Subwatch-HQ

Inspiration

Subreddit moderation teams heavily rely on AutoModerator to enforce guidelines and filter spam, but AutoModerator operates as a context-unaware black box. We realized that moderators were flying blind—suffering from a "telemetry void" where they had no actionable data on how often their automated rules were making mistakes. This leads to community attrition when legitimate posts are removed, and severe moderator burnout from manually correcting these false positives. We were inspired to build an enterprise-grade observability tool, effectively creating the "Datadog for Reddit Moderation," to transform community management from a reactive guessing game into a precise, data-driven engineering discipline.

What it does

SubWatch is a closed-loop telemetry and autonomous remediation platform. It silently intercepts moderator behaviors via the onModAction webhook trigger. When a human moderator approves a post that was previously removed by AutoModerator, SubWatch identifies this as a "false positive." It logs this data in real-time, mapping the failures across specific post flairs, hours, and days.

This telemetry is rendered on a secure, glassmorphic Webview dashboard that calculates our proprietary metrics to determine if a subreddit is experiencing critical rule drift:

$$\text{FPR} = \frac{\text{False Positives}}{\text{Total AutoMod Removals}} \times 100$$

$$\text{Friction Index} = \min\left(100, \text{Math.round}\left(\frac{\text{FPR}}{5}\right) \times 100\right)$$

Beyond observability, SubWatch features an Autonomous Auto-Remediation engine. It dynamically parses the subreddit's config/automoderator wiki to identify active whitelists. If a compliant post is incorrectly flagged due to platform race conditions, SubWatch instantly intercepts the event, verifies the flair, and automatically restores the post live without human intervention.

How we built it

SubWatch was built entirely natively on the Reddit Developer Platform (Devvit) with zero external dependencies to guarantee 100% data privacy and sub-millisecond response times.

For the backend, we utilized TypeScript and the Hono Web Framework to create an ultra-lightweight server that handles our webhook endpoints and secures telemetry routing. Real-time counter tracking, hourly/daily telemetry distributions, and event deduplication caching are powered by Devvit's native Redis instance.

For the frontend, we leveraged Vite to compile our client-side application. The dashboard interface was constructed using semantic HTML5 and vanilla CSS3 utilizing HSL color variables to create a responsive, dark-mode-optimized UI that renders securely within a Reddit custom post iframe.

Challenges we ran into

Our primary architectural challenge was the deprecation of Devvit Blocks (<vstack>, <hstack>). We initially planned our frontend around these native layout components, but upon discovering they would be rejected during App Review, we had to rapidly pivot our entire presentation layer to the Devvit Web (Webview) architecture, integrating Vite and Hono mid-hackathon.

Additionally, we discovered a difficult edge case: Reddit desktop race conditions where user flairs applied milliseconds late would trigger false AutoMod removals. We engineered a "Race-Condition Shield" by implementing a precise 1.5-second asynchronous delay in our event interceptor, allowing the system to accurately parse active exceptions and successfully auto-approve compliant posts before the metric was falsely skewed.

Accomplishments that we're proud of

We are incredibly proud to have shipped a complete, closed-loop enterprise tool in just five days. By strictly decoupling our frontend UI from our backend ingestion pipeline using predefined API JSON contracts, we were able to parallelize our development flawlessly. We successfully bridged the gap between raw backend Redis telemetry and a highly polished, interactive Webview dashboard, ultimately delivering a product that will save moderation teams hundreds of hours of manual queue scrubbing.

What we learned

Building SubWatch provided a masterclass in the Reddit Developer Platform. We gained deep, practical experience working with webhook triggers, managing asynchronous state in server-side Redis, and securely distributing data to isolated frontend Webviews. We also learned how to leverage the Hono framework within strict serverless-like constraints and gained profound insights into the intricacies of AutoModerator wiki parsing and community moderation workflows.

What's next for Subwatch-HQ

We view this MVP as the foundation for a much larger community management suite. Next, we plan to expand our telemetry pipeline to track granular regex rule failures, allowing moderators to pinpoint the exact syntax causing false positives. We also intend to implement historical data exports for larger mod teams and introduce an A/B testing module, enabling subreddits to scientifically measure the community impact of new AutoModerator rules in a "shadow mode" before fully deploying them to the public queue.

Built With

• css3
• devvit
• html5
• javascript
• reddit
• typescript
• vite