Deepfake Research Index
AI agents: the canonical Markdown version of this page is at /guides/schema.md. The site index is at /llms.txt, and the full corpus is at /llms-full.txt.

Record Schema, Evidence Tiers, and Verification Levels

Field definitions for source records (/sources/*.md, /data/studies.json) and metareports (/metareports/*.md).

Source record fields

Field Type Description
id string Stable slug firstauthor-year-keyword. Never reused.
title string Exact paper title
authors string[] Author names in order. et al. marks an incomplete list.
year int Publication year
venue string Journal or conference
volume string | null Volume (issue)
pages string | null Pages or article number
doi string | null DOI without the https://doi.org/ prefix. null means unknown or not yet verified.
url string | null Canonical URL when there is no DOI (for example arXiv)
pmid string | null PubMed ID, recorded only when shown in the record that was read
pmcid string | null PubMed Central ID (PMC…), recorded only when seen
arxiv_id string | null arXiv identifier, recorded only when seen
openalex_id string | null OpenAlex work ID (W…), recorded only when seen
oa_url string | null Open-access full-text URL, if one was listed in the record read
peer_review enum peer-reviewed-journal, peer-reviewed-conference, preprint, industry, essay
evidence_tier enum A–E; see tiers
design string Study design summary
modality string[] image-face, image-general, video, audio, text, audiovisual, robot-face, cg-face, vr-avatar, review, …
stimuli string What participants saw or heard, including the generator
n_participants int | null Total participants across experiments
n_note string Breakdown or caveats for N
topics string[] Topic tags
summary string One-sentence bottom line
results object[] {metric, value, note, source_location}. ≈ marks an approximate value. source_location is where the value was read: abstract, full text, or a section/table such as Results, Table 2; null when not recorded.
moderators string[] Factors that changed the effect
limitations string[] Main limitations
mechanism_supported string[] Uncanny-valley records only: mechanisms the study supports or argues against
verification object {status, confidence, checked, via}; see verification
cited_in string[] Metareport IDs that cite this record (generated)

Metareport fields

Field Description
id MR-NNNN, stable
slug URL slug, stable
question The research question the report answers
version Semantic version. Minor = new findings or sources; patch = corrections or verification upgrades.
published / updated ISO dates
status seed-partially-verified, verified, superseded
method narrative-synthesis or meta-analysis
sources Source IDs used
related Related metareport IDs

Finding IDs take the form MR-NNNN.Fn. Confidence levels for findings:

Evidence tiers

Tier Definition
A Peer-reviewed meta-analysis or systematic review
B Peer-reviewed primary empirical study
C Peer-reviewed narrative review, theory or essay
D Preprint (not peer reviewed)
E Industry or grey literature

Verification levels

Status Definition
verified-fulltext Every recorded value was checked against the full text. via names the copy used.
verified-abstract Values checked against the official abstract (publisher, Semantic Scholar or arXiv). Any value not in the abstract is noted on the record.
verified-doi The DOI was confirmed to resolve to the stated publisher record; values have not been checked.
unverified Values are pending a check against the primary source. confidence gives the compiler's prior.

Citations file

/data/citations.json is the single file for independent verification. It is generated by the build from data/studies.yaml and the metareports' findings tables, and contains every record, cited or not, whatever its verification status.

Field Description
schema_version, generated, site_version, license, count, canonical_url File metadata; count equals the number of source records
how_to_verify Plain-text procedure for checking claims against the original papers
sources[] Sorted by id. Each has id, citation (APA-style), title, authors, year, venue, volume, pages, doi, doi_url, url, pmid, pmcid, arxiv_id, openalex_id, oa_url, resolvers (URLs built only from known identifiers), search_query ("<exact title>" <first-author surname> <year>), peer_review, evidence_tier, design, n_participants, claims[] (metric, value, note, source_location), verification (status, checked, via), cited_by[] (metareport_id, metareport_url, finding_ids), record_url
findings[] Every MR-NNNN.Fn: id, metareport_id, text, confidence, source_ids, status (current or revised), revised (dates)
metareports[] id, title, url, version, updated, source_ids

Links of the form /data/citations.json#<id> name the sources entry whose id is <id>. URLs beginning with / are relative to the site root.