Reviewed evidence
Public totals come from approved records only. Approval means the record passed the tracker workflow, not that raw source material is publicly displayed.
Public glossary
Terms, boundaries, and source context
Definitions for reading OGBV Tracker evidence views without exposing restricted source details.
How to read tracker data
Evidence is reviewed, model-assisted, and aggregate-only
Model decision support
Classifier categories, confidence, language, location, and target-profile signals guide reviewers and threshold processing. They can be corrected before publication.
Reviewer annotations
Manual trend events and reviewer notes add context for internal work. Public trend annotations are safe summaries, not raw post excerpts.
Safe aggregates
Charts, heatmaps, and explorer previews use anonymized aggregate fields and intentionally suppress identifying source details.
Separate reporting lanes
OGBV incidents and OGBV awareness and advocacy use the same review workflow but publish to separate totals, trends, and breakdowns.
Evidence terms
Glossary
- Approved aggregate
- Reviewed evidence that can contribute to public totals after raw content, handles, URLs, account IDs, precise locations, and operational notes are suppressed.
- Candidate record
- A collected public-platform item or manual entry that still needs classification, enrichment, and reviewer checks before it can influence public views.
- Content kind
- The record's primary reporting lane. Each reviewed record is either an OGBV incident or OGBV awareness and advocacy, so it can contribute to only one set of totals.
- OGBV Awareness & Advocacy
- A reviewed post whose primary purpose is education, prevention, survivor support, counter-speech, or campaigning against OGBV. These posts are presented separately and never counted as incidents.
- Harm topic
- An optional existing OGBV category attached to an awareness or advocacy post to describe the harm it discusses. It does not turn that post into an incident.
- Public source
- A publicly accessible platform, page, post, video, or account lane that can be collected without private access or login-gated content.
- Source record
- An admin-managed collection setting such as a keyword, hashtag, account URL, public Facebook URL, or Apify actor lane.
- Scraping keyword, hashtag, or account lane
- A configured search term, tag, or public account/page URL used by collection jobs. Inactive and removed lanes are kept for audit context but excluded from active collection.
- Apify-backed lane
- A collection route that uses an Apify actor for supported public sources such as Instagram, TikTok, Facebook, YouTube, or supplemental X collection.
- Manual review item
- A reviewer or admin-entered incident, awareness post, or advocacy post that enters the same review and publishing workflow as collected public-source records.
- Trend event
- Reviewer-approved context that explains a visible spike or shift in aggregate volume over time. It is an annotation, not raw incident evidence.
- Classifier confidence
- A model score used to route records for review and threshold processing. It is decision support, not proof that an incident is verified.
- Uncertainty signal
- A public aggregate cue that helps readers see where confidence, volume, or model agreement is weaker and where interpretation should be cautious.
- Publishing threshold
- A runtime value that helps decide whether records are ignored, rejected, routed for review, or eligible for threshold-based approval.
- Language enrichment
- Automatic or reviewer-requested detection of language context used for routing and aggregate analysis. Reviewers can correct it before publication.
- Location enrichment
- Automatic or reviewer-requested location inference from safe available evidence. Public views use coarse geography and never expose exact coordinates.
- Undisclosed location
- A record whose reviewed evidence does not reveal where it happened. Location can be unknown because the post never mentioned one, the account hides it, or reviewers removed an unsafe precise location. These records still count in totals but cannot appear on the map.
- Target profile
- An aggregate audience, survivor group, or public-role grouping used for analysis without identifying individuals.
- Learning pool
- Reviewed examples nominated for classifier improvement. Admin activation controls whether examples are used in prompts or training support.
- Passkey-gated workspace
- A restricted review or admin area protected by role-based passkeys. Public glossary and tracker pages do not expose passkeys.
- Activity log
- An internal audit trail for user, source, review, taxonomy, and manual-input changes. Activity logs are not public evidence.
- Validation-only row
- A record used for operational validation or provider testing. It is excluded from public aggregate evidence.
- Platform availability
- Collection depends on public availability, provider terms, rate limits, and active admin-managed source records. Missing provider data should be read as collection context, not absence of harm.
- Facebook and social-source limits
- Facebook collection is URL/page/post based where provider access allows it. Facebook and other providers can change access, actor behavior, returned fields, or error patterns.
- Publication boundary
- Public pages show aggregate evidence only. Raw text, handles, source URLs, account IDs, passkeys, reviewer notes, and precise locations remain restricted.
- Admin-managed source records
- Keywords, hashtags, account URLs, Facebook URLs, and Apify actor lanes are managed in restricted workspaces. Removed or inactive records do not drive active collection.