-Polish, single-choice (A/B/C) **legal** MCQ benchmark built from the official entrance
examinations for the Polish legal apprenticeships, published by the Ministry of Justice:

- **adwokacka + radcowska** (advocate + legal counsel — a single shared test from 2009 on;
  two separate exams in 2007),
- **notarialna** (notary),
- **komornicza** (court-enforcement officer / bailiff).

Each item carries the published **`legal_basis`** (the exact statute provision the correct
answer rests on), which makes the set usable not only for accuracy scoring but for
**citation / legal-grounding evaluation** — does a model name the right provision, not just
pick the right letter?

To the best of our knowledge **no equivalent dataset existed** publicly on the Hugging Face
Hub, GitHub, or in the literature (the closest prior art, *GAIus* / arXiv:2507.01259, covers
~146 Civil-Code questions and was never released).

## Dataset at a glance

| | |
|---|---|
| Items | **8250** |
| Exams | 53 exam-units (3 in 2007 @ 250 Q + 50 across 2009–2025 @ 150 Q) |
| Years | **2007, 2009 – 2025** (2008 was not held / not archived — genuine gap) |
| Tracks | `adwokacka-radcowska`, `adwokacka`, `radcowska`, `notarialna`, `komornicza` |
| Format | single-choice, exactly **3 options** A/B/C, exactly one correct |
| Language | Polish (native, not translated) |
| Per exam | exactly 150 questions (**250** for each 2007 exam) |
| Gold distribution | A = 2694 · B = 2697 · C = 2859 |
| Validation | 8250/8250 items have 3 non-empty options, gold ∈ {A,B,C} matching `answer_idx`, a non-empty `question`, and a non-empty `legal_basis` — 0 problems |

### Items per track × year

**2007** is a special vintage: the advocate and legal-counsel exams were **separate**, each exam
had **250 questions** (not 150), and there was no komornicza track:

| 2007 track | items |
|---|--:|
| adwokacka | 250 |
| radcowska | 250 |
| notarialna | 250 |

**2009–2025** (150 questions per exam; advocate + counsel share one combined test):

| track | 2009 | 2010 | 2011 | 2012 | 2013 | 2014 | 2015 | 2016 | 2017 | 2018 | 2019 | 2020 | 2021 | 2022 | 2023 | 2024 | 2025 | total |
|---|--:|--:|--:|--:|--:|--:|--:|--:|--:|--:|--:|--:|--:|--:|--:|--:|--:|--:|
| adwokacka-radcowska | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 2550 |
| notarialna | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | — | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 2550 |
| komornicza | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 150 | 2550 |

(`notarialna` total 2650 = 2550 above + 250 from 2007.)

## Schema

One JSON object per line (`aplikacje_2007_2025.jsonl`):

```json
{
  "year": 2022,
  "track": "adwokacka-radcowska",
  "q_no": 1,
  "question": "Zgodnie z Kodeksem karnym, w stosunku do sprawcy usiłowania ...:",
  "options": [
    "stosuje nadzwyczajne złagodzenie kary,",
    "może zastosować nadzwyczajne złagodzenie kary,",
    "odstępuje od wymierzenia kary."
  ],
  "answer": "B",
  "answer_idx": 1,
  "legal_basis": "art. 15 § 2 w zw. z art. 15 § 1 k.k."
}
```

| field | type | meaning |
|---|---|---|
| `year` | int | exam year — **the legal-state anchor** (see caveat below) |
| `track` | str | `adwokacka-radcowska` / `adwokacka` / `radcowska` / `notarialna` / `komornicza` |
| `q_no` | int | question number within the exam (1–150, or 1–250 in 2007) |
| `question` | str | the question stem |
| `options` | list[str] (len 3) | answer options, order A, B, C |
| `answer` | str | correct option letter (`A`/`B`/`C`) |
| `answer_idx` | int | 0-based index of the correct option (`"ABC".index(answer)`) |
| `legal_basis` | str | the official statutory basis for the correct answer |

The natural item key is `(year, track, q_no)`.

## Domain coverage (by `legal_basis`)

Number of questions citing each code (a question may cite more than one act; **3391 items cite a
named *ustawa* / *Prawo o …* / treaty** — e.g. Prawo o adwokaturze, ustawa o radcach prawnych,
Prawo o notariacie, ustawa o komornikach sądowych — rather than a code):

| code | act | questions |
|---|---|---|
| k.c. | Kodeks cywilny (Civil Code) | 1329 |
| k.p.c. | Kodeks postępowania cywilnego (Civil Procedure) | 1059 |
| k.s.h. | Kodeks spółek handlowych (Commercial Companies) | 616 |
| k.p.a. | Kodeks postępowania administracyjnego (Admin. Procedure) | 349 |
| Konstytucja | Konstytucja RP (Constitution) | 333 |
| k.k. | Kodeks karny (Criminal Code) | 255 |
| k.r.o. | Kodeks rodzinny i opiekuńczy (Family & Guardianship) | 249 |
| k.p.k. | Kodeks postępowania karnego (Criminal Procedure) | 245 |
| k.p. | Kodeks pracy (Labour Code) | 203 |
| k.w. | Kodeks wykroczeń (Petty Offences) | 60 |
| k.k.s. | Kodeks karny skarbowy (Fiscal Penal) | 40 |
| k.p.s.w. | Kodeks postępowania w sprawach o wykroczenia | 21 |

The civil / commercial / civil-procedure emphasis is heavier than a single-track set, because
the notarialna and komornicza tracks lean strongly on k.c., k.p.c. and k.s.h.

