Famulus Elucidation

Famulus · The semantic core

Famulus Elucidation.
What a sentence says, and on which evidence.

A critical edition explains a passage in a note: who is meant, where and when, what the sentence asserts and what it leaves open, and on which evidence each reading rests. Famulus Elucidation is the part of the platform that writes such notes for the sentences of a collection. Normalisation, hermeneutics, proof, logic, language and the data structure each contribute one layer; this page brings them together and leads to one page for each.

Working state of October 2026. Papers and technical documents are drafts; the documentation sites are copies of the living platform documentation. Every number on this page is read from the workspaces when the page is built.

2,270sentences in the Kármán lab copy
6,556footnotes written for them
1,036sentences of Euler's E57 read as a book
10frozen probes
263test functions in famulus-proof
13,081persons in the global register

The goal

An elucidation beside the transcription

A reader of the Kármán Nachlass marks a phrase in the transcription. In the edition viewer the phrase offers one badge; opening it reads the whole sentence and shows its explication — the meaning, what it is about, each referent with its register identifier, the questions that stay open — and hands the sentence to the probe set. Beside it a Bedeutung view lays the normaliser's footnotes over the page as marks with their grounds, and a dossier turns the page into sourced facts of a person. The names of these surfaces are not fixed yet. 2 sentences have been read from the reader so far; the layers they draw on are described on the pages below.

What the reader sees

The workbench

Probe Lab

Where the reading is developed. A collection with a lab sidecar is chosen, then a probe. The explication of that sentence is shown at once, and the audit keeps to the few factors that decide it. Learning compares a candidate with the package rules on the probe set, which grows when a review verdict is marked. 2 collections have a lab sidecar; 1 expectations come from such verdicts.

The probe lab

Layer one

Normalisation

The first stage of understanding. Every expression of a sentence is read as what it stands for: a register person, a place, a day, the writer or the addressee, an earlier person taken up by a pronoun. Each reading is a footnote with its grade and grounds. In the Kármán lab copy 2,270 sentences carry 6,556 footnotes.

Normalisation

Layer two

Hermeneutics

Reading a sentence against the world of its source. Specialists answer what the text presupposes — a calendar date, a position in the sky as the tables of the time gave it — and an explication states the sentence's meaning, its referents and what the page does not settle.

Hermeneutics

Layer three

Proof

From a question to a justified answer. The evidence space narrows the objects of a collection by their attributes; an argument takes addressed facts as premises, and a conclusion is accepted only when it follows from them.

Proof

Layer four

Logic

What follows from what. The Famulus Normal Form writes a sentence as literals over atoms so that entailment and opposition between the propositions two sentences express can be decided by joins in DuckDB; a test bench measures classifiers of these relations.

Logic

Layer five

Language

The rules of use. LWgrammar holds 528 term entries and 127 usage rules in its reference store, and a dictionary for each collection built from the corpus itself. The reader's dictionary tab already uses it.

Language

Foundation

Data

Every layer reads and writes DuckDB files next to a collection: the Stanza analysis, the fact sidecar, the sentence record, the dictionary, and one global register of 13,081 persons, 2,892 places and 6,357 institutions.

Data

How the parts become one

One home

The layers grew in parallel workspaces. famulus-proof becomes their single home: famulus-logic and famulus-normalisation are merged into it, and one documentation site replaces five.

Improvements are made in general terms only: declared attributes, rules and instructions that a learning loop may revise and frozen probes may refute. A fix that names a case is not accepted.

Famulus Proof documentationMap of the workspaces

Papers and technical documents

Drafts and technical write-ups of the layers; the PDF files are bundled with this page.

Sources and open points

What this page shows comes from the following workspaces.

famulus-proof

Sentence record, expression normaliser, explication, logical form, evidence space, arguments, goal enfolding, the probe lab.

5,653 lines in the normaliser · 263 test functions · 10 frozen probes

famulus-logic and famulus-normalisation

The relation bench and a separate statement model with its own time module; both are to be merged into famulus-proof.

5,243 and 1,800 lines · 51 test functions in famulus-logic

hermeneutic-experts, LWgrammar, the global register

The calendar and astronomical specialists, the language institution, and the register of individuals on PDAI-1.

2 specialists · 528 term entries · 13,081 persons

Open points

  1. The first stage has not yet been applied to the Kármán collection itself: the sentence sidecar exists only as lab copies (the Kármán sample and E57), not in the collection folder.
  2. In the Kármán viewer the explication and the dossier answer for any letter; the Bedeutung view reads the collection's own sentence record, which does not exist yet (0 sentences carry one), and falls back to the lab copy, labelled as such — pages outside the lab sample show no marks.
  3. The reader's surfaces are three panels under three names — Bedeutung and Explication in the viewer, Erläuterung / Elucidation on this page. One name and one panel are still to be decided; the explication's labels are English inside a German viewer.
  4. Against the frozen probes the package rules pass 6 of 7 train, 1 of 2 held-out, 5 of 6 book and 19 of 33 guard expectations. Learning rounds on the relevance model: 3 run, 0 kept.
  5. famulus-logic and famulus-normalisation are still separate workspaces; the merge is planned.
  6. Several documentation pages are in German; the copies are those of the date of the build.