# Corpus register — German originals

All German originals are public domain in life+70 jurisdictions since 2026-01-01 (Mann d. 1955).
US column: publication + 95 years. The puller records this register beside every download; the
training manifest carries each work's row. Word counts approximate.

| work | year | words | US status (2026) | tier |
|---|---|---|---|---|
| Buddenbrooks | 1901 | 360k | PD | core |
| Frühe Erzählungen (Tonio Kröger, Tristan, etc.) | 1893–1903 | 120k | PD | core |
| Königliche Hoheit | 1909 | 130k | PD | core |
| Der Tod in Venedig | 1912 | 33k | PD | core |
| Herr und Hund | 1918 | 25k | PD | core |
| Betrachtungen eines Unpolitischen | 1918 | 190k | PD | core — ESSAY WEIGHT ×2 |
| Rede und Antwort (essays) | 1922 | 90k | PD | core — ESSAY WEIGHT ×2 |
| Von deutscher Republik | 1922 | 20k | PD | core — ESSAY WEIGHT ×2 |
| Der Zauberberg | 1924 | 350k | PD | core |
| Unordnung und frühes Leid | 1925 | 15k | PD | core |
| Bemühungen (essays) | 1925 | 80k | PD | core — ESSAY WEIGHT ×2 |
| Mario und der Zauberer | 1930 | 20k | PD | core |
| Die Forderung des Tages (essays) | 1930 | 90k | PD | core — ESSAY WEIGHT ×2 |
| Joseph und seine Brüder I–IV | 1933–43 | 650k | US © until 2029–2039 | extended |
| Leiden und Größe der Meister (essays) | 1935 | 90k | US © until 2031 | extended — ESSAY ×2 |
| Lotte in Weimar | 1939 | 140k | US © until 2035 | extended |
| Doktor Faustus | 1947 | 200k | US © until 2043 | extended |
| Der Erwählte | 1951 | 70k | US © until 2047 | extended |
| Bekenntnisse des Hochstaplers Felix Krull | 1954 | 130k | US © until 2050 | extended |
| Deutschland und die Deutschen (speech) | 1945 | 10k | US © until 2041 | extended — ESSAY ×2 |

**Tiers.** `core` (~1.5M words, all US-PD) is the default training corpus — enough for the
adapter, unambiguous everywhere. `extended` adds the late masterpieces under EU-PD status only;
enabling it is an operator flag on the puller and is recorded in the training manifest. The
dialectic set-pieces (Naphta–Settembrini debates, Kretzschmar lectures) live in Zauberberg (core)
and Faustus (extended).

**Essay weighting.** Essay-register works are duplicated (×2) in the CPT mix and dominate the
reverse-instruction SFT set: the target faculty is argument, and the essays are where Mann argues
in his own narrator's voice rather than through characters.

**Orthography.** Period orthography is preserved (no modernization): diction fidelity includes
spelling rhythm. The tokenizer sees what Mann's typesetters saw.

**Sources.** Project Gutenberg (Gutendex-resolved), Zeno.org, Deutsches Textarchiv. The puller
strips boilerplate, keeps a sha256 per text, and writes corpus/manifest.json.