CampaignLens

• 

• 

• 

• 

Inspiration

Moderators often see harmful or promotional campaigns only after the pattern has already spread. A single post can look harmless: one recommendation, one repeated phrase, one unusual link. The real signal is distributed across threads, timing, wording, reports, and link obfuscation.

CampaignLens was inspired by that gap. I wanted to build a Devvit-native mod tool that helps moderators see the pattern without turning the app into an automatic enforcement system or a user-profiling tool.

The goal was simple: make coordinated behavior easier to investigate, while keeping moderators in control.

What it does

CampaignLens detects campaign-shaped activity inside a subreddit and turns it into explainable evidence dossiers.

It watches Reddit triggers for posts, comments, reports, deletions, Automod filters, and moderator actions. It extracts signals such as repeated domains, brand-like phrases, configured watchlist terms, obfuscated links, near-duplicate phrasing, thread spread, timing bursts, and reports.

When enough signals line up, CampaignLens creates a dossier showing:

• total score and score breakdown
• why the pattern surfaced
• evidence excerpts and matched fragments
• timeline
• campaign replay
• category such as commercial promotion or possible harmful narrative
• safe moderation action preview
• item-level action history

CampaignLens does not automatically ban, remove, lock, accuse, profile users, or send content to external AI services.

How I built it

CampaignLens is built with Devvit Web, React, TypeScript, Hono, Redis, and Reddit’s Developer Platform APIs.

The app has three major layers:

1. Devvit integration layer
   All Reddit, Redis, realtime, trigger, and context access is isolated behind src/devvit/*. This keeps the rest of the app protected from SDK changes.

2. Detection and scoring layer
   The backend extracts compact evidence samples and scores them with deterministic signals:

   • domain burst
   • brand burst
   • thread spread
   • near-duplicate similarity
   • participation/timing pattern
   • obfuscation
   • reports
   • moderator-configured watch terms

3. Moderator workflow layer
   The dashboard shows dossiers, review queue, settings, diagnostics, action preview, confirmation, and action history.

For Reddit compliance, the inline custom post is only a compact launch view. The full dashboard opens in user-initiated Expanded Mode so the inline view does not trap Reddit scrolling gestures.

Challenges I faced

The biggest challenge was making a real Devvit app behave reliably inside Reddit, not just building a local dashboard.

Some specific challenges:

• Devvit SDK instability
  I hit opaque SubmitCustomPost runtime errors in older Devvit builds. I solved this by pinning a tested Devvit version and isolating SDK calls behind an adapter layer.

• Inline scroll trap review feedback
  Reddit review flagged the first dashboard version because the inline webview could scroll inside a Reddit post. I redesigned the flow so the inline post is only a launch card and the full app opens in Expanded Mode.

• Fresh subreddit behavior
  New subreddits start with no baseline and no dossiers. I had to make sure this felt like a normal empty state instead of an error state.

• Moderator safety
  Because CampaignLens can preview and execute moderation actions, I added explicit confirmation, item-level results, rollback warnings, subreddit scope checks, disabled ban/mute actions, and no automatic enforcement.

• Hackathon demo friction
  A judge should not need to create ten posts and guess the right thresholds. I tuned the defaults so a safe three-post test can demonstrate the core loop clearly.

What I learned

I learned that mod tools need a different product mindset than normal dashboards. The UI must be fast and clear, but also careful. A moderation tool should avoid overclaiming. “This pattern is worth review” is safer and more useful than “this user is bad.”

I also learned that Devvit Web is powerful, but production readiness depends on details: permissions, view modes, realtime fallback, Redis failure handling, app review expectations, and clean install instructions.

The most important technical lesson was to keep the Devvit SDK behind a boundary. That made it much easier to respond to API/version changes without rewriting the whole app.

What makes CampaignLens different

CampaignLens is not another AI moderation classifier. It is an evidence-first pattern detection tool.

The app does not ask moderators to trust a black box. It shows the signals, the evidence, the timeline, and the reason a dossier surfaced. Moderators decide what to do.

That matters because coordinated campaigns are often not obvious at the single-post level. CampaignLens gives moderators a broader view while preserving human judgment.

What’s next

The next version would focus on:

• better onboarding for first-time moderators
• richer replay visualization
• exportable diagnostics with content redaction
• reviewer assignment workflows
• more frontend interaction tests
• optional semantic harmful-narrative classification, only with clear safety review and moderator control

CampaignLens v1.0.1 is focused on the complete hackathon-ready loop: detect, explain, review, preview, confirm, execute, and audit.

Built With

• devvit
• html5
• typescript