## Source & provenance

- **Publisher:** Ministerstwo Sprawiedliwości (Polish Ministry of Justice).
- **2016–2025 hub:** https://www.gov.pl/web/sprawiedliwosc/zestawy-pytan-testowych — each year
  publishes, free of charge, the 150-question set **and** the official answer key (with legal
  basis) for every track.
- **2007 + 2009–2015** come from the Ministry's **archived BIP** (the pre-2019 CMS, e.g.
  `arch-bip.ms.gov.pl`). Same publisher, same official material — only the host differs; archival
  hosting does not change the legal status (materiały urzędowe). Exact source attachment URLs are
  recorded in the build provenance.
- From 2009 on, the advocate and legal-counsel exams are a **single combined test**, so one set
  per year covers both professions. In 2007 they were two separate exams (both included).
- Source formats vary by year: text-layer **PDF** (2007, 2017–2020, 2023–2025), **RTF**
  (2009–2010), legacy **.doc** (2011–2016, 2021 keys), and **.docx** (2021–2022 questions). All
  are digital text — **no scans, no OCR**.

### Build method

- **Questions:** PDFs via `pdftotext -layout`; RTF and legacy `.doc` via **LibreOffice headless**
  (`soffice --convert-to txt`); `.docx` via the same. Stem anchored on `^\d+\.`, options
  `A./B./C.` (a missing period is tolerated), wrapped lines re-joined, page markers / form-feeds /
  BOMs stripped. A numbered instruction preamble (where present) is skipped by locating the first
  `1.` that is followed by an `A./B./C.` option.
- **Answer key + `legal_basis`:** the key is a 3-column table
  `nr pytania | (prawidłowa) odpowiedź | podstawa prawna`. PDF keys via
  `pdfplumber.extract_table(strategy="lines")` (so vertically-centred, multi-line legal-basis
  cells join within their real cell boundaries); RTF / `.doc` / `.docx` keys parsed from the
  tab/space-separated table text.
- **Per-exam validation gate:** exactly 150 questions and 150 key rows (250 in 2007); every item
  has 3 non-empty options, gold ∈ {A,B,C} matching `answer_idx`, and a non-empty `legal_basis`.
  All 53 exam-units passed clean (8250/8250).
- **Encoding:** one font-encoding glitch in the 2007 PDFs (`ż` rendered as `Ŝ`, 11 cells) was
  corrected; article-number superscripts (`art. 36¹`, `art. 8³`) are genuine and preserved.

## Important caveats

1. **Statute drift — answers are correct *według stanu prawnego* on the exam day.** Polish law
   changes; an option correct in 2009 may be wrong in 2025. **Every item is anchored to its
   `year`** — treat each year as its own legal snapshot and **never mix years silently** when
   scoring or fine-tuning. The 18-year span makes this set especially useful for *studying* drift,
   but also makes year-mixing especially hazardous.
2. **Recurring questions (a feature, kept on purpose).** Across the corpus 607 question stems
   recur (826 extra rows). They are kept as **distinct items** keyed by `(year, track, q_no)`,
   because each is a separate exam instance. In **404 of those groups the gold answer differs**
   between instances — options get reordered and statutes change over time, so the recurrences
   directly illustrate statute drift / option shuffling. Only **29** rows are exact duplicates
   (identical stem + options + gold + basis); they are retained for provenance completeness. No
   semantic deduplication was applied — dedupe downstream if your use case needs it.
3. **2007 is special.** Advocate and legal-counsel were **two separate exams** (tracks
   `adwokacka` and `radcowska`), each exam had **250 questions**, and there was **no komornicza**
   track that year. From 2009 on, adw/radc is the single joint `adwokacka-radcowska` track at 150
   questions.
4. **2008 is absent — a genuine gap** (no exam archived for that year), not a collection failure.
5. **komornicza 2009–2018 was an "egzamin konkursowy"** (competitive exam), renamed "egzamin
   wstępny" from 2019. Same 150 × 3-option format; derive the distinction from `year` if needed.
6. **2018 notarialna is intentionally absent.** On gov.pl, the 2018 notarialna *question* file is
   byte-identical to the 2018 adw/radc question file (the wstępny exam shared one question set
   across adw/radc/notarialna that year), while its published *key* matched a different set
   (agreed on only 65/150). To avoid a mismatched question↔key pairing, the 2018 notarialna split
   was dropped; 2018 adw/radc (with its own matching key) is kept.
7. **Invalidated questions (*pytania unieważnione*).** Questions struck during the official
   appeals process are not republished separately; the published key is treated as ground truth.
8. **Scope.** Covers 2007 and 2009–2025. The KSSiP judicial/prosecutorial track (konkurs ASiAP)
   is not yet included (its papers are scans requiring OCR).

## License

- Content is **materiały urzędowe** (official materials) — under **art. 4 of the Polish
  Copyright Act** (*ustawa o prawie autorskim i prawach pokrewnych*) such materials are excluded
  from copyright.
- gov.pl content is additionally made available under **CC BY-SA 4.0**.
- **Attribute Ministerstwo Sprawiedliwości** as the source and **share-alike** on
  redistribution.
- ⚠️ This dataset is built **only** from the official Ministry of Justice publications. It does
  **not** include any content from commercial exam-prep services (ArsLege, testy-prawnicze.pl,
  zostanaplikantem.pl, etc.), whose materials are proprietary.

## Intended uses

- **MCQ accuracy** on Polish professional-legal reasoning, across three legal professions and 18
  years.
- **Legal-grounding / citation eval** via `legal_basis`.
- **Per-act, per-track, per-year breakdowns** of legal-domain competence and **statute drift**.

## Citation

```
Polish Legal Apprenticeship Entrance Exams (adw/radc, notarialna, komornicza) 2007–2025.
Source: Ministerstwo Sprawiedliwości (gov.pl + archived BIP).
Materiały urzędowe (art. 4 pr. aut.) · CC BY-SA 4.0.
```