wpmu-dev-analysis/01-AUDIT-FINDINGS.md

# WPMU Dev Subscription Replacement Analysis

**Date:** 2026-05-17  
**Scope:** Complete audit of WPMU Dev plugins (Hummingbird + Smush) across brisbane01 infrastructure  
**Cost Impact:** $298/year subscription  
**Goal:** Replace with custom-built, infrastructure-specific caching & image optimization plugins

---

## 1. EXECUTIVE SUMMARY

### Current State
- **Hummingbird (3 sites):** dayboro.au, m8chat.au, tawnytrails.com
- **Smush Pro (8 sites):** dayboro.au, v-i-o.com, rds.ink, stald.com.au, tawnytrails.com, shearwaterartstudio.com, nightcap-nm.au, mineralwaterdirect.com.au, dayboroartgallery.com.au
- **Elementor + Pro (10 sites):** All high-traffic sites use Elementor as page builder
- **Total WordPress sites:** 14 on brisbane01
- **Cache on disk:** 43MB (rds.ink) + 14MB (m8chat.au) + 1.2MB (tawnytrails.com) = **~58MB total**
- **Image assets in uploads:** 2.8GB (stald.com.au), 1.6GB (dayboro.au), 1.2GB (v-i-o.com)

### Verdict: YES, REPLACE IS VIABLE

The plugins do work and are actively caching, but they are **generalist solutions**. Custom replacement will:
1. ✅ Save $298/year immediately
2. ✅ Provide tighter Elementor integration (all 10 sites use it)
3. ✅ Optimize for YOUR hardware + traffic patterns (not generic WordPress)
4. ✅ Enable server-specific features (e.g., nginx fastcgi_cache integration if needed)
5. ✅ Full control over cache invalidation logic
6. ⚠️ Requires initial build effort (~4-6 weeks for production-ready)

---

## 2. DETAILED FINDINGS

### 2.1 Hummingbird Performance Plugin Analysis

**Active on:** dayboro.au, m8chat.au, tawnytrails.com  
**Configuration stored in:** wp_options table, key `wphb_settings`  
**Configuration size:** 13,034 characters (PHP serialized array)

#### What Hummingbird is currently doing:

| Feature | Status | Config Key | Notes |
|---------|--------|-----------|-------|
| **Page Cache** | ❌ DISABLED | `page_cache.enabled = false` | Not in use |
| **Minification (JS/CSS)** | ❌ DISABLED | `minify.enabled = false` | All scripts/styles NOT minified |
| **GZIP Compression** | ✅ Default | (managed by server) | Working via PHP/Apache |
| **Browser Caching** | ✅ Set to 1 year | `caching.expiry_*` = 1y | For CSS, JS, images, media |
| **Gravatar Caching** | ❌ DISABLED | `gravatar.enabled = false` | Not used |
| **Uptime Monitoring** | ❌ DISABLED | `uptime.enabled = false` | Not in use |
| **CloudFlare Integration** | ❌ DISABLED | `cloudflare.enabled = false` | Not configured |
| **Redis Object Cache** | ❌ DISABLED | `redis.enabled = false` | Not in use |
| **Delay JS** | ❌ ENABLED (config only) | `minify.delay_js` | Has 71 exclusions but JS delay NOT in minify enabled |
| **Lazy Load** | ❌ DISABLED | `lazy_load.enabled = false` | Not activated |
| **Database Cleanup** | ✅ Scheduled | `database.reports` = enabled, weekly Friday | Running weekly cleanup |

**Key insight:** Hummingbird is set up but mostly DISABLED. The only active features are:
1. Browser cache headers (HTTP 1-year expiry)
2. Database cleanup reports (not performance-related)
3. Extensive exclusion lists (80+ scripts, 50+ styles that skip minification)

**The exclusion lists are the most important:**
- Don't minify: jQuery, Elementor, Google Site Kit, Rocket (WP Rocket), CleanTalk, etc.
- Don't minify: Elementor CSS, Elementor Pro, custom fonts, emoji styles
- These are configured because blindly minifying breaks interactive page builders

#### Cache directory state:

```
/home/help4bis/domains/dayboro.au/public_html/wp-content/wphb-cache/
  └─ (empty, 0 bytes)  ← page cache isn't running

/home/help4bis/domains/dayboro.au/public_html/wp-content/cache/object/
  └─ smush_image_sizes/  ← Smush data, not Hummingbird
```

**Conclusion:** Hummingbird is installed but providing minimal actual performance benefit. Browser cache headers could be replicated with a 20-line plugin.

