Waste Detection

The Waste Detection page identifies patterns in your sessions that waste tokens and money.

The 7 waste patterns

1. Retry Storms

Repeating the same prompt with >80% word overlap within 60 seconds.

Impact: Duplicated tokens, no progress. Fix: Add constraints or error messages. Change approach.

2. Context Bloat

Sessions where input tokens grow >3x from the first turn without using /compact.

Impact: Higher cost per turn, slower responses, degraded output quality. Fix: Use /compact every 15–20 turns. Use /clear when switching tasks.

3. Model Overkill

Using Opus for tasks that produce <200 tokens of output.

Impact: 10–15x cost premium for simple work. Fix: Use Sonnet for simple tasks. Reserve Opus for complex reasoning.

4. Vague Prompts

Prompts with no specificity markers (file paths, function names, errors).

Impact: AI guesses, leading to corrections and wasted turns. Fix: Always include the file path and function name.

5. Multi-Task Bundling

Prompts with 3+ tasks bundled together.

Impact: AI handles none of them well, leading to partial results and corrections. Fix: One task per prompt.

6. Correction Cascades

Chains of corrections where each fix introduces a new issue.

Impact: Multiplied token spend with diminishing returns. Fix: Stop, re-read the code, plan a different approach.

7. Unverified Output

Sessions with no test runs, type-checks, or verification prompts.

Impact: Bugs ship, causing downstream rework. Fix: Run tests after every significant change.

Dashboard view

For each detected pattern, the page shows:

• Severity — how often and how badly this pattern appears
• Occurrences — count of instances in the selected time range
• Estimated waste — tokens and cost wasted by this pattern
• Trend — is this pattern getting better or worse?
• Affected sessions — links to specific sessions where the pattern was detected

Savings estimates

The page calculates potential savings if each waste pattern were eliminated:

Example:
  Retry storms:     $12.50/month  (eliminated by adding constraints)
  Model overkill:   $8.00/month   (eliminated by using Sonnet for simple tasks)
  Context bloat:    $5.00/month   (eliminated by regular /compact)
  Total potential:  $25.50/month savings