SK hynix 記憶體週期分析：端到端產品測試流程

You are running a source-grounded MadCowork Office E2E workflow.
Communicate final user-facing conclusions in Traditional Chinese, but all tool planning and tool arguments must be concise English.

Topic: SK hynix 2026 memory-cycle analysis: financial statements, memory pricing, inventory turnover days, and software/AI infrastructure insight.

Known source targets to fetch first:
- Official SK hynix Q1 2026 results: https://news.skhynix.com/q1-2026-business-results/
- Official SK hynix FY2025 results: https://news.skhynix.com/sk-hynix-announces-fy25-financial-results/
- Official SK hynix Q1 2025 results: https://news.skhynix.com/sk-hynix-announces-1q25-financial-results/
- SK hynix analyst interview on AI infrastructure and memory demand: https://news.skhynix.com/2026-analyst-interview-02/
- TrendForce DRAM price page: https://www.trendforce.com/price
- StockAnalysis SK hynix income statement: https://stockanalysis.com/quote/krx/000660/financials/
- StockAnalysis SK hynix balance sheet: https://stockanalysis.com/quote/krx/000660/financials/balance-sheet/

Hard requirements:
1. Do not rely on model memory. Use web_search, web_fetch, and web_crawl as needed before making factual or numerical claims.
2. Gather evidence from multiple sources, including SK hynix official newsroom / IR-style reports and at least two third-party market or financial data sources.
3. Create CSV files in this session workspace:
   - skhynix_sources.csv with columns: source_type,title,url,published_date,used_for.
   - skhynix_financials.csv with columns: company,metric,period,value,unit,source_url,note.
   - memory_prices.csv with columns: product,price_type,period,value,unit,source_url,note.
   - inventory_metrics.csv with columns: company,metric,period,value,unit,formula,source_url,note.
   - skhynix_chart_inputs.csv with columns: chart_name,series,period,value,unit,source_url,note.
   - session_metadata.csv with columns: artifact,type,path,created_at,source_inputs.
4. Every numeric claim used in the deck must have a source_url in one of the CSV files.
5. Include these real numerical analyses if the sources support them:
   - Q1 2026 revenue, Q1 2025 revenue, Q4 2025 revenue, Q1 2026 operating profit, Q1 2026 net income, operating margin, net margin.
   - FY2025 revenue, operating profit, net income, and operating margin.
   - 2025 cost of revenue / cost of sales and inventory from a financial-data source or official financial statements.
   - Annual inventory days / DIO using: average inventory / cost of revenue * 365. Use FY2025 and FY2024 inventory if available. If only one period is available, clearly label the limitation.
   - DRAM or DDR4/DDR5 price / price-change data from TrendForce or another fetchable third-party memory market source.
   - HBM / AI memory demand insight only when backed by a fetched source.
6. Create chart-ready rows in skhynix_chart_inputs.csv for:
   - Revenue and operating profit trajectory: Q1 2025, Q4 2025, Q1 2026, FY2025.
   - Profitability share: operating margin and net margin for Q1 2026 and FY2025.
   - Inventory / DIO: inventory, cost of revenue, and calculated DIO.
   - Memory pricing: every fetched DRAM/DDR price or price-change point.
   - Insight bridge: HBM/AI demand, price strength, inventory days, and software/AI infrastructure implication as sourced qualitative rows with value=1 if needed.
7. Call source_validate on skhynix_sources.csv with min_official_sources=3 and min_third_party_sources=2. If analyst/market data cannot be fetched, record the data gap in the deck instead of inventing it.
8. Before charting or xdeck, call list_skills, read xdata-office-deck-workflow and xdata-spreadsheet-workflow if available, call csv_profile on the financial/chart CSVs, then call csv_validate on every CSV.
9. Generate a self-contained web presentation through xdeck_generate. When calling xdeck_generate, set out_dir exactly to "skhynix-memory-web-deck" so the final file is skhynix-memory-web-deck/deck.html. Pass facts_paths, sources_path, chart_data_path, and validation_report_path when available.
10. The deck should be 8 slides, Traditional Chinese, investor / memory-cycle strategy briefing style, with speaker_note on each slide.
11. Include insight about software and AI infrastructure expectations: AI memory budgets, HBM capacity priority, possible tightness in commodity DRAM/NAND, inventory-risk caveat, and valuation-risk caveat. Ground all numerical claims in CSV rows.
12. Run xdeck_audit and xdeck_screenshots. If the audit finds layout issues, run one xdeck_revise pass.
13. Final answer: list generated artifact paths and briefly explain the evidence chain.

Important: if a source cannot be fetched, record the failed URL in skhynix_sources.csv with used_for="attempted but unavailable" and continue with fetchable sources.

Correction pass for the SK hynix E2E workflow.

You already fetched the StockAnalysis and TrendForce evidence files. The first pass left inventory_metrics.csv as unknown/N/A even though the evidence directory contains the needed values. Do not rely on model memory. Read the saved evidence files and update the CSVs from those files.

Required corrections:
1. Read evidence/stockanalysis.com-financials-b876eefd4e.md and evidence/stockanalysis.com-balance-sheet-40968329b2.md.
2. Update skhynix_financials.csv with StockAnalysis rows for Cost of Revenue:
   - FY2025 Cost of Revenue = 38455885 Million KRW from https://stockanalysis.com/quote/krx/000660/financials/
   - FY2024 Cost of Revenue = 34364814 Million KRW from the same source.
   Also preserve the official Billion KRW rows already present.
3. Update inventory_metrics.csv with:
   - Inventory FY2025 = 14289390 Million KRW from https://stockanalysis.com/quote/krx/000660/financials/balance-sheet/
   - Inventory FY2024 = 13313937 Million KRW from the same source.
   - Average Inventory FY2025 = (14289390 + 13313937) / 2 = 13801663.5 Million KRW.
   - Cost of Revenue FY2025 = 38455885 Million KRW.
   - DIO FY2025 = 13801663.5 / 38455885 * 365 = 131.02 days.
4. Read evidence/trendforce.com-price-9edf44a1bf.md and update memory_prices.csv with numeric TrendForce rows, including:
   - DDR5 16Gb 4800/5600 session average 41.50, session change 0.40%.
   - DDR4 16Gb 3200 session average 60.857, session change 1.19%.
   - DDR5 RDIMM 32GB session average 975.00, average change 1.56%.
   - DDR4 UDIMM 16GB session average 150.761, average change 1.57%.
5. Update skhynix_chart_inputs.csv so Inventory / DIO and Memory Pricing charts contain the numeric values above, not N/A or qualitative strings.
6. Run csv_profile on skhynix_financials.csv, inventory_metrics.csv, memory_prices.csv, and skhynix_chart_inputs.csv. Run csv_validate on every CSV and source_validate on skhynix_sources.csv.
7. Regenerate the same 8-slide web presentation by calling xdeck_generate with out_dir exactly "skhynix-memory-web-deck", passing the updated CSVs. Then run xdeck_audit and xdeck_screenshots.
8. Final answer in Traditional Chinese: explain that this was a correction pass fixing inventory/DIO and memory-price numeric extraction.