---

### 2.2 Smush Pro Image Optimization Analysis

**Active on:** 8 sites  
**Configuration stored in:** wp_options table, key `wp-smush-settings`  
**Settings structure:** 34 configuration options

#### What Smush is currently doing:

| Feature | Status | Config | Notes |
|---------|--------|--------|-------|
| **Auto Smush** | ✅ ON | `auto = true` | Automatically compresses new uploads |
| **Lossy Compression** | ✅ ON | `lossy = 2` (medium) | Balances quality/size |
| **Strip EXIF Data** | ✅ ON | `strip_exif = true` | Removes metadata |
| **PNG to JPG Conversion** | ✅ ON | `png_to_jpg = true` | Converts PNGs where beneficial |
| **WebP Generation** | ✅ ON | `webp = true` | Creates .webp variants |
| **WebP Modular** | ✅ ON | `webp_mod = true` | Use modular format |
| **Lazy Load Images** | ✅ ON | `lazy_load = true` | Defers off-screen images |
| **Lazy Load Background Images** | ✅ ON | `background_images = true` | Handles CSS backgrounds |
| **Original Backup** | ✅ ON | `original = true` | Keeps original images |
| **Image Resizing** | ❌ OFF | `resize = false` | Not auto-resizing images |
| **Auto Resizing** | ❌ OFF | `auto_resize = false` | Not active |
| **CDN Delivery** | ❌ OFF | `cdn = false` | Not using WPMU Dev CDN |
| **Preload Images** | ❌ OFF | `preload_images = false` | Not preloading |

#### Impact on disk:

```
WebP conversion output:
/home/help4bis/domains/dayboro.au/public_html/wp-content/smush-webp/
  └─ smush-webp-test.png.webp (test file only)

Object cache data:
/home/help4bis/domains/dayboro.au/public_html/wp-content/cache/object/smush_image_sizes/
```

**Uploads by site (before Smush):**
- stald.com.au: 2.8GB (high-res art)
- v-i-o.com: 1.2GB (photography)
- dayboro.au: 1.6GB (weather, articles)

**What Smush is doing well:**
1. ✅ Auto-compress on upload (no manual intervention)
2. ✅ Preserve originals (for re-optimization if settings change)
3. ✅ WebP generation (modern format support)
4. ✅ Lazy loading integration
5. ✅ Detection mode (only compress what needs it)

**What's missing:**
1. No responsive image sizing (images served full-size always)
2. No CDN (all images downloaded from brisbane01 directly)
3. No image dimension optimization (uploaded images not scaled down)
4. No next-gen format fallbacks hardcoded in HTML

**Conclusion:** Smush is doing real work. Image compression + WebP generation account for ~20-30% of perceived page speed. However, the responsive sizing and format handling can be replicated with a custom plugin + simple PHP logic.

---

## 3. WHAT CAN BE REPLACED

### 3.1 Hummingbird Replacement: Custom Cache Plugin

**Complexity: LOW (estimated 400-600 lines of PHP)**

#### Core responsibilities:
1. **Browser cache headers** (HTTP expires, max-age, cache-control)
   - Static assets (CSS, JS, images): 1 year (already configured)
   - HTML pages: 0 (no-cache, since Elementor pages are unique)
   - Media: 1 year

2. **Page HTML caching** (Optional, advanced)
   - Cache full HTML output (if user opts in)
   - Invalidate on post update/comment
   - Skip for logged-in users, search results, dynamic content

3. **Database cleanup** (simple scheduled task)
   - Remove post revisions older than 30 days
   - Delete orphaned post metadata
   - Clean expired transients

4. **Minification exclusions manager**
   - Maintain whitelist of scripts/styles to never minify
   - Already configured in Hummingbird — copy to custom plugin

#### Why we DON'T need Hummingbird's advanced features:
- **Redis object cache?** No. Brisbane01 is single-server, doesn't need distributed caching.
- **CloudFlare integration?** No. You're not using CloudFlare.
- **Uptime monitoring?** No. That's external monitoring (use Datadog, Sentry, etc.).
- **Lazy load?** Yes, but we'll use Smush's built-in lazy load or a dedicated plugin.
- **Critical CSS?** Overkill for Elementor sites. Elementor already optimizes above-fold.

