Reading the Erowid Archive

Anyone who searches for what ibogaine is like arrives, sooner or later, at Erowid. It is the oldest and most careful collection of first-hand drug accounts in existence, and it is not a dataset. The difference is the whole subject of this page.

What it is

Erowid Center is a small American non-profit, granted charitable status in 2007 and collecting since 1995, based in Grass Valley, California. Its public tax filings show an organisation running, in most years, on between a quarter of a million and seven hundred thousand dollars, of which donations are usually most and sometimes nearly all. In half of the last ten filed years it spent more than it took in.

That scale matters when weighing what follows. This is not an institution with a research department. It is a couple of people, a few dozen volunteers, and thirty years of submissions.

How a report becomes a published report

More than a hundred thousand reports have been submitted. Around a third have been published.

The process, described by its operators under peer review, is that each submission is read by two of a few dozen trained volunteers, who rate it and pass it to a senior reviewer for a decision on publication. Published reports are then tagged with primary categories describing the type of experience, and secondary categories describing the circumstances.

Two thirds of what people write is therefore not what you are reading. What the rejection criteria are, and how consistently they are applied, is not something we could establish.

Nobody checked what was in it

This is the single most important thing to hold in mind, and it is not a criticism from outside. It was written by Erowid’s founders, as co-authors of a 2018 paper in Frontiers in Neuroscience:

it is impossible for the curators to assess the validity of the circumstances associated with the reports (e.g., the identity of the consumed substance or its dosage …).

Every account in the archive rests on the writer’s belief about what they took and how much. For a substance sold in the grey market as iboga or ibogaine of unstated purity, and where total alkaloid extract and purified hydrochloride behave differently at the same stated dose, that gap is not small.

The shape of the crowd

The archive’s own analysts have published what its contributors look like, from a snapshot of more than 36,000 published reports.

They also name the two biases they think this produces: reviewers may favour reports about unusual substances, and contributors write about what they themselves found interesting enough to write down. Their own summary, in the same paper, is that collections of anonymous self-reported drug experiences “are by no means a representative sample and in fact are likely a highly biased sample”.

The categories, and what the counts do not mean

Published reports carry tags from a fixed set of about fifteen primary categories. Across the whole archive in that snapshot, the largest were General, First Times and Combinations. Among the outcome tags, Difficult Experiences and Glowing Experiences each covered around one report in seven, Bad Trips one in twenty, and the category for outright disasters around one in thirty. Tags overlap, so these do not sum to the archive.

The tag for reports where the writer suspected the drug was not what it had been sold as is among the smallest, at about one in a hundred. Note what it does not mean. It counts suspicion of a discrepancy, not the far larger number of people who simply never knew.

It is tempting to read those proportions as a risk profile. The same paper suggests why that fails. Trying to predict, from the drug taken, the dose, the combination, the setting and the writer’s gender, whether a report would end up tagged as glowing, as difficult, or as tipping into habituation, the authors’ models found almost none of them. Overall accuracy looked high, in the eighties and nineties, because most reports carry none of those tags and a model that predicts nothing is right most of the time. On the tags themselves the models caught around a tenth of the glowing reports, a seventh of the addiction ones, and fewer than one in two hundred of the difficult ones.

Whatever decides how an experience goes, it is not legible in what the archive records about it.

The ibogaine corpus is thin

In the snapshot behind the 2018 analysis, restricted to reports where only one substance was involved, the archive held 32 ibogaine reports. LSD had 718. Salvia divinorum had 1,267.

At thirty-two, a single vivid account is roughly three per cent of everything there is. Read on those terms, one report is a story someone told about a night they had. It is not a sample, and no proportion drawn from thirty-two reports survives contact with a denominator.

When the archive found something real

The case for taking it seriously is not hypothetical.

Researchers working with Erowid staff went through reports of people who had taken a classic psychedelic while on a mood stabiliser. Among 62 reports involving lithium, nearly half described a seizure. Among 34 involving lamotrigine, none did. That finding changed what clinicians tell patients, and no trial would ever have produced it.

Note what it is and is not. It is a signal, drawn from a set of reports selected by the fact that someone wrote them. There is no denominator, so it cannot say how often lithium and psychedelics produce a seizure. It says the combination appears in the record often enough, and starkly enough against a comparator, to take seriously. That is the correct use of this archive.

The timings that came back through the front door

There is one documented case of the archive being laundered into the literature, and it concerns ibogaine specifically.

A 2017 review in Current Neuropharmacology sets out ibogaine’s experience in three phases, with the timings that appear in almost every description since: a visionary phase beginning within one to three hours and lasting four to eight, an introspective phase from four to eight hours lasting eight to twenty, and a residual phase setting in twelve to twenty-four hours after the dose.

The entire paragraph carries one citation. That citation is an Erowid vault page.

The timings themselves are not invented, and they are not necessarily wrong. They trace back to a clinical review whose author described it as a composite built from interviews and case reports. What happened here is that a peer-reviewed pharmacology journal reproduced them without going to that literature, and cited a website instead. Anyone citing the 2017 review now inherits an anecdote-derived composite with a peer-reviewed stamp on it.

Who describes the archive

One structural fact belongs on any page assessing this source. Erowid’s own staff are co-authors on most of the peer-reviewed work that characterises the collection, including every paper cited above except the 2017 review. This is disclosed each time and it is a consequence of how data access works: the archive requires coordination with its research team.

It also means the archive’s operators have shaped how the archive is scientifically described. They have done so candidly, and the most damaging admissions on this page are theirs. But no independent group has published a study whose central subject is whether the vaults represent anything, and that gap is worth naming.

How to read one report

If you have found an account of an ibogaine experience there, four questions are worth asking before you weigh it.

1. Was the substance identified by anyone but the writer? Almost certainly not.
2. What else was taken? Combination reports are a fifth of the archive, and an interaction may be doing work the writer attributes to iboga.
3. What is the writer’s stake? Erowid pays nothing and asks nothing, which removes the incentive a clinic testimonial carries. It does not remove the wish to have had a meaningful experience.
4. Is this a story or a rate? With thirty-two reports, always the first.

The account is still worth reading. It is one of the few places where the experience is described by someone with nothing to sell. Reading the Reddit communities covers the other, which has different problems, and when it does not work covers what the follow-up studies found once people had gone home.