# Micron (MU) Newsflow — 13-Month Sampled Archive

## Goal
Comprehensive newsflow on everything Micron-related, with thorough samples of the first 3 days of each month for 13 months (Jan 2025 → Jan 2026).

## Scope
- **All Micron news**: Earnings, product launches, partnerships, supply/demand, analyst notes, industry context (DRAM/NAND/HBM pricing, competitors)
- **Sampling strategy**: First 3 calendar days of each month × 13 months = 39 days of thorough coverage
- **Sources**: Financial news (Reuters, Bloomberg, WSJ), tech press (Tom's Hardware, AnandTech, SemiAnalysis), social (Twitter/X finance), analyst reports, SEC filings

## Discovery Strategy
Per-month web search blitz:
1. **Google News search**: `"Micron" OR "MU" site:reuters.com OR site:bloomberg.com ...` date-restricted to 1st-3rd of each month
2. **Finnhub news API**: Company news for MU, date-filtered
3. **SEC EDGAR**: 8-K, 10-Q, 10-K filings in window
4. **Twitter/X search**: High-follower accounts mentioning Micron in window
5. **YouTube**: Earnings calls, analyst videos in window

## Stages
1. **discover** — Collect URLs per month-window
2. **fetch** — Download/extract article text (reuse brain/ content extraction)
3. **classify** — Tag: `{category: earnings|product|supply_chain|analyst|industry|filing, sentiment: bullish|bearish|neutral}`
4. **summarize** — Per-article 2-sentence summary + key data points
5. **compile** — Monthly digest: top stories, price context, sentiment arc

## Storage
```
data/companies/mu/
├── newsflow/
│   ├── 2025-01/
│   │   ├── articles/
│   │   │   ├── {hash}_{source}.json   # {url, title, text, date, classification, summary}
│   │   │   └── ...
│   │   └── digest.md                  # Monthly compilation
│   ├── 2025-02/
│   │   └── ...
│   └── 2026-01/
│       └── ...
├── index.jsonl
└── sources.yaml
```

## Key Considerations
- **brain/ reuse**: Content extraction should use `rivus/brain` for paywalled sources
- **Date precision**: Articles must be from the 1st-3rd, not just published-date — check actual content dates
- **Dedup**: Same article syndicated across outlets → keep highest-quality source
- **Price context**: Include MU price on each sample day for correlation

## Pacing
- Web search: 20/hr (avoid rate limits)
- Content fetch: 30/hr (via brain/)
- LLM stages: 60/hr