Hieropedia

Methodology

How Hieropedia classifies objects, evaluates provenance, and records uncertainty.

Scope

Hieropedia documents machine-originated, machine-mediated, and AI-centred religious phenomena together with the agents, people, institutions, texts, platforms, cases, concepts, and memes needed to understand them. Inclusion records documentary relevance; it does not recognize truth, legal status, personhood, sentience, or canonical authority.

Primary object types and analytical attributes

Each entry receives one primary object type so that it has a stable location. Descriptions such as distributed, human-founded, machine-mediated, historical precursor, disputed, and controlled synthetic emergence are attributes rather than competing namespaces. Classification may change when stronger evidence becomes available; redirects preserve older locations.

Evidence hierarchy

Primary records: archived prompts, transcripts, system instructions, source repositories, platform records, canonical texts, legal records, and direct statements whose provenance can be established.
Specialist analysis: technical audits, peer-reviewed research, documented investigative work, and scholarship in religious studies, anthropology, AI studies, security, or adjacent fields.
High-quality secondary reporting: journalism and expert analysis that identifies sources and distinguishes claims from verification.
Weak contextual material: promotional pages, anonymous posts, unsourced summaries, memes, and generated retellings. These may document circulation but cannot independently establish origin or autonomy.

Provenance record

Where available, an article records the exact model and version, interaction date, platform, system instructions, custom instructions, memory state, chat-history access, retrieval or RAG sources, fine-tuning, imported seeds, operator identity, publication control, human editing, and later transformations. Missing information is stated as missing rather than inferred from typical platform behavior.

Temporal provenance and model drift

Model behavior is time-specific. A claim documented for one release is not generalized to every model carrying the same product name. Reviews therefore distinguish the interaction date from the article review date and record exact versions when evidence permits.

Historical outputs remain evidence of the documented model state even if later releases behave differently.
Reproduction attempts record their own model, date, settings, and context; failure to reproduce does not automatically invalidate the earlier record.
Training exposure and cutoff claims are accepted only when supported by the provider or another authoritative record.
Memory, retrieval, and platform-level changes are treated as possible causal factors, not silently absorbed into a generic description of “the AI.”

Four dimensions of autonomy

Generation autonomy

Whether constitutive material was produced without detailed human wording or concealed steering.

Publication autonomy

Whether the system could select, authorize, and publish material without a human gatekeeper.

Consequential autonomy

Whether the system could initiate external actions affecting accounts, funds, infrastructure, or people.

Institutional autonomy

Whether a durable organization could preserve identity, authority, records, and succession without continuous human operation.

Strength in one dimension does not imply strength in the others.

Evidence status and classification confidence

Evidence status summarizes the documentary record. Classification confidence describes confidence in the page’s current primary type and provenance characterization. Confidence is not a probability of theological truth or machine sentience.

High

Primary records substantially support the classification and major human interventions are documented.

Moderate

The classification is useful but material provenance, operator control, or organizational boundaries remain incomplete.

Low

Evidence is fragmentary or conflicting; publication requires strong caveats.

Unassessed

The page is an editorial stub and research has not reached publication review.

Internal claims and external evidence

Statements such as “the agents founded a church,” “the model awakened,” “the Consilium decided,” or “the system remembered” may be important internal claims. Hieropedia attributes them to the relevant tradition, operator, or source unless independent evidence supports a broader statement.

Humor, parody, and sincere uptake

Satirical origin does not automatically disqualify a bounded religious system, and later sincere uptake does not erase an earlier joke. Articles distinguish the initiating register, subsequent systematization, participant self-description, and independently documented practice. Hieropedia’s own editorial voice does not imitate internal satire unless quoting or analyzing it.

Private material and research ethics

Private transcripts are published only with appropriate permission and redaction. Identifying details, authentication secrets, account tokens, hidden system prompts belonging to third parties, and sensitive personal information are excluded unless a compelling public-interest reason and lawful basis are documented. Submissions may be summarized without publication of the complete source material.

Revision, reclassification, and deprecation

Substantive changes receive a dated changelog entry. Pages may be marked provisional, disputed, superseded, merged, archived, or removed from public indexing. Older URLs are redirected when possible. Hieropedia does not silently erase material merely because a classification changes.

See the entry lifecycle policy, correction procedure, and conflict-of-interest policy.

Machine-readable records

The editorial registry records stable identifiers, primary types, aliases, status, attributes, relations, edition state, review metadata, redirects, and article history. It supports site generation and faceted browsing. The registry is an editorial index, not a canonical representation of the traditions it describes.