Refund Request: Significant Token Waste Due to Repeated Claude Code Errors - Account rorudit52@gmail.com
Formal Refund Request
Account: rorudit52@gmail.com
Product: Claude Code CLI
Model: claude-sonnet-4-6
---
Project Context
I used Claude Code to build VEMO FuelIQ, a full maritime compliance SaaS platform at vemofuel.com. The project included a Next.js application, database integration, marketing website, and compliance calculation features across multiple weeks and sessions.
---
AI Mistakes That Wasted Tokens
1. Swiper.js Integration Broke Homepage Completely
Claude replaced a working hero carousel with Swiper.js without testing for CSS conflicts. All three slides rendered simultaneously overlapping. Required 7 extra fix deploys and multiple full reads of a 1000+ line HTML file.
2. AOS.js Made Entire Page Invisible
Claude removed trigger JS while leaving opacity:0 CSS. Every section became permanently invisible. Required 3 more fix cycles.
3. Same Files Read 15-20 Times
index.html (1000+ lines), CLAUDE.md, memory files read repeatedly across sessions instead of being retained efficiently.
4. Imprecise sed Commands Applied Multiple Times
Domain migration and color changes were incomplete and had to be repeated across multiple sessions.
5. Multiple Context Compaction Cycles
4 to 5 compaction events requiring expensive re-loading of already-processed context.
---
Estimated Token Waste
| Category | Tokens |
|---|---|
| Legitimate work | 6-8 million |
| Wasted on repeated reads | 3-4 million |
| Wasted fixing AI mistakes | 2-3 million |
| Wasted on redo of changes | 1-2 million |
| Total waste | 6-9 million |
Estimated total cost: $150-$280 USD. AI errors responsible for 40-50%. Financial loss from mistakes: $60-$140 USD.
Claude Code itself acknowledged these mistakes when asked directly.
---
Request
- Review of token usage on account rorudit52@gmail.com
- Refund for tokens consumed by AI errors
- Compensation for 8-12 hours of user time lost
Udit Kumar | Owner, VEMO FuelIQ | hello@vemofuel.com
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