跳到主要内容
interview-defense · 代码
查看代码目录
Graphics Learning Lab Interview Defense / Source

Observe · Change · Recompute · Explain

把回答写成可复盘的实验记录

Read-only · exact teaching model

评分不是为了给人贴标签,而是为了指出下一条可验证的证据。

这一课怎么学

  1. 观察
  2. 预测
  3. 改一个量
  4. 看证据
  5. 再读代码
01 / RUBRIC

A story is a measurable interface

Score context, action, result, failure and ownership instead of memorizing a paragraph.

// Interview Defense Lab — executable rubric
const evidence = weighted({ context, action, result, failure, ownership });
const clarity = weighted({ specificity, metrics, stakeholders, tradeoff });
const timing = answerSeconds / targetSeconds;
02 / RISK

Tradeoffs need an explicit boundary

The answer becomes senior when it names the cost and fallback.

if (!aiDisclosure || !aiVerification) risk = "high";
if (evidence < 55 || clarity < 55) risk = "high";
else if (evidence < 75 || clarity < 75) risk = "medium";
03 / FOLLOW-UPS

Missing evidence becomes the next question

A defense lab turns weak spots into a practice queue.

followUps = missingEvidence.map((gap) => promptFor(gap));
reportId = stableHash({ scene, evidence, clarity, timing });