About this resource

Methodology, sources, and limits of the Conrad Weiser conversation tool

This page documents how the Conrad Weiser conversation tool is built — which sources it draws on, how the underlying system works, what it cannot do, and how student data is handled. The information is intended for educators, curriculum developers, and IT administrators who wish to evaluate the tool before classroom use.

Sources

What the system draws on

  1. Paul A. W. Wallace, Conrad Weiser, 1696–1760: Friend of Colonist and Mohawk University of Pennsylvania Press, 1945.

    The standard academic biography. Forms the largest part of the source corpus.

  2. Andreas Paul John, Between Two Worlds — Conrad Weiser 1696–1760 2024.

    The author's narrative nonfiction work, English and German editions, fully indexed.

  3. Frederick S. Weiser (ed.), The Weiser Family: A Genealogy of the Family of John Conrad Weiser the Elder John Conrad Weiser Family Association, 1960.

    For genealogical and family detail.

  4. Conrad Weiser's autobiography ("Johan Friedrich Weisers Buch")

    Birth, baptism, and marriage records from Schoharie and Tulpehocken, referenced through Wallace and the genealogical anchor.

  5. Henry Melchior Muhlenberg, Hallische Nachrichten

    The eyewitness source for Conrad Weiser Sr.'s death in 1746 and Conrad's own death in 1760.

  6. Pennsylvania Council Minutes, July 12, 1742, and Benjamin Franklin's A Treaty Held at the Town of Lancaster… in June 1744 Philadelphia, 1744.

    Verbatim Iroquois speeches, including Canasatego's address to the colonies.

Not included as direct sources: the unabridged Logan-Weiser correspondence in Pennsylvania Archives Series I, Conrad's account book, and Sir William Johnson's papers. These are referenced indirectly through Wallace, who cites them extensively.

Method

How the system generates responses

Retrieval-Augmented Generation

Each user question is matched against an indexed corpus of approximately 1,140 text passages from the sources above. The seven most semantically relevant passages are retrieved before any response is generated. Responses are grounded in these retrieved passages rather than in free-form generation, which substantially reduces the risk of factual hallucination.

Anti-fabrication safeguards

The system operates under explicit constraints against fabricating dates, persons, or events outside the documented record. A canonical-facts block fixes Weiser's birth and death dates, family relations, and other non-negotiable details so that the language model cannot drift on basic biographical facts. A two-pass quality check screens each response for contradictions, anachronisms, and breaks in period-appropriate language.

Off-stage mode

Questions about events after Weiser's death in 1760 — for example about his grandson Frederick Muhlenberg's role as first Speaker of the U.S. House in 1789 — are answered with an explicit posthumous marker ("From where I now rest, I have been told that…"). This makes the distinction between primary knowledge and historical hindsight visible to students.

Limitations

What this tool is not

  • The system is not a substitute for primary-source reading. It is a conversation interface, not an archive.
  • Responses can occasionally drift in tone or detail. The two-pass quality check catches most issues, but not all.
  • The historical figure represented is reconstructed from documented sources. Conrad Weiser's actual interior life, opinions on events not recorded in his lifetime, and unrecorded conversations are not recoverable.
  • Voice mode is limited to three minutes per session due to API cost constraints. Text mode is unlimited.

Student Privacy

How student data is handled

The tool requires no registration and collects no personally identifying information. Conversation history is held only in the active browser session and is not persisted on the server. Voice mode uses Google's Live API for speech-to-speech generation; audio is streamed but not stored. The tool sets no advertising cookies and shares no data with third parties beyond the model provider (OpenRouter), the speech service (Google AI Studio), and Cloudflare Web Analytics for cookieless aggregate visit measurement — the first two required for the tool to function, the third for assessing reach. Cloudflare Web Analytics sets no cookies, does not fingerprint users, and collects no personally identifying information.

Adaptation

Other historical figures and topics

This methodology can be adapted to other historical figures or curricular topics where authoritative source material is available. For inquiries about custom implementations, see codeclover.com or contact the author at andreas-john.net.