Conversation
Fills the gap between mechanical theme index (themes/低軌衛星.md, auto-generated by build_themes.py from wikilink reverse lookup, covers 42 TW companies) and curated narrative that maps technology blocks → chip → module → system → tickers. Deliverables: - docs/analysis/2026-06-18-LEO-satellite-deep-dive.md (~6k 字 Traditional Chinese, 81 unique wikilinks, 4 constellations × 6 chip blocks × 8 tickers) - docs/superpowers/specs/2026-06-18-leo-satellite-analysis-design.md (full design spec from brainstorming session) - docs/analysis/research-notes/00-verification-log.md (6-point adversarial verification ledger that caught 2 worker confabulations: W06 fabricated "2026 Q1 80.81% LEO" and W04 fabricated "7717 50% satellite H1 2025") - scripts/build_themes.py: 5-line additive patch — optional analysis_doc field on THEME_DEFINITIONS injects 📖 deep-dive header link when set. Backward compatible: other themes unaffected. Reusable for future CoWoS / 矽光子 / HBM deep-dives. - themes/低軌衛星.md: regenerated; now header-links the deep-dive. Methodology: 12 parallel research worker agents (4 tech-chapter + 8 ticker deep-dive) gathered material; permission denial in worker worktrees forced all to return inline findings rather than persisted files. Coordinator aggregated inline findings into the main report after adversarial cross-verification of 6 high-priority quantitative claims (4/6 confirmed, 2/6 confabulations caught, 1 source URL unverifiable kept as 市場傳聞). Honesty markers throughout the report: - §3.6 explicitly documents TW ADC/DAC structural gap (TASA Beyond 5G LEO payload built by US vendor, per Focus Taiwan 2025-04-01). - §5.3 全訊: no public Starlink design-win evidence — framed as defense-LEO SAR + ground-station mmWave PA, not Starlink supplier. - §5.7 聯發科 中性贏家 thesis verified against 2 official MediaTek press releases (2025-02-24 first 5G NR-NTN over OneWeb LEO with ITRI gNB; 2025-11-03 first 3GPP Rel-19 5G-Advanced NR-NTN over OneWeb LEO with ESA + Sharp + ITRI + R&S). Confidence marker convention applied: unmarked = IR/法說會/年報 confirmed, *（市場傳聞）* = press only, *（推測）* = inferred from product line. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This was referenced Jun 18, 2026
Per the newly-established analytical framework in CLAUDE.md: 1. Add §6.3 對偶 long/short candidate mapping — every catalyst/risk now names symmetric winner + loser pair (e.g., Direct-to-Cell long 聯發科 → short 啟碁/攸泰科技/台揚) 2. Expand Appendix B to include channel-check 加權 (B.1) — explains per-source trust rationale based on 第幾手 / 位置 / 誘因 / track record. Renumbered original verification summary to B.2. Framework principle applied: "同一個變化必然伴隨對稱輸家". The report already implicitly identified some losers (e.g., D2C shrinking UT market), but now they're explicitly paired and tabulated. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
leehao921
added a commit
to leehao921/Semiconductor-Supply-Chain
that referenced
this pull request
Jul 9, 2026
Unit 8: Docker scheduler container
leehao921
added a commit
to leehao921/Semiconductor-Supply-Chain
that referenced
this pull request
Jul 9, 2026
Triggered by 2026/4 Infineon-led 5-15% pricing cycle, Nexperia (Wingtech) Dutch-government export control, and AI server 800V HVDC architecture upgrade — three convergent catalysts hitting TW power semi cluster. Deliverables: - docs/analysis/2026-06-18-power-semi-pmic-deep-dive.md (~21k chars Traditional Chinese, 61 unique wikilinks, 13 individual ticker deep-dives) - docs/superpowers/specs/2026-06-18-power-semi-pmic-design.md (full design spec) - scripts/build_themes.py: (1) restore analysis_doc rendering logic (additive, backward compatible) — auto-injects 📖 header link when a THEME_DEFINITION provides analysis_doc field; (2) add new 功率半導體 theme pointing to the deep-dive - themes/功率半導體.md: new auto-generated theme index with 📖 deep-dive link Coverage spans 3 sub-clusters per user direction: - Discrete IDM/fabless: 強茂 2481, 德微 3675, 台半 5425, 富鼎 8261, 大中 6435, 杰力 5299 - Wafer foundry / Epitaxy: 漢磊 3707 (SiC/GaN), 世界先進 5347 (8" BCD), 嘉晶 3016 (SiC/GaN epi) - PMIC integrated: 矽力-KY 6415, 致新 8081, 茂達 6138, 天鈺 4961 Three catalysts each verified to ≥2 sources: - Infineon 4/1 5-15% hike confirmed via 工商時報 + 數位時代 + Smart 自學網 - TI 7/1 PMIC+MOSFET second wave confirmed via Pocketstock + 數位時代 - Nexperia event timeline (Dutch 10/13 takeover, China retaliation, Wingtech $1.2B lawsuit) via Wikipedia + Nexperia official press - NVIDIA 800V HVDC 2026Q4 timeline via SemiAnalysis + Szsanyi - Open Compute ORv3/ORv4 specs via OCP + Molex Honesty markers: - §3.2 IGBT TW gap (high-voltage power modules still dominated by Infineon/Mitsubishi/Fuji — TW structural weakness) - §3.5 SiC 8" wafer lag (1-2 years behind Wolfspeed/STM) - §5.6 漢磊 still loss-making (2025 GM -3.30%) — call this out explicitly - §6 risks include Nexperia US capacity ramp narrowing TW transfer window Confidence markers: unmarked = IR/法說會/年報 confirmed, *（市場傳聞）* = press-only, *（推測）* = inferred from product line. Note: this PR's build_themes.py adds analysis_doc rendering logic. The sibling LEO satellite PR (Timeverse#4) adds the same logic plus the 低軌衛星 analysis_doc field. Merge order: either PR first; the second merge will need to resolve the duplicate rendering-logic block in build_themes.py around line 226 (additive lines, easy to merge by hand). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
docs/analysis/2026-06-18-LEO-satellite-deep-dive.md)，涵蓋 4 大星座 (Starlink v1/v2/v2-full, Kuiper, OneWeb/Eutelsat, 台星 B5G) × 6 個晶片 block × 8 家個股 deep-divescripts/build_themes.py：新增 optionalanalysis_doc欄位 (additive 5-行 patch)，使themes/<topic>.md自動產出時可注入 📖 deep-dive header link。向後相容，其他主題不變；可重用於 CoWoS / 矽光子 / HBM / EUV 之後的 deep-divethemes/低軌衛星.md重生：頂部新增 📖 連結，雙向 link 機制完成docs/superpowers/specs/2026-06-18-leo-satellite-analysis-design.md：完整 brainstorming spec（產品需求、章節骨架、研究方法論）docs/analysis/research-notes/00-verification-log.md：6 點 adversarial verification log，攔截 2 個 worker confabulation報告誠實揭露點
Adversarial verification (見 Appendix B + research-notes/00-verification-log.md)
方法論
12 個 parallel background research worker agents（4 個 tech-chapter + 8 個 ticker deep-dive）並行收集材料；workers 在 worktree 內遇 permission denial 無法寫檔/commit，全部以 inline final message 回傳 findings。coordinator 從 inline findings 跨 worker 整合，再以 6 個 high-priority 量化主張 WebFetch 交叉驗證後串行寫主報告。
Test plan
python3 scripts/build_themes.py 低軌衛星重生，頂部出現 📖 linkif theme_def.get(\"analysis_doc\")guard）🤖 Generated with Claude Code