Methodology & Rights

A technical account of how this index was built, what safeguards protect its accuracy, and how much confidence each claim warrants. A plain-language summary is on the How this was made page.

Unofficial study aid. This site is a personal, non-commercial research index. It is not affiliated with or endorsed by Ligonier Ministries or St. Andrew's Chapel, and it is not a substitute for R.C. Sproul's primary works.

1. Purpose and scope

This project indexes positions — discrete theological claims — that R.C. Sproul (1939–2017) stated in his publicly available recorded and written teaching, and weights each by how strongly the available sources attest it. The goal is a navigable, source-linked map of what he taught and how consistently, not a republication of his work and not an interpretation of it.

It is deliberately not exhaustive: it covers freely-available web text from Ligonier Ministries plus a small set of public video, and excludes his books (see §7).

2. Sources and corpus

818 source documents across four independent channels:

Channel	Count	Acquisition	Fidelity
Sermons (St. Andrew's Chapel)	281	Official Ligonier transcripts	Human-edited
Articles (incl. Tabletalk)	422	Official Ligonier transcripts	Human-edited
Q&A answers	105	Official Ligonier transcripts	Human-edited
Video	10	YouTube captions (ASR)	Machine, lower fidelity

The 808 Ligonier items were enumerated from learn.ligonier.org XML sitemaps (permitted by robots.txt; /api and /auth were never accessed) and filtered to Sproul by author/teacher metadata. Sermons, articles, and Q&As were retained only where the page identified R.C. Sproul as author/primary teacher. From these, ~10,356 quote-grounded claims were extracted and reduced to ~8,700 canonical positions.

3. Pipeline (reproducible)

All code is in the repository; the corpus regenerates from ./run.sh.

ingest  → extract → weight (embed → topic → cluster → corroborate) → render

1. Ingest each channel to local transcripts (raw transcripts are not republished — see §8).
2. Extract claims with a locally-run LLM (gemma4:e4b via Ollama), constrained to emit, for every claim, a verbatim quote plus a one-sentence neutral summary and a topic label.
3. Weight: embed every claim (nomic-embed-text); assign each to one of 33 canonical theological topics by nearest-anchor similarity; cluster semantically within a topic so one cluster = one canonical position; then apply the corroboration gate (§4).
4. Render the published positions to this static site.

4. Validity safeguards

• Verbatim-quote grounding (code-enforced). The model's quote is re-checked in code as a literal substring of the transcript; ungrounded quotes are dropped, not trusted. No position exists without a real quote.
• Source-trust tiers. Official human-edited transcripts (sermons, articles, Q&As) = trust 1.0; whisper transcription = 0.8; YouTube auto-captions = 0.5.
• Independent-corroboration gate. A position publishes only if it is attested by ≥2 distinct sources, or by a single trust-1.0 (official transcript) source. A position resting on a single low-trust source is quarantined to a review queue and never presented as settled.
• Why this matters: transcription errors self-eliminate. A real example — an auto-caption mis-rendered "amillennial" as "Arminian"; because no other source corroborated the garbled claim, it was quarantined automatically while the correct statement (from an official transcript) was published.
• Contradiction audit. The corroborated core was searched for internal contradictions (same-topic positions about a near-identical point) and adjudicated by hand; none were found (see contradiction-audit.md).
• Human review gate. Nothing is presented as settled without passing the above; quarantined and flagged items are held back transparently.

5. Reading the confidence signal

Each position shows a corroboration count. Interpret it as evidentiary weight:

• Well-attested (≥2 independent sources, often across genres — a sermon and an article and a Q&A) — high confidence this is a settled, repeated emphasis of his.
• Single official source — confidently his (verbatim, from a human-edited transcript), but stated in one place in this corpus.

We deliberately do not use the language model's self-reported confidence: it was found to be uncalibrated (near-1.0 on virtually everything). Corroboration and source trust are the only weighting signals.

6. Limitations (read these)

• Summaries are machine-paraphrased; quotes are authoritative. The bold one-line summary on each position is an LLM's neutral restatement and may compress or slightly skew nuance. The verbatim quote and its source link are the ground truth — always defer to them.
• Topic assignment and clustering are approximate. A position may sit under a topic you'd file differently, and distinct points are occasionally not merged (or vice-versa).
• "A view he describes" vs. "his own view." Sproul often explains positions he rejects (e.g., Roman Catholic justification, semi-Pelagianism). These are retained with their subject named ("Rome teaches…"), but a careless reader could misread an exposition as an endorsement. Stance is not yet machine-tagged.
• Coverage is partial. Books are excluded; video transcription (10 items) is machine-generated and lower-fidelity; the corpus is a large sample, not the whole of his output.
• Not peer-reviewed and not official.

7. Out of scope

Sproul's books are not ingested. They are published variously by Reformation Trust (Ligonier) and other houses, carry distinct rights, and are not freely crawlable. Daily devotionals, conference event pages, and teaching-series landing pages were evaluated and excluded — devotionals are predominantly other authors', and conference/series pages are aggregators whose underlying messages are already captured as sermons or sit behind /api.

8. Rights and attribution

All indexed content was authored by R.C. Sproul. The sermons, articles, and Q&A answers are published by Ligonier Ministries (learn.ligonier.org, publisher metadata confirmed) and are governed by Ligonier's Copyright Policy; Ligonier holds or administers their copyright. The 10 video items were transcribed from third-party re-uploads on YouTube; the original recordings remain the property of their respective copyright holders (in most cases Ligonier Ministries), and the re-uploading channel is not a rights holder. No books are included.

This site reproduces only brief quotations for study and indexing, claims no ownership, stores no full transcripts publicly, and links every quotation to its source so readers can consult the original. If you are a rights holder and have concerns, the project will respond promptly.