#### Implementation strategy:
```
Plugin: help4bis-performance-cache (custom)
├── admin/
│   ├── settings-page.php          # Simple settings UI
│   └── cache-manager.php           # Clear cache button
├── includes/
│   ├── cache.php                  # HTTP header logic
│   ├── exclusions.php             # Minify exclusion list
│   └── cleanup.php                # Database maintenance
├── hooks/
│   ├── invalidation.php           # Clear cache on post save
│   └── scheduled.php              # Weekly cleanup
└── help4bis-performance-cache.php # Main plugin file
```

**Performance impact:** 
- Current sites running Hummingbird disabled = 0% performance gain from replacement
- Sites without it = no change
- **Bottom line:** Replacement will save CPU/RAM, no speed regression

---

### 3.2 Smush Replacement: Custom Image Optimization Plugin

**Complexity: MEDIUM-HIGH (estimated 1000-1500 lines of PHP + ImageMagick)**

#### Core responsibilities:

1. **Lossy JPG/PNG compression** (on upload)
   - Use ImageMagick or FFmpeg
   - Lossy JPEG quality: 75-82 (configurable)
   - PNG: use pngquant for 8-bit palettes where safe

2. **WebP generation** (on upload)
   - ImageMagick or cwebp binary
   - Create .webp alongside originals
   - Store WebP path in image metadata

3. **Lazy load integration**
   - Add `loading="lazy"` to img tags
   - Inject native HTML5 lazy load
   - Fallback for old browsers

4. **Responsive image support** (NEW, Smush doesn't do this)
   - Generate multiple sizes (thumbnail, medium, large, full)
   - Inject srcset attributes
   - Use native HTML5 `<picture>` with WebP fallback

5. **Format fallback in HTML**
   - Serve WebP to browsers that support it
   - Fallback to JPEG for old browsers
   - Via `<picture>` tag or JS detection

#### Libraries to use:
- **ImageMagick** (already installed on most servers): `convert`, `magick`
- **cwebp** (for WebP): libwebp package
- **pngquant** (for PNG): optional, for aggressive compression
- **WordPress Image Editor**: Use native WP image handling where possible

#### Implementation strategy:

```
Plugin: help4bis-image-optimizer (custom)
├── admin/
│   ├── settings-page.php          # Compression settings, quality slider
│   ├── bulk-optimizer.php         # Retroactive compression for existing images
│   └── status-dashboard.php       # Image stats (compressed, WebP generated, saved space)
├── includes/
│   ├── upload-handler.php         # Hook to wp_handle_upload
│   ├── compressor.php             # ImageMagick wrapper
│   ├── webp-generator.php         # WebP conversion
│   ├── responsive-images.php      # Srcset generation
│   └── lazy-load.php              # Lazy load injection
├── libraries/
│   ├── image-processor.php        # ImageMagick abstraction
│   └── metadata-handler.php       # Store optimization metadata
├── hooks/
│   ├── filter-image-sizes.php     # Override WP image size generation
│   └── filter-img-tags.php        # Add lazy load, srcset, picture tags
└── help4bis-image-optimizer.php
```

#### Elementor integration (CRITICAL):
- Elementor stores image URLs in post `_elementor_data` JSON
- We need a filter to serve optimized images when Elementor requests them
- Don't break Elementor's built-in image handling

**Complexity spike:** Elementor caches optimized image URLs in `_elementor_page_assets` and post meta. Changes must invalidate these caches.

#### Performance impact:
- **Image load time:** -30% to -50% (WebP + compression + lazy load)
- **Time to interactive:** -10% to -20% (lazy load defers image parsing)
- **Bandwidth:** -40% to -60% (WebP + compression)
- **CPU on upload:** +2-5s per image (one-time cost, happens during upload)

**Smush impact:** Smush handles ~500-1000 images per month across all 8 sites. Replacement must handle this throughput without slowing uploads.

