Technical notebook

Writing

Notes, experiments, and longer writing around systems, engineering software, simulation, and things I am currently trying to understand.

Recent

  1. 微电子学习工具笔记

    课程工具、操作说明、官方配图与RC练习。

  2. A result object is not an evaluated result

    Why automated ANSYS Mechanical work needs separate geometry, setup, solver-input, solved, evaluated, and credibility gates.

  3. Separating DNS failure from slow CDN delivery

    A redacted remote-development download investigation that isolates resolver, redirect, edge-selection, and throughput stages.

  4. Separating measured cough-model runs from the experiment plan

    A subject-disjoint Coswara pipeline, compact teacher–student baselines, and the evidence boundary before any clinical or SOTA claim.

  5. What the HarmonyOS prototype could and could not verify

    ArkTS UI, browser OAuth and deep links, native Mihomo libraries, VPN Extension limits, and repository-level evidence.

  6. Why ForgeSpec keeps a typed specification between a prompt and CAD

    Separating requirement interpretation, domain planning, deterministic CadQuery generation, and artifact validation.

  7. The trend survived mesh refinement; the amplitude did not

    A three-level finite-element audit where response ordering remained stable but the absolute output missed the registered mesh-independence criterion.

Featured

  1. A result object is not an evaluated result

    Why automated ANSYS Mechanical work needs separate geometry, setup, solver-input, solved, evaluated, and credibility gates.

  2. Separating measured cough-model runs from the experiment plan

    A subject-disjoint Coswara pipeline, compact teacher–student baselines, and the evidence boundary before any clinical or SOTA claim.

  3. Why ForgeSpec keeps a typed specification between a prompt and CAD

    Separating requirement interpretation, domain planning, deterministic CadQuery generation, and artifact validation.

  4. The trend survived mesh refinement; the amplitude did not

    A three-level finite-element audit where response ordering remained stable but the absolute output missed the registered mesh-independence criterion.

  5. One shell, two observers: the persistent terminal architecture

    How a native broker owns one PTY or ConPTY while an agent and a human share observation and controlled input.

Research / Build Notes

  1. A result object is not an evaluated result
  2. Separating measured cough-model runs from the experiment plan
  3. What the HarmonyOS prototype could and could not verify
  4. Why ForgeSpec keeps a typed specification between a prompt and CAD
  5. The trend survived mesh refinement; the amplitude did not
  6. One shell, two observers: the persistent terminal architecture

Browse by type

Topics

Series

All writing

Essay

No public entries yet.

Build Note

  1. A result object is not an evaluated result
  2. What the HarmonyOS prototype could and could not verify
  3. Why ForgeSpec keeps a typed specification between a prompt and CAD
  4. One shell, two observers: the persistent terminal architecture
  5. What I tested in the persistent terminal broker
  6. A narrow Mechanical delivery protocol
  7. A T-Rex runner on a 16×2 character LCD

Research Note

  1. Separating measured cough-model runs from the experiment plan
  2. The trend survived mesh refinement; the amplitude did not
  3. When a parameter sweep cannot identify its own model

Note

  1. 微电子学习工具笔记
  2. Separating DNS failure from slow CDN delivery
  3. Known limits of a shared persistent terminal
  4. Failure taxonomy for automated Mechanical handoffs