# Coverage and limits — Bound by Fear dataset, release v1

Generated 2026-07-30T02:10:59.513Z by `scripts/build-release.mjs` from the live corpus
(`src/lib/data/cases.js`). Every number below is computed at build time from the 1408
records in the corpus (1117 of them core, spanning 1819–2026;
the full corpus, register included, spans 1819–2026). None of these
figures are asserted by hand; if the corpus changes, this document changes with the next build.

This document exists because a dataset that documents religious-authority abuse can be
misread as a map of where abuse *occurs*, rather than a map of where it has been
*investigated, prosecuted, and reported in a form we could verify and cite*. Those are
different things. Read the counts below with that distinction in mind.

## What this release is

- `boundbyfear-core-v1.json` / `.csv` — the 1117 core records: those that
  passed the site's inclusion criterion (schema/CODEBOOK.md §5, conditions N1–N4) against a
  person or institution whose authority nexus is supported by the cited record, at a
  qualifying evidence tier. Pending matters remain allegations and are labeled as such.
- `boundbyfear-full-v1.json` — all 1408 case records (core, institutional-event,
  historical, and context-only) plus the 9-record separate historical register.
  Context-only and register records did NOT pass the inclusion criterion and are included for
  transparency, not as additional substantiated allegations.

## Country coverage

`place` is free text (typically "City, Region, Country"), not a normalized jurisdiction code,
so the country counts below are an approximation — they under-count multi-jurisdiction and
historical entries whose `place` does not end in a clean country name. Treat this table as
indicative, not authoritative.

Countries appearing in at least one record (by this approximation), most-represented first:

- USA: 310
- United Kingdom: 88
- Russia: 85
- Canada: 60
- India: 50
- Brazil: 42
- Australia: 41
- Ireland: 40
- Germany: 37
- Indonesia: 35
- South Africa: 24
- New Zealand: 21
- Mexico: 19
- Nigeria: 18
- France: 18
- Spain: 16
- Italy: 16
- Poland: 16
- Argentina: 15
- Ukraine: 15
- Netherlands: 14
- Thailand: 11
- South Korea: 11
- Uganda: 10
- Israel: 10

...and 120 more countries with at least one record.


**This list has zero entries for the large majority of the world's roughly 195 sovereign states.**
A country's absence from this list is NOT evidence that religious-authority abuse does not
happen there. It most often means: no case from that country has yet been found, verified to
this project's evidence tier (N4: court judgment, charging document, official inquiry,
regulatory finding, or verifiable journalism, labelled as what it is), and entered. Coverage
here tracks *researcher reach and source availability*, not the true distribution of abuse.

## Tradition overrepresentation

- Christian institutions & contexts (`christian`): 797 (56.6%)
- Islamic institutions and contexts (`islam`): 122 (8.7%)
- Divination and fortune-telling services (`divination-fraud`): 85 (6.0%)
- Spiritual healing, divination, and mediumship (`spiritual-healing`): 77 (5.5%)
- New religious movements (`nrm`): 69 (4.9%)
- Buddhist institutions and contexts (`buddhism`): 42 (3.0%)
- Jewish institutions and contexts (`judaism`): 37 (2.6%)
- Hindu and Hindu-derived contexts (`hinduism`): 29 (2.1%)
- Nigerian juju (`juju`): 23 (1.6%)
- Indigenous spiritual authority (`indigenous-spiritual`): 16 (1.1%)

Christian-tradition records (`tradition: "christian"`, which includes Catholic and Protestant
sub-traditions under `traditionDetailId`) make up 56.6% of all 1408 records. This
is a real skew, and it reflects **research history and source availability, not necessarily
the real-world distribution of religious-authority abuse.** Institutional Catholic abuse in
particular has been the subject of decades of official inquiries, church-commissioned audits,
and investigative journalism in English-speaking and European jurisdictions — producing far
more N4-qualifying documentation than exists for many other traditions, not necessarily
because the underlying rate of abuse is higher. Smaller or less-institutionalized traditions,
and traditions concentrated in jurisdictions with weaker official reporting infrastructure,
are structurally under-documented here regardless of actual prevalence.