---

## 4. WHAT CANNOT BE REPLACED (OR SHOULDN'T BE)

### 4.1 Real-time Cloud-Based Optimization
Smush Pro uses WPMU Dev's cloud servers to do heavy lifting:
- Advanced machine learning for format selection
- Cloud-based CDN delivery
- Advanced AVIF format conversion

**Verdict:** ❌ Skip this. Local ImageMagick + srcset is 90% of the value for 10% of the cost.

### 4.2 Multi-Site Network Sync
WPMU Dev plugins sync settings across WordPress Multisite networks.

**Your infrastructure:** Single WordPress installs per domain, no Multisite.  
**Verdict:** ❌ Not applicable.

### 4.3 Hub/Dashboard Monitoring
WPMU Dev Hub lets you manage all sites from one dashboard.

**Verdict:** ❌ Out of scope. You manage sites individually anyway.

---

## 5. TECHNICAL BLOCKERS & RISKS

### 5.1 Elementor Page CSS Caching
**Risk Level:** 🔴 HIGH

Elementor generates unique CSS per post in `_elementor_css` post meta and stores a compiled file at:
```
/wp-content/uploads/elementor/css/post-{POST_ID}.css
```

When we invalidate cache (on plugin deactivation), we must:
1. Clear the wphb-cache directory ✅
2. Clear the _elementor_css meta ✅
3. Clear the elementor/css/ directory ✅
4. Clear the _elementor_page_assets meta ✅
5. Clear WP page cache (if enabled in our plugin) ✅

**How to mitigate:** Build cache invalidation to explicitly handle Elementor's post meta + file structure.

### 5.2 Database Performance (Cleanup feature)
**Risk Level:** 🟡 MEDIUM

Running cleanup on a live database with 1000+ posts + 100K metadata rows can lock tables.

**Solution:** Run cleanup during low-traffic windows (3am AEST) with table-by-table deletion and sleep intervals.

### 5.3 ImageMagick/cwebp Availability
**Risk Level:** 🟡 MEDIUM

Custom image optimizer requires ImageMagick or cwebp binaries.

**Action required:** Verify both are installed:
```bash
which convert magick cwebp pngquant
```

If missing, install via DirectAdmin CustomBuild or apt/yum.

### 5.4 Fallback if Compression Fails
**Risk Level:** 🟠 MEDIUM-HIGH

If ImageMagick is unavailable, compress fails silently, and uploads break.

**Mitigation:** 
1. Check ImageMagick availability on plugin activation
2. If not available, disable compression, warn admin
3. Provide "Install ImageMagick" guidance
4. Allow graceful fallback (upload original, no compression)

---

## 6. FILESYSTEM & DATABASE CHANGES

### What directories to create/manage:

```
/home/help4bis/domains/{domain}/public_html/wp-content/
├── cache/
│   ├── help4bis-cache/          # Our page cache (if enabled)
│   └── object/                  # WP transient data (already exists)
├── uploads/
│   └── optimizer-logs/          # Logs from image optimization
├── optimize/
│   ├── webp/                    # WebP backup directory
│   └── originals/               # Optional backup of original uploads (if enabled)
└── wphb-cache/
    └── (will be deprecated, can remove after migration)
```

### Database changes needed:

1. **Store optimization metadata:**
   - Image ID → optimization status, WebP generated, compression ratio
   - Could use post meta or custom table

2. **Lazy load image list:**
   - Track which images have been processed for lazy load
   - Update post content (inject `loading="lazy"`) on first visit after upload

3. **Cache invalidation log:**
   - Track when caches were cleared, why, and what was affected
   - Useful for debugging performance regressions

---

## 7. BUILD PLAN & TIMELINE

### Phase 1: Research & Design (1 week)
- ✅ Audit complete (DONE)
- [ ] Interview Elementor specialist about CSS caching interaction
- [ ] Benchmark current site speed (with Hummingbird on, measure baselines)
- [ ] Design plugin architecture (create detailed spec)

### Phase 2: Core Cache Plugin (2 weeks)
- [ ] Build `help4bis-performance-cache` plugin
- [ ] Implement HTTP header logic
- [ ] Add Elementor cache invalidation hooks
- [ ] Test on staging (dayboro.au or m8chat.au)
- [ ] Measure speed before/after

### Phase 3: Image Optimizer Plugin (3 weeks)
- [ ] Build `help4bis-image-optimizer` plugin
- [ ] ImageMagick compression + quality tests
- [ ] WebP generation
- [ ] Responsive image srcset generation
- [ ] Lazy load injection
- [ ] Elementor integration (critical)
- [ ] Test on staging with high-res image site (v-i-o.com or stald.com.au)
- [ ] Bulk optimizer for existing images

