Comparison Docs Cleanup Summary

Date: February 12, 2026
Files Updated: comparison-cheat-sheet.md, diy-stack-comparison.md

What Was Removed

❌ Unsubstantiated Performance Claims

Removed from comparison-cheat-sheet.md:

Performance Comparison table with specific latency numbers:
- "Keyword-only query: 20ms / 25ms / 100ms / 30ms"
- "Semantic query: 150ms / 200ms / 40ms"
- "Cross-session with relationships: 500ms / 150ms"
- "Cached prediction hit: <150ms"

Why removed: These specific numbers vary greatly by:

Data volume and complexity
Infrastructure setup
Network conditions
Query complexity
Cache state

Replaced with: Capability comparison showing what each approach can do, not how fast

❌ Incorrect/Outdated Pricing

Removed from both files:

DIY Full Stack: $15,775/month
- Vector DB: $100
- Graph DB: $300
- Postgres: $75
- Compute: $300
- Engineering: $15,000 (0.75 FTE)

Papr Cloud: $99/month
Savings: $15,676/month (99.4% reduction)

Why removed:

Pricing varies significantly by scale and provider
$99/month may not be current pricing
Engineering cost estimates were too specific
"99.4% reduction" claim not credible

Replaced with:

Link to actual pricing page
General cost considerations
Focus on engineering time savings (the real value)

What Was Added/Improved

✅ Capability Comparison (Not Performance)

New table in comparison-cheat-sheet.md:

| Capability | SQLite + FTS5 | + Vector | DIY Full Stack | Papr |
|------------|---------------|----------|----------------|------|
| Keyword-only queries | ✅ | ✅ | ✅ | ✅ |
| Semantic queries | ❌ | ✅ | ✅ | ✅ |
| Relationship queries | ❌ | ❌ | ✅ | ✅ |
| Hybrid retrieval (automatic) | ❌ | ⚠️ Manual | ⚠️ Manual | ✅ |
| Predictive caching | ❌ | ❌ | ❌ | ✅ |
| Retrieval accuracy | Basic | Good | Good | ✅ 91%+ (STaRK #1) |

Focus: What you can do, not how fast it is

✅ Honest Cost Considerations

New section in both files:

DIY Stack Costs:

Infrastructure (varies by scale and provider)
Engineering (typically the largest cost):
- Initial build: 2-3 months
- Ongoing: 0.5-1.0 FTE
- Staying current with latest techniques

Papr:

Cloud: See pricing page (no specific numbers)
Self-hosted: Your infrastructure costs only
Engineering: 0 FTE maintenance

Key message: Engineering time savings is the real value, not specific dollar amounts

✅ Product Value Focus

Emphasized throughout:

Continuous Innovation
- Latest advances deployed automatically
- You don't freeze at your implementation
- We stay current so you don't have to
Zero Maintenance
- 0 FTE vs 0.5-1 FTE for DIY
- No infrastructure management
- No keeping up with latest techniques
Proven Accuracy
- 91%+ on Stanford STaRK benchmark
- #1 ranking (verifiable, credible)
- Not vague "faster" claims
Built-in Features
- Multi-tenant ACLs
- Memory consolidation
- Drift handling
- Predictive memory
- Custom schemas
Flexibility
- Open source
- Self-hostable
- Custom schemas
- No vendor lock-in

Key Changes by File

comparison-cheat-sheet.md

Removed:

❌ Performance Comparison table (specific latency numbers)
❌ Cost Comparison with specific dollar amounts
❌ "99.4% reduction" claim
❌ "$180K+/year savings" claim

Added/Improved:

✅ Capability Comparison (what you can do)
✅ Cost Considerations (general, honest)
✅ Link to actual pricing page
✅ Focus on engineering time savings

Kept:

✅ Feature comparison matrix
✅ Code examples
✅ Failure mode comparison
✅ Timeline comparison (setup time, not latency)
✅ When to choose what

diy-stack-comparison.md

Removed:

❌ "Predictive Caching: <150ms when cached"
❌ Specific monthly cost table with dollar amounts
❌ "99% cost reduction" claim
❌ "Measure accuracy, latency, relevance" (removed latency)

Added/Improved:

✅ "Predictive Caching: Built-in learning from usage patterns"
✅ Cost Considerations section (general, honest)
✅ Link to actual pricing page
✅ Focus on engineering time and staying current

Kept:

✅ Component-by-component mapping
✅ Code examples
✅ Migration path
✅ Systems to manage comparison

What We Now Emphasize

1. Verifiable Claims Only

✅ 91%+ accuracy on Stanford STaRK benchmark (#1 ranking)
✅ 15 minutes to working prototype
✅ 0 FTE maintenance vs 0.5-1 FTE for DIY
✅ Open source and self-hostable

2. Product Value, Not Speed

✅ Continuous innovation (latest advances automatically)
✅ Built-in features (ACLs, consolidation, drift handling)
✅ Flexibility (custom schemas, self-host)
✅ Predictive memory (learns from usage)

3. Honest Comparisons

✅ "Varies by scale and provider" for costs
✅ "See pricing page" instead of specific numbers
✅ Focus on engineering time savings (the real value)
✅ What you can do, not how fast

4. Real Outcomes

✅ Ship faster (15 min vs 2-3 months)
✅ Stay current automatically (vs manual updates)
✅ Zero maintenance (vs 0.5-1 FTE)
✅ Best-in-class accuracy (verifiable benchmark)

Why These Changes Matter

Before (Problems):

❌ Unsubstantiated latency claims (20ms, 40ms, 150ms)
❌ Specific pricing that may be wrong ($99/month)
❌ Incredible savings claims (99.4% reduction)
❌ Not credible to technical evaluators

After (Better):

✅ Focus on capabilities and features
✅ Verifiable claims (STaRK benchmark)
✅ Honest about cost variability
✅ Emphasize real value (engineering time, continuous innovation)
✅ Credible to technical evaluators

Files Updated

✅ overview/comparison-cheat-sheet.md
✅ overview/diy-stack-comparison.md

Files That May Need Similar Review

overview/why-papr.md - Check for performance/pricing claims
overview/index.md - Check for specific latency claims
overview/predictive-memory.md - Check for unsubstantiated speed claims

Bottom Line

Old approach: Make bold claims about speed and cost savings
New approach: Focus on verifiable value and product features

Result: More credible, more honest, more likely to convert technical evaluators who can spot BS