Skip to content

Review Interface

The Review interface is where you manually approve or reject posts for export to Curated Queue.

Accessing Review

  1. Click Review in the main navigation
  2. You'll see a list of posts awaiting review
  3. Each post shows spam score, content preview, and action buttons

Understanding the Interface

Post Display

Each post shows: - Spam Score (0-100): FenLiu's automated heuristic assessment - 0-25: Low spam confidence - 25-75: Medium (requires judgment) - 75-100: High spam confidence - ML Confidence (0.0–1.0): Machine learning model's approval probability — see ML Confidence Scores for full details - Content Preview: First 280 characters of post - Author: Fediverse account that posted it - Hashtags: Tags in the post - Media: Indicator if post has attachments - Stream: Which hashtag stream sourced this post

Action Buttons

  • ✓ Approve: Post is quality content, export to Curated Queue
  • ✗ Reject: Post is spam or unsuitable, don't export
  • Score Slider: Adjust spam score (0-100) before approving/rejecting

A ↑ Back to top link appears at the bottom of the page for quick navigation back to the top after reviewing a long list of posts.

Reviewing Posts

Basic Review

  1. Read the post content and preview
  2. Look at the spam score (guide only, not definitive)
  3. Click Approve ✓ or Reject ✗
  4. Move to next post

Adjust Spam Score

If you disagree with the automatic score:

  1. Use the Score Slider (0-100)
  2. Adjust to match your judgment
  3. Click Approve or Reject
  4. Your adjusted score is recorded

Spam Score Guide

  • 0-25 (Green): Legitimate content, safe to approve
  • 25-50 (Yellow): Questionable, review carefully
  • 50-75 (Orange): Likely spam or low quality
  • 75-100 (Red): Almost certainly spam

Filtering Posts

Use the filter bar above the queue to narrow what you see.

By Stream

Select a specific hashtag stream to review posts from that source only. Useful for batch-reviewing by topic or when one stream needs urgent attention.

By Spam Score

Set a min and/or max spam score to focus on a score range: - Min 0, Max 30 — content the heuristic thinks is clean - Min 50, Max 100 — content the heuristic flags as probable spam - Min 25, Max 50 — the uncertain middle band

By ML Confidence

Choose a preset from the ML Confidence dropdown to focus on what the model thinks: - Likely Approve (≥ 0.7) — posts the model is confident are good - Unsure (0.4–0.6) — the uncertain zone where your decision matters most - Likely Reject (≤ 0.3) — posts the model is confident are not what you want

Posts with no ML score (—) always pass through the confidence filter. The spam score and ML confidence filters work independently and can be combined.

See ML Confidence Scores for more filter examples.

Review Workflow

Quick Approve/Reject

For obvious posts (clear quality or clear spam): 1. Glance at content 2. Click Approve or Reject 3. Move on

Careful Review

For questionable posts (spam score 25-75): 1. Read full content 2. Check author account 3. Review hashtags 4. Adjust score if needed 5. Make decision 6. Click Approve or Reject

Batch Review

Review many posts efficiently: 1. Filter by spam score range (e.g., 0-30) 2. Approve/reject in rapid succession 3. Switch to next score range 4. Continue until all reviewed

Decisions Recorded

Every review decision is recorded: - Post approval/rejection status - Your adjusted spam score - Timestamp of review - Used for future ML model training

Export Queue Connection

When you Approve a post: 1. Post enters the queue as Pending 2. Curated Queue consumer can fetch it via API 3. Consumer processes and acknowledges 4. Post moves to Delivered status

When you Reject a post: 1. Post is marked as rejected 2. Never exported to Curated Queue 3. Remains in database for reference

Tips

Build Consistent Judgment

  • Define your own quality standards
  • Apply consistently across posts
  • Use manual score adjustment for borderline posts

Don't Over-Adjust Scores

  • Trust automated scoring for obvious cases
  • Only adjust for borderline posts
  • Your adjustments will train future ML models

Review Regularly

  • Review new posts frequently
  • Keep queue from growing too large
  • Maintain responsive export pipeline

Monitor Queue Status

  • Check Queue page to see approved posts being consumed
  • If queue backs up, increase review frequency
  • If queue empties, consider lower thresholds

Keyboard Shortcuts (Planned)

Future versions will support keyboard shortcuts:

  • A - Approve current post
  • R - Reject current post
  • ← / → - Navigate between posts
  • / - Focus search/filter

Common Issues

Too Much Spam

  • Block common spam sources in "Don't Reblog Users"
  • Disable low-quality streams
  • Consider lowering approval thresholds manually

Can't Decide on Post

  • Trust your judgment
  • If borderline, slight approve bias is ok (consumer can nack if needed)
  • Your decision will be recorded for ML training

Queue Not Consuming Posts

  • Check Queue page for errors
  • Verify Curated Queue consumer is running
  • Check API key is correct

Next Steps