A record of significant additions, removals, and methodology changes, so you can see exactly how the collection has evolved and why.
Some people in this collection have told their story more than once, sometimes to several different interviewers. Until now each of those conversations was counted as a separate account, which overstated how many people the collection actually represents and treated one person's story as though it were evidence from several. Every conversation is now linked to the person who gave it, and the figures on this site count people rather than interviews. The individual conversations are all kept and studied separately, because a story told twice is not the same as a story told once. The effect is that the totals go down and become more accurate. Where a single conversation contains two people each describing their own experience, it now counts as two.
Not every interview in a collection like this is someone describing what happened to them. Some are teachers explaining ideas, some are panels where several voices blend together, and some are people discussing other people's experiences. Two independent AI systems now read every full transcript and decide, using the same written criteria, whether the speaker describes an experience of their own in enough detail to study. Where they disagree, or where either is unsure, a person decides. The test is whether there is enough first-hand material to work with, not how much of the conversation it takes up: an interview where the host talks for most of the time still counts, because the part that matters is the part the speaker says about themselves.
Near-death experiences and awakenings are different things and deserve to be studied separately. But they often arrive together: many people describe a near-death event that opened into a lasting change in how they experience themselves. Those accounts stay here, with the near-death event recorded as what set the change in motion. Only an account that is the near-death experience alone moves to a separate collection. Deciding it any other way would have made a whole category of awakening impossible to see in the data.
A review of the earlier work found that the two AI systems reading these accounts had not been given the same instructions, and had not been shown the same amount of each transcript. That made the differences between them look like disagreement about the accounts when some of it was really a difference in what they had been asked and given. Every definition now lives in one place, every system receives exactly the same wording, and every system reads every word of every transcript rather than the first portion of it. Nothing is labelled at scale until a small sample has been run and checked by hand first. The figures on this site will change when the new reading is finished, and they will be more trustworthy.
The collection is now stored in a versioned archive rather than as a single working copy. Every time it changes, a complete snapshot is recorded, so any earlier state can be restored exactly and any two versions can be compared directly. A second copy is held separately from the working files. Both are checked automatically after every update, and the process refuses to record an unusually large loss of accounts without a person confirming it first. This also underpins a research requirement: published figures can point at one fixed, identifiable version of the collection, so any number in a paper can be traced back to the exact data behind it.
A second batch of hand-picked YouTube awakening stories was added to the collection, bringing the YouTube Mixed archive to 76 episodes and the total corpus to 759 accounts. All new episodes passed the full five-pass tagging pipeline and dual-model cross-validation before inclusion.
Word counts, percentages, and aggregate numbers across the site (insights page, sources page, integration challenges, physical phenomena, NDE) were previously hardcoded and going stale. They are now all driven from the live corpus data at page load, the same way episode counts have always worked. An automated check now blocks any future hardcoded statistic from reaching the site.
Round Trip Death (Eric Bennett's NDE interview show, one guest per episode) and Anthony Chene Production (long-form documentary NDE interviews) were added. Both archives are NDE-focused and complement the broader spiritual awakening accounts already in the collection. All episodes passed the standard five-pass tagging pipeline and dual-model cross-validation before being included.
A source mislabeling bug caused 163 episodes from four sources, Simply Always Awake, YouTube Mixed, Guru Viking Keep, and Brent Spirit, to receive incorrect source context when sent to the AI tagging model. This meant the model was calibrating its analysis against the wrong interview format and tradition. All 163 episodes were retagged from scratch and put back through full GPT-4o cross-validation and adjudication before being restored to the corpus.
19 episodes from Brent Spirit's "Kundalini Conversation with…" series, each featuring an individual describing their own kundalini awakening experience in depth. The series is tightly focused on first-hand accounts and high quality throughout.
After importing 37 episodes from the Guru Viking channel, a manual quality review found that 20 did not meet the standard for first-hand individual accounts, they were too short, primarily educational, or not sufficiently centered on the guest's own experience. Those 20 were removed. The 17 that remained cleared full tagging and cross-validation.
63 episodes from Angelo DiLullo's "Awakening Stories" playlist on YouTube. Each episode features an individual recounting their own direct awakening experience. DiLullo's interviews are patient, unhurried, and focused on the experiencer's account rather than commentary or teaching.
Every episode is now independently tagged by two AI models, Claude and GPT-4o, with no shared context between the two passes. Where the models agree, the tag is confirmed. Where they disagree, a third adjudication pass reviews both tag sets alongside the original transcript and makes a final determination with reasoning. This substantially raises the evidentiary standard: no tag in the corpus rests on a single model's judgment.
Shaman Oaks is a dedicated NDE interview archive featuring long-form conversations with near-death experiencers. Its addition expanded the NDE sub-corpus significantly and introduced the first archive specifically focused on this category of experience.
Public launch with the initial corpus spanning three interview archives: Buddha at the Gas Pump, Conscious TV, and a hand-curated set of YouTube interviews. All accounts had been through at least one full tagging pass at launch. The site has been expanded and refined continuously since.