Back to Details

verify-this

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Verify This

验证此声明

Adapted from Cursor's 
cursor-team-kit/verify-this
(MIT, see THIRD_PARTY.md), with this app's surfaces.
Verification is not a recap. It proves or disproves a specific claim with repeatable evidence.
改编自 Cursor 的 
cursor-team-kit/verify-this
（MIT，见 THIRD_PARTY.md），并适配了本应用的验证入口。
验证不是简单的复述。它通过可重复的证据证明或推翻一项具体声明。

When To Use

何时使用

  • The person asks "verify this", "prove it works", "did this fix it", or "show me the evidence".
  • A bug fix needs a before/after repro.
  • A page, API, agent-output or performance claim needs measurement.
  • A test passes but the user-visible behaviour still needs confirmation.
Do not use this for vague claims like "the code is cleaner". Ask for a measurable claim first.
  • 用户提出“验证这个”“证明它有效”“这是否修复了问题”或“展示证据”。
  • 某个 bug 修复需要修复前后的复现对比。
  • 页面、API、Agent 输出或性能声明需要实际测量。
  • 测试已经通过，但用户可见的行为仍需确认。
不要将此流程用于“代码更清晰”之类的模糊声明。请先要求用户提供可测量的声明。

Workflow

工作流

  1. Restate the claim in falsifiable form: condition, metric, and threshold.
  2. Pick the smallest local surface that can disprove it (below).
  3. Capture a baseline from the old state: the merge base, the parent commit, or the current broken repro (a git worktree of the base gets its own verify instance).
  4. Capture the treatment from the changed state with the same command, data and environment.
  5. Compare raw artifacts: numbers, snapshots, HTTP responses, run rows, eval tables.
  6. Return exactly one verdict: 
    VERIFIED
    , 
    NOT VERIFIED
    , or 
    INCONCLUSIVE
    .
  1. 将声明改写为可证伪的形式：条件、指标和阈值。
  2. 选择能够将其证伪的最小本地验证入口（见下文）。
  3. 从旧状态采集基线：合并基点（merge base）、父提交（parent commit），或当前故障的复现结果（基线的 git worktree 应拥有独立的 verify 实例）。
  4. 使用相同的命令、数据和环境，从变更后的状态采集处理组（treatment）。
  5. 对比原始产物：数值、快照、HTTP 响应、运行记录行、评测表格。
  6. 严格返回以下判定之一：
    VERIFIED
    、
    NOT VERIFIED
    或 
    INCONCLUSIVE
    。

Local surfaces in this app

本应用中的本地验证入口

  • Code behaviour: a focused unit test (
    node --test test/unit/<file>
    ).
  • The page: 
    node scripts/verify.mjs
    (the verify-app skill): 
    snapshot
    , 
    screenshot
    , 
    network-summary --since-mark
    , 
    trace
    .
  • The API and the wire: 
    verify api <path>
    before and after.
  • What an agent writes: 
    node scripts/verify.mjs up --real
    and drive it, or the evals of hachej/boring-stack for the expert's skill.
  • Performance: 
    verify perf
    , or timings from the same machine for baseline and treatment.
  • 代码行为：聚焦的单元测试（
    node --test test/unit/<file>
    ）。
  • 页面：
    node scripts/verify.mjs
    （verify-app skill）：
    snapshot
    、
    screenshot
    、
    network-summary --since-mark
    、
    trace
    。
  • API 及其传输结果：在变更前后分别运行 
    verify api <path>
    。
  • Agent 写入的内容：运行 
    node scripts/verify.mjs up --real
    并进行操作，或使用 hachej/boring-stack 中针对 expert skill 的 evals。
  • 性能：运行 
    verify perf
    ，或在同一台机器上分别采集基线与处理组的计时数据。

Verdict Rules

判定规则

  • VERIFIED
    : baseline and treatment differ in the predicted direction, by the claimed threshold, with no obvious confound.
  • NOT VERIFIED
    : the behaviour is unchanged, moves the wrong way, or misses the threshold.
  • INCONCLUSIVE
    : no valid baseline, noisy signal (agent outputs on 3 cases are noisy), failed measurement, or an environment difference invalidates the comparison.
  • VERIFIED
    ：基线与处理组在预期方向上存在差异，差异达到声明的阈值，且不存在明显的混杂因素。
  • NOT VERIFIED
    ：行为没有变化、朝错误方向变化，或未达到阈值。
  • INCONCLUSIVE
    ：缺少有效基线、信号存在噪声（Agent 在 3 个用例上的输出噪声较大）、测量失败，或环境差异导致比较无效。

Output

输出

text
VERIFIED | NOT VERIFIED | INCONCLUSIVE
Claim: <falsifiable claim>

Evidence:
<metric/artifact>: baseline=<...>, treatment=<...>, delta=<...>, threshold=<...>

Reasoning:
<one tight paragraph naming the evidence and any confounds>
Do not soften a negative result. A clear 
NOT VERIFIED
is useful. Never put real user data in an artifact you commit.
text
VERIFIED | NOT VERIFIED | INCONCLUSIVE
Claim: <falsifiable claim>

Evidence:
<metric/artifact>: baseline=<...>, treatment=<...>, delta=<...>, threshold=<...>

Reasoning:
<one tight paragraph naming the evidence and any confounds>
不要弱化负面结果。明确的 
NOT VERIFIED
同样有价值。绝不要将真实用户数据放入要 commit 的产物中。