# Exp 4 Timeline

## Parameters
- **Models:**
  - Planning, Relevance, Gap Analysis: **glm-4.7-flash** (input $0.07/MTok, output $0.4/MTok)
  - Synthesis: **glm-5.1** (качественный финал)
- **lightContext:** true
- **Attachments:** JSON state files
- **Start balance:** z.ai $10.50, Perplexity $29.20

## Timeline

| Step | Start (UTC) | End (UTC) | Duration | Tokens | Notes |
|------|-------------|-----------|----------|--------|-------|
| 1. Planning | - | - | 30s | 2.8k (in 1.8k/out 990) | glm-4.7-flash, 5 dimensions, 8 threads |
| 2. Search | - | - | 3.5s | - | 8 Perplexity queries, 40 results |
| 3. Relevance | - | - | 1m34s | 26.1k (in 22.4k/out 3.7k) | glm-4.7-flash, 20 URLs filtered, caught misidentifications |
| 4. Reading | - | - | 30s | - | 8 URLs fetched, 2 failed (429 Citilink) |
| 5. Gap Analysis | - | - | 44s | 4.7k (in 3.3k/out 1.5k) | glm-4.7-flash, 9 gaps identified |
| 6. Iterative Search | - | - | 5s | - | 9 queries, FOUND induction with rotary switches! |
| 7. Synthesis | - | - | 1m16s | 10.4k (in 7.6k/out 2.8k) | glm-5.1, 3 bundles created |
| **TOTAL** | - | - | **~4m** | **~44k** | - |
| 2. Search | - | - | - | - | Perplexity |
| 3. Relevance | - | - | - | - | glm-4.7-flash |
| 4. Reading | - | - | - | - | web_fetch |
| 5. Gap Analysis | - | - | - | - | glm-4.7-flash |
| 6. Iterative Search | - | - | - | - | Perplexity |
| 7. Synthesis | - | - | - | - | glm-5.1 |

## Token Tracking

| Step | Input | Output | Model | Cost |
|------|-------|--------|-------|------|
| 1. Planning | - | - | glm-4.7-flash | - |
| 2. Search | - | - | Perplexity | - |
| 3. Relevance | - | - | glm-4.7-flash | - |
| 4. Reading | - | - | web_fetch | - |
| 5. Gap Analysis | - | - | glm-4.7-flash | - |
| 6. Iterative Search | - | - | Perplexity | - |
| 7. Synthesis | - | - | glm-5.1 | - |
| **Total** | - | - | - | - |

## Hypothesis

**Expected cost:** $0.50-0.80 (50-70% cheaper than Exp 3's actual $2.07)
**Expected quality:** Same 10/10 (flash model for simple filtering tasks, 5.1 for final synthesis)