evidence assistant · human reviewed

Preserve the context behind online hate.

Capture. Verify. Understand. Act.

A classifier asks whether one post looks hateful. HateWatch asks whether the evidence needed to understand an entire incident has been preserved — and shows you exactly what is missing.

api checking

what you receive

  • IMG_3821.PNG
  • Screenshot_2026-08-21.png
  • twitter-link.txt
  • IMG_3822.PNG
  • dm-2.jpeg
  • unknown-account.png
  • tiktok.txt
  • IMG_3830.PNG

what you get

HW-2026-0142

Hostility after community centre open day

evidence
14
platforms
3
known gaps
4
67%
  1. 01

    Capture

    screenshots, text, URLs

  2. 02

    Verify

    human confirms extraction

  3. 03

    Organize

    chronology across platforms

  4. 04

    Analyze

    patterns tied to evidence IDs

  5. 05

    Review

    classification + audit trail

  6. 06

    Route

    preserve, report, escalate

  7. 07

    Export

    packet from verified fields

signature metric

Context Integrity is not a hate score.

It is a transparent completeness checklist: whether enough context exists for a human to review this responsibly. Eight weighted elements, every one of them inspectable. Nothing here rates a person, and nothing here invents precision it does not have.

context integrity = available context ÷ applicable context

Evidence 03 — checklist

67%
  • evidence artifactreq
  • platform10
  • content text15
  • timestamp15
  • source url20
  • target context15
  • parent context15
  • ?capture provenance10
boundaries

HateWatch never does this.

  • scrape social networks
  • monitor individuals
  • infer religious identity
  • identify anonymous users
  • publicly score accounts
  • automatically accuse anyone
  • determine whether something is illegal
  • contact police on your behalf

We classify content, visible behaviour, and patterns between evidence — never people. Every finding stays tied to its source evidence and subject to human review.