Methodology

Standards before scale.

Campus Evidence Lab tracks public-source records related to campus civil-rights matters, identity-based conflict, public enforcement activity, and institutional responses. The dataset is documentation infrastructure, not a verdict system.

Review Artifacts

• Public codebook
• Coverage limits
• Methodology stress-test memo
• Methodology examples JSON

Current Scope

The current dataset covers U.S. campus civil-rights records across religion, race, ethnicity, national origin, sex, pregnancy, disability, and athletic-equity categories. Community labels are assigned only when the public source supports them, and they describe what is documented in the source record rather than an independent project finding.

Event Record Unit

The event record is the basic unit of publication. Each event record should describe one public-source-supported matter or source-backed update with enough structured fields for another reviewer to inspect the school, source, date precision, category, affected-community label, response-depth status, confidence, and record hash.

Import-Wave QA Candidate Unit

Accepted import-wave QA candidates are official-source rows that passed deterministic source-family, source-locator, school-identity, duplicate, prohibited-field, and claim-limit gates. Acceptance is not individual human certification, a legal finding, a school ranking, a safety score, or a prevalence estimate.

Inclusion Criteria

• Campus-related.
• Connected to identity-based civil rights, discrimination, harassment, public policy, legal action, or institutional response.
• Supported by at least one public source.
• Written in neutral, attributed language.
• Reviewed before publication for event records; deterministic QA before publication for accepted import-wave candidates.

Minimum Record Fields

A publishable event record must identify the school, date or date precision, location, category, affected community label, summary, description, source IDs, source types, verification status, confidence label, update date, and record hash. If a field cannot be supported by public source material, the record should remain unpublished or use an explicit limited-value field instead of inference.

Accepted Source Types

• Campus newspaper reporting.
• University statements and public policy notices.
• Public safety notices.
• Public legal filings.
• OCR complaints, resolutions, and government releases.
• Local or national journalism.
• Nonprofit reports when methodology and source basis are clear.

Source Hierarchy

Official records and public datasets receive the strongest evidentiary weight. Reliable journalism and campus newspaper reporting can support inclusion when they identify the school, date, event type, and source basis. Advocacy or nonprofit reports require clear methodology and public source support before they can be used.

Source Reliability Protocol

The Source Reliability Protocol asks reviewers to identify publisher type, direct URL, publication date, public accessibility, source-to-record fit, locator specificity, and whether the source itself distinguishes allegations, findings, institutional statements, and agency-described actions.

Deduplication Logic

Records should not be duplicated because the same public matter appears in multiple sources. A second source should usually be attached to the existing record unless it documents a distinct date, public action, legal step, institutional response, or source-backed update. Structured public datasets may use scoped deduplication keys such as school, year, geography scope, offense, and bias-code so aggregate cells remain auditable without being merged incorrectly.

Event Category Definitions

• Harassment or threat: targeted harassment, threats, intimidation, or public allegations of identity-based hostile treatment.
• Vandalism: property damage or defacement tied to identity-based hostility.
• Discrimination allegation: public complaints, lawsuits, OCR matters, or institutional records alleging unequal treatment.
• Protest-related incident: campus protest events where public sources identify civil-rights or identity-based conflict.
• Institutional response: public actions by a school, agency, court, or accreditor responding to a civil-rights matter.
• Policy change: public changes to rules, enforcement processes, access, training, or institutional obligations.
• Public safety notice: campus or law-enforcement notices relevant to the dataset scope.
• OCR complaint: public OCR complaints, investigations, letters, resolutions, or monitoring actions.
• Lawsuit or legal filing: public court filings or legal actions.
• Criminal investigation: public-source records of criminal investigation or prosecution.
• Community response: public-source responses by student, faculty, alumni, nonprofit, or civil-rights organizations.

Affected Community Definitions

Affected communities are assigned only when supported by public source material. The field identifies the community named in the source record; it does not imply that Campus Evidence Lab independently determined motive, legal liability, or institutional fault. Multiple communities may be listed when a public source identifies more than one affected group.

The known taxonomy limitation is that broad source language does not always support narrow community labels. When the source is broad or ambiguous, the record should preserve that uncertainty rather than forcing a narrower label.

Exclusion Criteria

• Private testimony.
• Private screenshots or direct messages.
• Unverified social media-only claims.
• Records without public source links.
• Private personal information.
• Legal conclusions not present in source material.

Verification Status

Verification describes the kind of public support behind a record. It does not determine legal truth and does not measure severity.

• Verified from public source.
• Verified from multiple public sources.
• Public allegation.
• Institutional statement only.
• Updated after correction.

Confidence

Confidence describes source support, not moral severity.

• High: official documentation or multiple reliable public sources support the record.
• Medium: one reliable public source supports the record, or public sources leave some details incomplete.
• Low: a public source exists, but important details remain limited or disputed.

Review Workflow

• A source is discovered or submitted.
• A draft event record is created from public information only.
• A reviewer checks the source, school, date, affected community, category, and attribution language.
• The reviewer assigns verification status and confidence based on source support.
• The record is published only after review.
• Record hashes and snapshot hashes are regenerated after approved changes.

Correction Process

Corrections must identify the record ID, the disputed field, and a public source supporting the change. Accepted corrections update the changelog and produce a new record hash. Rejected corrections should preserve a short rationale in review notes before the project has a formal public review queue.

Submit Public Sources through the structured intake page when a correction, source addition, duplicate report, or school metadata update can be supported by a public URL.

Reviewer Standard

A reviewer should be able to reproduce the record from public material without private context. Review checks include source availability, source type, date precision, school identity, category choice, affected community label, legal-status wording, privacy risk, duplicate risk, neutral language, and whether confidence describes source support instead of severity.

No Ranking System

Campus Evidence Lab does not rank schools by hate, safety, severity, or moral standing. Record counts reflect public-source availability, source discovery, reporting practices, school size, jurisdiction, institutional transparency, and reviewer capacity. If future coverage signals describe documentation density or source diversity, they should remain documentation signals rather than comparative school judgments.

Privacy Limits

The MVP does not collect private testimony or sensitive evidence. Public records should avoid naming private individuals unless a name is necessary to understand the public record and already appears in source material from a reliable public source.

AI Use

AI may assist with extraction, summarization, duplicate detection, and brief drafting. AI does not publish event records, determine legal meaning, or assign final event classification without human review. Import-wave candidates can be accepted through deterministic QA gates when the source family, fields, locators, and claim limits meet the bulk-publication standard.

Integrity

Each event receives a deterministic record hash. Each dataset snapshot receives a snapshot hash. Hashes are used to make silent retroactive edits easier to detect.

The current manifest is published at data/snapshot-manifest.json. The current archived manifest is linked from the data downloads page. Weekly briefs reference the event dataset hash used for their record set.

Limitations

The dataset is not a complete census of campus civil-rights incidents. It reflects public-source availability, reviewer capacity, source discoverability, and the current scope of the MVP. Records should not be used as rankings, severity scores, prevalence measures, or proof of legal liability.

Versioning and Audit Policy

Public changes regenerate record hashes, dataset hashes, snapshot manifests, the snapshot index, CSV exports, source audit metadata, and the public changelog. Archived snapshot manifests are listed in data/snapshot-index.json. Before launch or major publication cycles, maintainers should run the advisory live source audit and manually review any failures.

Update Cadence

Update Cadence is snapshot-based rather than continuous. Records can be added, corrected, reclassified, or removed through public review, then published with regenerated hashes, release notes, and snapshot artifacts.