A13: DISSINET-Relevant Use Cases

Use cases specifically relevant to the research profile of DISSINET: medieval religious movements, social network analysis, historical sources in Latin and vernacular languages, inquisition records, digital humanities infrastructure.

The research profile

DISSINET researchers work with:

The use cases below map Claude capabilities directly onto this profile.

Use case 1: First pass on a new source

Scenario: A researcher has a new PDF scan of an inquisition register — 80 pages, Latin, notarial hand. They need to understand what it covers before deciding how much time to invest.

In Claude Code:

I have a PDF of an inquisition register: montauban_1241.pdf
It is in Latin. Please:
1. Give me a 2-paragraph overview of what it covers
2. List all named persons you can identify, with the folio reference
3. Note any place names
4. Flag any unusual features (gaps, interpolations, different hands if visible in text)
Save the results to montauban_1241_first_pass.md

What Claude does well here: Rapid first-pass extraction; identifying the structure of the document; flagging named entities. Latin is well within Claude's capabilities for Medieval Latin of this type.

What requires human review: Interpreting the significance of findings; assessing diplomatic features from text alone (physical manuscript features are invisible to Claude); catching Claude's misreadings of ambiguous name forms.

Time saved: A trained researcher might spend 2–3 hours on a first pass. Claude produces a draft first pass in minutes, which the researcher then verifies and corrects. Total time: 20–40 minutes.

Use case 2: Name variant reconciliation across a corpus

Scenario: A researcher has transcriptions of 30 depositions. The same persons appear under multiple name forms across documents (Guilhem / Guillelmus / Guilhem de Cassers / W. de Cassers). They need a consolidated person list for database entry.

In Claude Code:

I have 30 deposition transcriptions in the folder /transcriptions/
Each mentions persons by name. The same person often appears under variant forms.
Please:
1. Read all transcriptions
2. Compile a list of all name forms encountered
3. Group likely variants of the same person together, noting your confidence
4. For each person, list which documents they appear in
Output to persons_consolidated.csv with columns: name_canonical, variants, documents, confidence_note

What Claude does well: Identifying obvious variants (Latin/vernacular forms, abbreviations, patronymic variants). Grouping high-confidence identifications.

What requires human review: Borderline cases where two persons might share a name; genealogical identifications that require external knowledge; distinguishing father/son with the same name.

Why this matters: This task is tedious and error-prone when done manually across 30 documents. Claude produces a first draft that a researcher can correct rather than building from scratch.

Use case 3: Batch extraction for database entry

Scenario: A researcher needs to extract structured data from depositions for entry into DISSINET's database: deponent name, date, location, inquisitor, persons mentioned, charges.

In Claude Code:

I have deposition transcriptions in /transcriptions/
For each file, extract the following fields:

- deponent_name

- deponent_status (defendant/suspect / witness only / unknown)

- date (as given in the document)

- location

- inquisitor_name

- persons_mentioned (comma-separated list)

- charges_summary (1 sentence)

Output all as a CSV file: extractions.csv
Note your confidence for any field where the source is ambiguous.

What Claude does well: Consistent field extraction across many documents; flagging ambiguities; handling formulaic notarial language.

What requires human review: Every row needs verification before database entry. Claude will make errors — wrong dates, missed persons, misread charges. Treat as a draft, not a final product.

The key productivity gain: Building the extraction template once, running it across the corpus automatically, then correcting errors — is significantly faster than manual entry from scratch.

Use case 4: Grant proposal drafting

Scenario: A researcher is writing a section of a grant proposal describing the project methodology and expected outputs.

In Claude Desktop (or Code):

I am writing a grant proposal for [funding body].
The project: [description]
The section I need: Methodology (max 800 words)
Key points to cover: [list]
Tone: academic, confident, not overclaiming
Funder priorities (from the call text): [paste relevant text]
My previous draft: [paste]

Please revise my draft to make the argument clearer and align better with the funder's language,
without changing my core claims or adding commitments I have not made.

What Claude does well: Restructuring for clarity; tightening prose; mirroring the funder's vocabulary back to them; ensuring the abstract claims match the detailed description.

What requires human review: Every factual claim; all budget and timeline implications; any language that could be read as a commitment.

Caveat: Claude will sometimes strengthen claims beyond what the researcher intended ("this project will definitively resolve..."). Always read the output critically for overclaiming.

Use case 5: Cross-linguistic source comparison

Scenario: A researcher is comparing a Latin source with its Occitan vernacular equivalent, or tracking how a theological concept is formulated across languages in the corpus.

In Claude Code or Desktop:

I have two versions of the same document:

- latin_version.txt (Latin original)

- occitan_version.txt (vernacular translation/adaptation)

Please compare them and:
1. Identify passages that differ significantly between the versions
2. For significant differences, quote both versions and describe the nature of the difference
3. Note any additions or omissions in the vernacular version
4. Pay particular attention to passages about [specific concept]

What Claude does well: Parallel text comparison; identifying structural differences; handling both Latin and Occitan.

What requires human review: The interpretation of why differences exist (scribal variation, ideological adaptation, audience considerations). This is the scholarly contribution — Claude identifies what is different; the researcher explains what it means.

Use case 6: Preparing seminar reading notes

Scenario: A researcher needs to prepare reading notes on 5 recent articles for a seminar or supervision meeting.

In Claude Code:

I have 5 PDF articles in /reading/
For each one, please write:
1. Main argument (2 sentences)
2. Key evidence and methods
3. How it relates to social network approaches in medieval studies
4. One critical question I might raise in seminar

Output each as a section in reading_notes.md

What Claude does well: Rapid structured summaries; identifying methodological approaches; generating discussion questions.

What requires human review: Whether the "critical question" is actually worth asking (Claude sometimes generates generic questions). Whether the characterization of the argument is fair and accurate.

Use case 7: Project organisation and task management

Scenario: A researcher has a complex multi-year project with many moving parts and needs help keeping track of what is happening.

With CLAUDE.md as project memory:

Set up a CLAUDE.md with project structure, current phase, and key deadlines. Keep a progress.md as a running log. Then in conversation:

Read my progress log and CLAUDE.md.
What are the outstanding tasks for the Montaillou subproject?
Draft a prioritised task list for the next two weeks,
given that I have a conference deadline on March 28.

Claude reads the files, synthesizes, produces a task list. The researcher corrects it, saves it back to the project folder.

What this is not: A replacement for proper project management tools. Claude cannot send reminders, track time, or integrate with calendars. It is a thinking partner for organising complexity, not a project management system.

The DISSINET-specific note on data sensitivity

DISSINET works with historical materials (long-deceased subjects, not living ones), so data privacy concerns around AI processing are lower than for researchers using interview transcripts or contemporary data. That is a separate question from whether the team's own critical editions and transcriptions should be freely exposed to AI training — editorial labor has its own value, independent of the underlying source's age or copyright status. However:

Related