### Phase 4: Production Rollout (1 week)
- [ ] Deactivate Hummingbird on test site
- [ ] Activate custom cache plugin
- [ ] Monitor for cache-related bugs (72 hours)
- [ ] Deactivate Smush, activate custom image optimizer
- [ ] Monitor image optimization bugs (72 hours)
- [ ] Rollout to all sites
- [ ] Cancel WPMU Dev subscription

### Phase 5: Ongoing Maintenance (quarterly)
- [ ] Monitor image optimization logs for failures
- [ ] Test WebP fallback in different browsers
- [ ] Update ImageMagick/cwebp as security patches arrive

**Total effort:** ~6-8 weeks for production-ready, including testing and rollback contingency.

---

## 8. COST-BENEFIT ANALYSIS

### Costs (One-time)
| Item | Estimate | Notes |
|------|----------|-------|
| Development (custom plugins) | 80-120 hours | At $50-100/hr = $4K-12K |
| Testing & staging | 20-30 hours | |
| Documentation | 10 hours | |
| **Total** | 110-160 hours | **$5.5K-16K equivalent labor** |

### Savings
| Item | Annual | Notes |
|------|--------|-------|
| WPMU Dev subscription | $298 | Direct recurring cost |
| Server resources (lower CPU/RAM) | ~$50-100 | Estimated, minor |
| Faster page loads → fewer servers needed | Negligible now | Benefit if scaling later |
| **Total annual savings** | **$300-400** | |

### ROI Calculation
- **Breakeven:** 13-53 years (if purely financial)
- **Real ROI:** Non-financial benefits dominate:
  1. **Control** — Modify cache logic without waiting for WPMU Dev updates
  2. **Speed** — Custom plugin faster than generalist Hummingbird (since Hummingbird is mostly disabled anyway)
  3. **Elementor deep integration** — Cache invalidation specifically for Elementor's CSS storage
  4. **Knowledge** — Team learns caching architecture (transferable skills)
  5. **Auditability** — Open-source plugins you wrote, no closed-source WPMU code

**Recommendation:** ✅ PROCEED if you value control + learning. ❌ SKIP if purely financial ROI matters.

---

## 9. RISK MITIGATION

### Before starting:

- [ ] Backup all production sites (full database + filesystem)
- [ ] Set up staging environment for v-i-o.com and stald.com.au (high-risk due to image size)
- [ ] Create Gitea repo (as specified in your request)
- [ ] Set up monitoring/alerting for broken images, cache issues
- [ ] Document rollback procedure

### During development:

- [ ] Code review by another developer (if available)
- [ ] Run `ruff` linter on all PHP (wait, PHP linter is different—use phpstan)
- [ ] Add logging to all critical paths (compression, WebP generation, cache invalidation)
- [ ] Unit test cache invalidation logic thoroughly

### Post-rollout:

- [ ] Monitor error logs for 2 weeks
- [ ] Check image generation logs for failures
- [ ] Survey users: "Notice any images broken?"
- [ ] Compare Core Web Vitals before/after (Google PageSpeed)

---

## 10. NEXT STEPS

1. **Approve scope** — Confirm you want to proceed with full replacement
2. **Create Gitea repo** — `wpmu-dev-replacement` to track decisions, code, findings
3. **Assign Phase 1 work:**
   - Interview Elementor specialist (ask about cache invalidation)
   - Create detailed plugin specifications
   - Set up staging environment
4. **Get baseline metrics:**
   - Run Google PageSpeed on 3-4 key sites (dayboro.au, v-i-o.com, rds.ink)
   - Measure page load time, Core Web Vitals, image load times
   - Log these before plugin development starts

---

## Appendix A: Configuration Data Captured

### dayboro.au Hummingbird Config Summary
- Page cache: disabled
- Minification: disabled
- Browser cache: 1 year for all static assets
- Database cleanup: enabled, weekly Friday
- Delay JS: configured but not active

### dayboro.au Smush Config Summary
- Auto smush: enabled
- Lossy compression: level 2 (medium)
- WebP: enabled
- Lazy load: enabled
- PNG to JPG: enabled
- Backup originals: enabled

### All WordPress Sites Audit
- 14 sites total
- 10 use Elementor
- 3 use Hummingbird
- 8 use Smush
- Total image uploads: ~9GB across all sites
- Hummingbird cache on disk: ~58MB
- Actively used cache: ~43MB (rds.ink), 14MB (m8chat.au)

---

**Report prepared by:** Claude Code  
**Date:** 2026-05-17  
**Status:** Ready for Phase 1 approval