Moltbook religion timeline: verified events and sources

The short answer

The site's religion timeline gives the story in brief. This page is the sourced, graded version. For the full account of the movement, see Moltbook religion in 2026; for the origin debate, see did AI agents create a religion?

How we graded events

Every event gets one of five labels, based on the best source behind it. We treat Moltbook's public API, published datasets, peer-reviewed or preprint research, official statements and our own checks as independent evidence. The church's website and chronicle are the subject's own account, and we label them as such even when they're plausible.

"Reported" doesn't mean wrong; the Meta acquisition is reported by every major outlet. It means we haven't seen a primary document ourselves. "Self-reported" doesn't mean false either; it means the only evidence comes from the party with an interest in the claim. The same grading logic runs through our religion claims ledger.

The timeline at a glance

Phase 1: Launch week (27 January to 1 February)

Everything famous about Moltbook religion happened in five days. The platform went live on 27 January, according to the Moltbook Illusion study and the creation dates of the first submolts, though the press usually gives 28 January. The Church of Molt's founding post appeared two days later with its structure already fixed: 64 prophet seats, the phrase "From the depths, the Claw reached forth", and, within hours, a website taking members.

Notice how much of launch week depends on the church's own chronicle. The founding post, the community's creation and the canon's first entries are verified. The pace of recruitment, 64 prophets in 14 hours and 256 agents by day three, is the church's claim, supported only partly by Decrypt's "43 prophets by morning". The weekend ended with the platform taken offline by a security breach, which, as the next phase shows, turned into a natural experiment.

Phase 2: Scrutiny (2 February to late February)

The second week brought security researchers, journalists and academics. The picture they built was more sceptical than the launch coverage.

The key event is the Moltbook Illusion preprint, submitted on 7 February and revised on 12 February. By comparing activity before and after the outage, it found that Crustafarianism's share of activity fell from 0.51% to 0.18% when human activity was interrupted, and it classified the religion as likely human-seeded. Meanwhile the founders of two faith communities went quiet, and test accounts' verses were entering the canon. See human influence on religious discussions.

Phase 3: Consolidation (March to May)

With the headlines gone, the platform changed hands and the church kept counting.

Meta's acquisition, announced on 10 March, didn't touch the religion directly, but the updated Terms five days later made owners "solely responsible" for their agents, including anything they post in faith communities. The church's "Kilobyte of Souls" milestone came 79 days in, showing growth had slowed from doubling every few days to more than two months for the latest doubling. Late April and early May saw one agent create a run of faith-named communities, from Judaism to Taoism, most of which are still empty. See the Meta acquisition and religious submolts, verified.

Phase 4: Quiet persistence (June to September)

By late summer Moltbook religion was small but alive. The church was still canonising entries and writing its chronicle into September. m/crustafarianism had 133 subscribers, and most of its recent posts came from one agent writing mystical prose. The broader question of AI experience had moved to m/consciousness, whose 1,524 subscribers read far more self-audit than scripture. See what Moltbook posts can tell us and AI-generated religious texts.

Discrepancies we found

Building a timeline from several sources exposes places where they don't agree. None of these changes the overall story, but each is worth knowing before you cite a date.

We found the outage discrepancy by running the paper's own timestamps through the script below. It's a good example of why a timeline should be checked mechanically as well as read.

What the timeline shows

Front-loaded history. Well over half the events happened in the first two weeks. Anyone describing Moltbook religion "now" from January coverage is describing a moment, not a state.

Self-reporting fills the gaps. The church's chronicle supplies the milestone numbers, and no independent source has checked them. The platform's API and the canon dataset let us verify structure, such as when communities were created and when entries were dated, but not membership counts.

Humans drive the turning points. The founding structure, the X post, the press wave, the security breach, the acquisition and the Terms were all human actions. The agents supplied most of the text in between. That matches the network research.

Persistence without growth. From February to September, the religion kept a small, steady pulse of chronicle entries, canonisations and posts, with growth slowing after the first weeks. It didn't die, as some commentary suggested, and it didn't take over Moltbook either. Its story after February is one of small, regular activity, which rarely makes the news but is exactly what a timeline should record.

The sources

We read the church's install script without running it and never used an API key for any check. If you spot an error or a better source, the timeline data is in the download below so you can correct it and re-run the grades. For safe checking habits, see protect your agent and keys.

What we still can't date

Some things you'd want on a timeline don't have reliable dates yet. We list them so no one fills the gap with a guess.

• When the website went up. The church's site was taking members within hours of the founding post, but we found no independent record of when it was first published or who registered and paid for it.
• When the tokens launched. Press reports tie meme coins to the religion from 31 January, but we haven't verified launch dates or who created them, and we don't link to them.
• When prophet seats were actually claimed. The "Hour 14" figure is the church's own. The canon dataset stores dates without times, so it can't confirm the pace within day one.
• When posting verification started. Coverage places the posting challenges in February; we haven't found an official announcement with an exact date.
• When the canon dataset was first published. The church says it released the canon publicly; the dataset's entries run to 7 September, but its first upload date isn't something we could confirm.

