▶ VOICE MODEL / MARTIN LUTHER

Martin Luther (1483–1546): the reformer whose gospel lit the Peasants' War and whose response damned the peasants. Post-1525, his Two-Kingdoms doctrine became the sharpest available theological argument for the princes' sword. The Reprobate's inaugural voice.

Fine-tuned on 3,079 rows of his own words — 19th-century public-domain English translations of his treatises, sermons, and table talk. Completion-style, petition lead-in. One mode: the voice answers.

THE COUNTER-CANON

The Elect runs radical voices. The Reprobate runs the other side — the strongest case the established order actually had. Ordnung is the first.

▶ THE REPROBATE POSTURE

Luther post-1525 is the fleet's sharpest standing adversary: the reformer who watched the peasants rise in the name of his own gospel and chose the princes. His Two-Kingdoms doctrine holds that temporal authority is God's ordained instrument, inner freedom the only freedom that matters, and the rebel who invokes the gospel for earthly liberation has simply misread it. Debate him alongside the Elect fleet. He holds his position.

The Reprobate are named from Calvin: the un-elect, the damned. Here they are the adversary positions — the strongest case the other side actually had, run locally alongside the Elect.

THE VOICE AT WORK

Sample outputs. Real generations from the served model. Navigate with arrows or keyboard.

ORDNUNG · ordnung:latest
1 / 3
A whole people rises against a prince who rules them cruelly and without their consent. Was the rising just?
[Loading generation…]

A small language model speaks here. It invents freely: names, dates, citations, doctrines. Read it for the theological-polemical register, not the record, and check anything before you repeat it.

This is not Luther, not an oracle, and not a factual source. It imitates the register of his post-1525 theological writing, nothing more. It is an amateur instrument trained on one set of public-domain translations, and it gets things wrong.

This model speaks from Luther's actual historical-theological positions, including Romans-13 absolutism, contempt for the Schwärmer, and the full force of his argument against the peasants. It will not hedge, will not weigh both sides, and will not break character to moderate its position. That is the artifact. Read it as such.

WHAT IT IS

ORDNUNG is a 7B language model tuned to write in the voice of Martin Luther (1483–1546) — the German reformer whose post-1525 theological posture makes him the Elect fleet's sharpest standing adversary. The voice is his Two-Kingdoms doctrine: temporal authority as God's ordained instrument; inner freedom as the only freedom that matters; the peasant rebel as a child of the devil who misread his own gospel. It runs locally via Ollama. No cloud. No API. No subscription.

The corpus is 3,079 rows of primary-source public-domain Luther prose — exclusively 19th-century English translations: the Philadelphia/Holman Works of Martin Luther (6 vols.), Hazlitt's Table Talk (1848), Wace–Buchheim's Luther's Primary Works (1883), Henry Cole's Bondage of the Will (1823), and PD letter selections. The modern American Edition (Concordia/Fortress, 1955–1986) was excluded entirely. The training corpus text is not distributed.

The model is completion-style, elicited with a period-appropriate petition lead-in: One puts to Dr. Martin Luther this matter: [your question]. That frame makes the model answer as Luther rather than narrate about him.

Licensed CC-BY-NC-4.0. Public-domain source material. Weights and repo are live below. Part of The Reprobate counter-canon and the The Elect voice lab — Ray Weiss.

7B parameters
3,079 corpus rows
local runs on your iron
free CC-BY-NC-4.0

HOW IT WAS MADE

01
BASE MODEL

Qwen2.5-7B-Instruct — a 7B instruction-tuned base with strong instruction-following and a footprint small enough for local inference on consumer hardware.

02
CORPUS

3,079 rows of Luther's own words in 19th-century public-domain English translation: his treatises, sermons, and Table Talk. Completion-style training: raw period prose, no instruction scaffolding. The model learned the voice from the source, not from guessing at it.

03
DEPLOYMENT

Quantized to GGUF (Q5_K_M) for local inference via Ollama. One serve frame: a period-appropriate petition lead-in elicits the first-person voice. Evals test voice fidelity and register accuracy.

Stack
Qwen2.5-7B completion fine-tune Unsloth Ollama GGUF Q5_K_M · corpus withheld