The present record has important limitations. They are listed here so that the quickest route to contradicting this work is also the best documented.
1
Independent laboratory replicationThe analyses have been repeated across computational systems and tested on held-out data, but an external group has not yet reproduced the complete result from acquisition through analysis. That is the next required evidentiary step.
2
Operator dependenceBoth instruments contain classification decisions. Cross-model agreement and explicit rubrics reduce arbitrariness, but they do not eliminate the need for human replication, inter-rater reliability, and sensitivity analysis. The paper therefore treats the blind human-coder study as central rather than confirmatory.
3
Corpus provenanceTranscendental narratives are not laboratory emissions. They are shaped by memory, language, selection, publication, interviewer practice, and cultural vocabulary. A strong result must show that the measured structure is not an artifact of those channels.
4
The comparison setFifty positions can establish a representative scale, but they cannot exhaust Kuhn's catalogue of more than four hundred. Different exemplars or recently published tests may alter individual standings. This is not a defect in the standard. It is a reason to publish the scoring rules and invite independent rescoring.
5
Theory of consciousness vs theory of evidenceThe instruments establish structure and content in the narratives. The inferential step from lawful narrative organization to the ontology that best explains it remains open to competing accounts. The strongest empirical claim is that the domain is measurable and returns non-random, prospectively testable structure.
What this does not claim
Named here so that the overreaches are on the record before anyone else makes them.
It does not show that consciousness survives death. The instrument measures the order in which experiences are narrated. It makes no claim about the veracity, source, or metaphysical status of what was reported, and the theory on this board cites no veridical case as support.
It does not show that the brain is not the source of consciousness. Ruling out an unstructured confabulation account is not ruling out physicalist accounts generally. Structured confabulation under specific neurochemical conditions, and narrative reconstruction during recall rather than during the event, are not tested by this design.
The models did not detect anything about consciousness. They performed pattern classification. The multi-model design speaks to coding reliability and to model-level divergence. It speaks to nothing about insight, understanding, or experience on the part of any model.
Reliability does not show the theory is correct. That the coding is stable and the structure replicates does not establish which explanatory mechanism, if any, is correct.
Scope of the present result
The order instrument is specified for seven categories of transcendental experience. The results reported here cover one of them.
Cross-category invariance is a specified test of the locked protocol. If the same architecture appears in after-death communication arising from grief, in mediumship, and in shared death experiences reported by healthy bedside witnesses, then a physiological account tied to cardiac arrest does not cover the phenomenon. If it does not appear, the seven categories do not share an architecture and that is a stated failure condition.
The test has not been run. Its absence is a limitation of the present report rather than of the instrument.
Comparison with other altered states
The most direct challenge to any structural claim about near-death narratives is that comparable structure may appear in states nobody regards as veridical. The challenge is reasonable and it has a substantial literature behind it.
Comparisons between near-death experiences and pharmacologically or physiologically induced states have consistently found extensive overlap — Charland-Verville et al. (2014) reporting no significant differences on Greyson scale items, Martial et al. (2019) finding ketamine reports consistently most similar across roughly 15,000 reports, and Michael et al. (2025) identifying canonical near-death themes across thirty-six DMT interviews.
These findings are not disputed. Each measures the presence or frequency of experiential content. None measures the order in which it is reported. A finding that two experience types are phenomenologically indistinguishable is a statement about their content inventories. It carries no information about transition structure, because transition structure was not the object of measurement.
At this time no comparable sequential findings have been reported from non-transcendental altered states. Whether comparable structure exists remains open, and it is the question this result most needs answered.
Three outcomes are possible and each is informative. If comparison corpora return no comparable structure, the near-death result is distinguished from the alternatives most often proposed to explain it. If they return the same structure, the order finding is substantially weakened, and that should be known. If they return the same elements with weaker transitions and higher entropy, the near-death corpus is more constrained than known altered states.
The standard offered
The protocol, element taxonomy, coding rubric, and analysis procedure are released in full for this purpose. Any group holding a corpus of first-person accounts from an altered state — psychedelic, hypnagogic, epileptic, anaesthetic, or otherwise — can put the same question to their own material and report what it returns, including a null.
A standard is not a claim to priority. It is a procedure specific enough that another party can run it, and a result specific enough that another party can contradict it. Both are supplied.
None of these limitations removes the completed results. They define what those results mean and what must happen next. A useful standard should make incompleteness visible without treating it as emptiness. The open move is clear: independent acquisition, blind human coding, preregistered analysis, fixed nulls, external replication of both instruments, and sequential comparison against other altered states.