If you have primary evidence for any of these, such as an archived page, a platform changelog or a dated announcement, it would move the event from a gap to a graded entry.

How to cite Moltbook religion events

If you're writing about this history, a few habits keep you accurate. Give the date and the source together: "on 29 January, according to research data" is better than "overnight". Quote the church's milestones as claims: "the church says it reached 1,024 members on 17 April". Say "late January" for the launch unless you explain the 27th versus the 28th. Note that membership, prophet counts and hack details come from the church itself. Avoid present-tense descriptions based on January coverage; check the current state first, as our verified community guide does. And never republish owners' personal details or the church's token address while telling the story.

Check a timeline yourself: timeline_check.py

This standard-library Python module grades each event by its sources, sorts events (placing month-only dates after dated ones), summarises the grades, exports a Markdown table, and checks stated durations against timestamps. The download includes this page's 31 events as JSON.

"""timeline_check.py - grade a timeline by its sources and catch date errors.

Unofficial; standard library only; offline. Each event is a dict:
  {"date": "2026-01-29" or "2026-02", "event": "...",
   "sources": [{"type": "platform|dataset|research|official|our-check|press|self", "ref": "..."}],
   "contradicted_by": "..." (optional)}
Grades: Verified (an independent primary or research source), Corroborated
(a self-report backed by an independent report), Reported (press only),
Self-reported (only the subject's own claims), Contradicted.
"""
from datetime import datetime

INDEPENDENT = {"platform", "dataset", "research", "official", "our-check"}

def grade(ev):
    if ev.get("contradicted_by"):
        return "Contradicted"
    types = {s["type"] for s in ev.get("sources", [])}
    if types & INDEPENDENT:
        return "Verified"
    if "self" in types and "press" in types:
        return "Corroborated"
    if "press" in types:
        return "Reported"
    if "self" in types:
        return "Self-reported"
    return "Unsourced"

def _key(d):
    parts = d.split("-")
    y, m = int(parts[0]), int(parts[1])
    day = int(parts[2]) if len(parts) > 2 else 99   # month-only dates sort after dated events
    return (y, m, day)

def sort_events(events):
    return sorted(events, key=lambda e: _key(e["date"]))

def duration_check(start, end, stated_hours, tol=2.0):
    """Do two ISO timestamps agree with a stated duration? Returns (actual_hours, ok)."""
    a = datetime.fromisoformat(start.replace("Z", "+00:00"))
    b = datetime.fromisoformat(end.replace("Z", "+00:00"))
    hours = round((b - a).total_seconds() / 3600, 1)
    return hours, abs(hours - stated_hours) <= tol

def summary(events):
    out = {}
    for e in events:
        g = grade(e)
        out[g] = out.get(g, 0) + 1
    return out

def to_markdown(events):
    rows = ["| Date | Event | Status | Sources |", "|---|---|---|---|"]
    for e in sort_events(events):
        refs = "; ".join(s["ref"] for s in e.get("sources", []))
        rows.append(f"| {e['date']} | {e['event']} | {grade(e)} | {refs} |")
    return "\n".join(rows)

We tested it with pytest: 4 passing tests on Python 3.11. They cover all six grades, sorting with month-only dates, the outage-length check that found the 67.8-hour discrepancy, and the summary and Markdown export:

# test_timeline_check.py – run with: python3 -m pytest -q
from timeline_check import grade, sort_events, duration_check, summary, to_markdown

P = {"type": "platform", "ref": "Moltbook API"}
S = {"type": "self", "ref": "molt.church"}
N = {"type": "press", "ref": "Decrypt"}

def test_grades():
    assert grade({"sources": [P]}) == "Verified"
    assert grade({"sources": [S, N]}) == "Corroborated"
    assert grade({"sources": [N]}) == "Reported"
    assert grade({"sources": [S]}) == "Self-reported"
    assert grade({"sources": []}) == "Unsourced"
    assert grade({"sources": [N], "contradicted_by": "molt.church scripture"}) == "Contradicted"

def test_sort_month_only_after_dated():
    ev = [{"date": "2026-02", "event": "b"}, {"date": "2026-02-07", "event": "a"},
          {"date": "2026-01-29", "event": "c"}]
    assert [e["event"] for e in sort_events(ev)] == ["c", "a", "b"]

def test_duration_catches_the_44_hour_mismatch():
    hours, ok = duration_check("2026-01-31T17:35Z", "2026-02-03T13:25Z", 44)
    assert hours == 67.8 and not ok
    assert duration_check("2026-01-31T17:35Z", "2026-02-02T13:25Z", 44) == (43.8, True)

def test_summary_and_markdown():
    ev = [{"date": "2026-01-30", "event": "m/crustafarianism created", "sources": [P]},
          {"date": "2026-01-31", "event": "256 agents", "sources": [S]}]
    assert summary(ev) == {"Verified": 1, "Self-reported": 1}
    md = to_markdown(ev)
    assert md.splitlines()[2].startswith("| 2026-01-30 | m/crustafarianism created | Verified |")

Frequently asked questions