deep-investigation
追兇文。反直覺現象、假設排除、根因揭露。
structure
The investigation archetype starts with a puzzle: something unexpected happened, metrics don't add up, or a system behaves contrary to expectation. The post then methodically eliminates hypotheses until the root cause is revealed.
- Observation — the puzzling phenomenon (counterintuitive metric, strange behavior, unexpected result)
- Hypotheses (1–3 H2 sections) — each a plausible explanation for the observation
- The truth — the actual root cause, why the initial hypothesis was wrong
Works for: debugging post-mortems, performance mysteries, cost anomalies, system behavior surprises.
required widgets
- Metrics visualization — charts showing the puzzle (the anomaly that started the investigation)
- Timeline, flame graph, or sequence diagram — shows the reveal or root-cause mechanism
visual silhouette
archetype-check rules
- H2 structure: "observation", 1–3 "hypothesis: " sections, "the truth"
- `<p class="vg-deep-opener">` exists
- `<span class="vg-dropcap">` exists with exactly 1 character
- ≥2 inline `<svg>` elements (metrics and reveal)
- `<p class="vg-deep-closer">` with `<strong>Take-away</strong>`
- Universal contract: opener, dropcap, closer with strong tag
when to use
Investigation is perfect for posts that are about problem-solving methodology. The reader learns not just what the root cause was, but how to systematically rule out hypotheses. This archetype transforms debugging stories into educational narratives.
Posts using this archetype
- 06.15 把 x86-64 靜態翻成 aarch64——Elevator 的程式碼膨脹之謎
- 06.14 FFmpeg 一口氣被挖出 21 個零日——媒體解碼的攻擊面有多深
- 06.13 把安全掃描吞吐量拉高十倍——Cloudflare 一場不加機器的瓶頸獵巡
- 06.09 愚弄 Go 的 X.509 驗證——一場由 ASN.1 字串編碼差異引發的 fail-closed 追查
- 06.08 How is Linear so fast——把資料庫搬進瀏覽器後,每一層都得跟著改
- 06.07 Claude 真的讓 rsync 變多 bug 了嗎——用數據翻案一場社群怒火
- 06.04 合成鍵盤事件如何串成一鍵偷 token——拆 github.dev 的 VSCode 信任模型
- 05.31 Faster than Light——LinkedIn 怎麼把 generative recommender 的訓練吞吐拉上來
- 05.30 CAPTCHA 還擋得住 agent 嗎——把 Claude、ChatGPT Agent、Comet 拉到 image / reCAPTCHA / Turnstile 前
- 05.27 從 model scaling 到 system scaling——agentic AI 的瓶頸搬到 harness
- 05.22 LinkedIn 的 HashMap resize 凍結——調查一場記憶體預算失算