## Language coverage

English-language sources dominate this corpus. As a rough proxy (source URLs on
English-market top-level domains — .com/.org/.net/.gov/.edu/.uk/.us/.ca/.au/.nz — against all
other TLDs; this is a domain heuristic, not a language classifier, and it both over- and
under-counts in known ways), approximately 63.8% of the 2324 sources
cited across case records resolve to an English-market domain.

Dedicated non-English-language sweeps (Spanish- and Portuguese-language sources; local-language
sources in the Asia-Pacific region) have found records that English-only search missed. That
pattern generalizes: it is near-certain that further non-English sweeps, in languages not yet
systematically searched, would surface additional records this corpus does not yet contain.
Absence of a record from a non-English-dominant region is at least as likely to reflect a
search-coverage gap as an absence of underlying cases.

## Evidence-tier requirement and its effect on coverage

A record only carries `evidenceStatus: "convicted"` if the underlying conviction **stood** —
was not overturned on appeal, so far as the record establishes. 1071 of 1117
core records (95.9%) currently carry that status; the remainder are charged,
civil-finding, official-finding, active-investigation, official-hearing, or
closed-without-charge, per `schema/CODEBOOK.md`.

This requirement is deliberate (see N4 in the inclusion criterion) and it has a coverage
consequence: **jurisdictions with weak, slow, or non-public court reporting are systematically
underrepresented**, independent of how much abuse actually occurs there. A conviction that
happened but was never reported in a form this project could verify and cite does not appear
here. A prosecution that is ongoing, or that concluded without a public record, does not
appear here as a conviction even if one occurred. This is a limitation of what can be verified
from outside the jurisdiction, not a claim about what happened inside it.

## Appellate verification — a disclosed backlog

A conviction that was later overturned does not meet this archive's criterion, so appellate
status is checked wherever it can be. **How far that check has got is itself published**, in the
`appellateStatus` field on every record recording a conviction.

Of the 1121 records carrying `evidenceStatus: "convicted"`:

| `appellateStatus` | Records | Share |
|---|---:|---:|
| `affirmed` — appellate court considered it; conviction stood | 205 | 18.3% |
| `modified` — conviction stood; sentence altered on appeal | 38 | 3.4% |
| `none-found` — searched; no appeal located | 183 | 16.3% |
| `pending` — appeal filed, or within the appeal window | 31 | 2.8% |
| `unchecked` — not yet verified | 664 | 59.2% |

**426 (38%) have an established appellate position** — the conviction was
affirmed, the sentence was modified while the conviction stood, or the appellate record was
searched and no appeal was found. **664 (59%) are marked `unchecked`: not yet
verified.** They are being worked by jurisdiction, largest and most tractable first.

This field exists because an empty appellate history previously meant two incompatible things —
*"we searched and found no appeal"* and *"nobody ever checked"* — and nothing in the data
distinguished them. That ambiguity is the reason to publish this table rather than a single
reassuring sentence: **an unchecked record is not a claim that no appeal exists.** It is an
open question this project has not yet closed, and the count of open questions is stated here
so that a reader can weigh it. Where a record does assert that a conviction stood, that
assertion is now attributable to a specific state rather than to silence.

One deliberate constraint: `disturbed` — a conviction quashed, vacated or reversed — is never
inferred automatically. Adverse language in a record usually describes a co-defendant, a
companion prosecution named for contrast, an extradition fight, or an appeal against sentence
rather than conviction, and a text match cannot tell those apart from a conviction that fell.
That determination is made by a person and recorded explicitly. See `schema/CODEBOOK.md` §8.

## The core claim of this document

Every gap described above points the same direction: **absence of a record is not evidence of
absence of abuse.** This corpus measures what has been investigated, prosecuted, or reported
to a verifiable standard, and what this project's researchers have so far been able to find,
read, and correctly classify. It does not measure, and was never designed to measure, the true
prevalence or geographic distribution of religious-authority abuse. Treat every count in this
document, and in the release files it accompanies, as a floor on what is documented — not a
ceiling on what occurred.

## Fields

See `schema/boundbyfear.schema.json` and `schema/CODEBOOK.md` in the source repository for
full field definitions and coding rules. The CSV release flattens `sources` to a count
(`sourceCount`) and omits nested fields (sources, proceedings, appellateHistory,
authorityRoles, coercionMechanisms, organizations) present in the JSON release — use the JSON
release for anything requiring that detail.