Data Methodology

The platform aggregates incidents from a wide array of publicly available news media. Articles are fetched from multiple global news APIs, including the ADL Incident Tracker and various regional reporting bureaus.

• Semantic deduplication: Beyond exact URL/title matching, a MinHash/LSH similarity engine merges paraphrased coverage of the same real-world event across outlets. Pairs scoring ≥0.75 on a composite of text similarity, shared named entities, date proximity (±2 days), and geographic proximity merge automatically; 0.50–0.75 pairs go to an analyst review queue. One event, N attached sources.
• Corroboration: Every confirmed cross-outlet merge counts as independent corroboration — it raises the record's credibility grade and feeds each outlet's reliability track record. Outlets sharing a parent organization or carrying wire copy do not count as independent.
• Stable identifiers: Every published incident carries a permanent AIT-YYYY-NNNNN id. Records are never silently rewritten — post-publication changes stream through the public corrections feed.
• Language Detection: Non-English articles are detected and tagged for translation.
• Source Integrity: Each incident maintains direct links back to every reporting outlet, plus a SHA-256 evidence-vault snapshot of the primary source.

Incidents are categorized algorithmically based on NLP analysis of the article text. The following categories are used:

Every record is graded by a deterministic engine that encodes professional analytic tradecraft — no machine-learning APIs, every judgment reproducible and auditable.

• NATO Admiralty System: source reliability (A–F) and item credibility (1–6) are assessed by separate functions sharing no inputs. Reliability comes from each outlet's evolving track record (40% structural baseline, 60% empirical accuracy from corroboration and detected retractions — published in full at /api/v1/sources). Credibility 1 (“Confirmed”) requires at least two independent outlets.
• ICD-203 estimative language:probabilities are stored as numbers and rendered with the IC standard vocabulary (“likely (55–80%)”), and every confidence call ships with its drivers (“because”) and what evidence would change it.
• Negation-aware classification:signals are matched at word boundaries with NegEx-style scoping, so “no weapons were found” or “debunked as a hoax” suppress rather than trigger severity. Every fired rule records the exact text span — the evidence ledger shown in the analyst workbench.
• Triple taxonomy coding:each incident is coded to CST's six-tier scale (UK), ADL audit categories, and FBI NIBRS offense + bias-motivation codes (Anti-Jewish = 12) so the dataset reconciles against any published official statistic.
• Competing hypotheses: reports grading doubtful are run through an ACH matrix (hate incident / non-bias crime / hoax / misattribution) ranked by weighted inconsistency — disconfirmation, not confirmation, drives the call.
• Human-on-the-loop with audit: analysts can override any machine grade, but only with a stated rationale; overrides and all post-publication changes are written to a SHA-256 hash-chained audit log and surfaced through the corrections feed.

The Policy Impact Analysis measures the observable shift in incident rates following the enactment of a specific piece of legislation.

All incident data is available via a free, public REST API. No authentication required. CORS is enabled for cross-origin requests from any domain.

Institutional Data Egress

Data Dictionary — /api/v1/export (full preset)

Code Examples

# Get all incidents as JSON
curl "https://antisemitic-incidents-tracker.onrender.com/api/v1/incidents"

# Filter by location and date range
curl "https://antisemitic-incidents-tracker.onrender.com/api/v1/incidents?location=new+york&since=2025-01-01&limit=100"

# Download as CSV
curl -o incidents.csv "https://antisemitic-incidents-tracker.onrender.com/api/v1/incidents?format=csv&limit=500"

import requests

response = requests.get(
    "https://antisemitic-incidents-tracker.onrender.com/api/v1/incidents",
    params={"category": "Physical Assault", "limit": 100}
)
data = response.json()
print(f"Total: {data['meta']['total']} incidents")
for incident in data["data"]:
    print(f"  {incident['date']} - {incident['title']}")

const res = await fetch(
  "https://antisemitic-incidents-tracker.onrender.com/api/v1/incidents?location=california&sort=severity"
);
const { meta, data } = await res.json();
console.log(`Page ${meta.page} of ${meta.totalPages}`);
data.forEach(inc => console.log(inc.title, inc.severity_label));

Example JSON Response

{
  "meta": {
    "total": 487,
    "page": 1,
    "pageSize": 50,
    "totalPages": 10,
    "lastUpdated": "2026-04-05T15:00:00.000Z"
  },
  "data": [
    {
      "id": "AIT-2026-00421",
      "internal_id": "abc123def456",
      "title": "Swastika found at university campus",
      "description": "...",
      "date": "2026-03-15",
      "location": "New York, NY, USA",
      "lat": 40.71, "lng": -74.01,
      "category": "Vandalism",
      "severity_score": 6,
      "severity_label": "MEDIUM",
      "threat_priority": "MEDIUM",
      "admiralty": "B2",
      "corroborating_outlets": 3,
      "source": "JTA",
      "url": "https://...",
      "status": "verified"
    }
  ]
}