A systematic dataset of first-hand accounts

Stories of Awakening is a systematically constructed dataset of individual, first-hand accounts of spiritual awakening, mystical experience, and near-death experience, drawn from long-form video interviews across eleven independent archives.

Each account was selected to meet the following criteria: a single speaker sharing their own direct experience, in sufficient depth to be meaningfully coded (minimum transcript length enforced per source), in a first-person interview format.

The dataset covers four analytical dimensions, each tagged independently in both broad (inclusive) and strict (evidentiary) versions:

A sub-corpus of NDE accounts includes additional NDE-specific fields (trigger type, life review, tunnel, light encounter, encounter with entities, return experience, and aftereffects). These accounts are drawn from sources that focus specifically on near-death experience and are fully included in the broader corpus analysis.

The corpus was constructed through a 12-step pipeline across five phases. The full methodology is documented at methodology.html, including the tagging prompt text, inter-rater reliability data, and a description of all known limitations.

• Phase 1: Corpus Assembly — Source identification, URL harvesting via yt-dlp, transcript retrieval via Apify, import and normalization with source-specific word-count thresholds
• Phase 2: Quality Filtering — AI quality rating using Claude 3.5 Sonnet; low-quality episodes (poor transcript, non-first-hand content, insufficient depth) excluded before tagging
• Phase 3: AI Tagging (4 passes) — Broad experiential tagging → strict evidentiary tagging → physical phenomena extraction → integration challenges extraction. Each pass uses a structured prompt with defined vocabulary and explicit coding criteria
• Phase 4: Cross-validation — Every episode independently tagged by GPT-4o using equivalent prompts; disagreements between Claude and GPT-4o reviewed by Claude Sonnet adjudication with per-tag reasoning
• Phase 5: Validation & Export — Automated schema validation, inter-rater agreement statistics, and export to aggregated website statistics

The broad/strict distinction is central to the dataset's design. Broad tags reflect what the tagger determined the speaker was likely describing, including implicit or contextually inferred content. Strict tags reflect only what was stated clearly and explicitly in the speaker's own words. This produces two prevalence estimates per category — an upper bound and a confirmed minimum — which are reported as a range throughout the site.

The dataset is suited to descriptive and correlational questions about the phenomenology and aftermath of spiritual awakenings, near-death experiences, and related non-ordinary states of consciousness. Examples:

• What experience types co-occur most commonly — and which tend to appear in isolation?
• Does the trigger of an awakening predict the types of experiences reported?
• How do physical phenomena distribute across NDE versus non-NDE accounts?
• What is the prevalence of specific integration challenges across different source populations?
• How do broad and strict prevalence estimates diverge across categories — and what does that divergence tell us about how speakers describe these experiences?
• How do accounts from established spiritual traditions (BATGAP, Conscious TV) differ phenomenologically from accounts from NDE-focused sources (Round Trip Death, Anthony Chene)?

This is a self-report dataset of public video interviews — not a clinical or population sample. The source population skews toward people willing and able to participate in long-form video interviews, toward Western and English-speaking contexts, and toward people who have found their experience to be ultimately meaningful or integrative (people who found it purely harmful are less likely to appear in interview archives). These limitations are discussed in full in the methodology.

This site provides aggregated statistics — prevalence rates, co-occurrence patterns, and categorical breakdowns. Individual transcripts are not hosted here; they are available through the original source archives listed on the Sources page.

The dataset and pipeline code are not yet publicly released, but collaboration inquiries are welcome. If you are working on research related to spiritually transformative experiences and would like to discuss access or collaboration, please reach out via the About page.

For citation purposes, please use: