c_Learn & Practice AILEARN/“Done” needs evidence
← Beginner guide
GUIDED PRACTICE

“Done” needs evidence

Check requirements, diff, scripts, and browser behavior separately. Produce an acceptance report that distinguishes a confirmed result from a check that was never run.

Before you start

Use the working copy from the bug-fix lesson, with its original requirements and verification script. Keep the initial failing result and the diff available.

Download this practice kit · ZIPWatch the V2 pilot video →
01

WHAT · What you will learn

The deliverable is a short acceptance report, not another code change. It names the tested file, the expected behavior, commands and actual outcomes, browser checks, and unresolved limitations.

02

WHY · Why this matters

A syntax check can pass while a link still goes to the wrong section. A test can pass against a reference file while the working file is broken. Each check answers a different question; none should be silently substituted for another.

03

HOW · Follow the steps

  1. Read REQUIREMENTS and VERIFY in your language before inspecting the answer. Write the acceptance target: Contact reaches #contact; neighboring navigation and content stay unchanged.
  2. Inspect the diff and name the file being checked. An unchanged test script matters because it preserves the original check.
  3. Run syntax and structural checks, recording each exit code separately. Then perform the mouse, keyboard, and narrow-screen checks on the working HTML.
  4. Summarize PASS, FAIL, and NOT RUN. If you cannot use a browser or a tool, name the missing check and avoid an unconditional completion claim.
Review the current practice fix without editing files.
Read REQUIREMENTS.en.md and VERIFY.en.md. Inspect git diff -- index.en.html.
Run node --check verify.mjs and node verify.mjs en. Report each actual exit code.
If a browser is available, check Contact with mouse and keyboard, and at 390px width.
Separate PASS, FAIL, and NOT RUN. A test pass does not prove browser behavior.
Do not change files or tests, install tools, commit, push, or deploy.
04

EXAMPLE · A worked example

For this static lab, node --check verify.mjs checks the script syntax. node verify.mjs en checks the HTML structure and anchor targets. Neither command sends a message, deploys the page, or proves keyboard behavior.

A report template is shown below. Replace each placeholder with your observed evidence. This lab has no application build pipeline, dedicated lint command, or type-check command; mark those as not configured instead of inventing successful runs.

File: [working path]
Requirement: Contact reaches #contact
Diff: [what changed / what stayed unchanged]
Syntax: [command, exit code]
Structure: [command, exit code]
Mouse + keyboard: [observed result]
390px: [observed result]
Not run / not configured: [name each check]
Decision: [accept / fix / still unverified]
05

TRY IT · Do it yourself

  1. In a fresh copy, intentionally break only the back-to-top target and run verification. Check whether your report identifies the actual failure instead of repeating the previous pass.
  2. Write one report with the browser unavailable. It should explicitly leave the mouse, keyboard, and visual checks as NOT RUN.
06

VERIFY · Check the evidence

  1. Every pass is tied to a command output or an observed browser action on the working file.
  2. The report names the one-line repair and confirms that the requirements and check script were not weakened.
  3. Missing tools and untested environments remain visible. “Not configured” is different from “ran and passed.”
  4. The next action follows the evidence: accept the bounded exercise, fix a specific failure, or continue a named missing check.
07

COMMON ERRORS · Recognize the failure

  1. Treating exit code 0 from a syntax check as proof that the page works.
  2. Reporting mobile or Safari as tested after checking only a desktop browser.
  3. Running commands in the wrong folder and accepting an unrelated passing result.
08

FIX · Recover deliberately

  1. Wrong environment: identify the actual working directory, reopen the intended file, and repeat the named check. Preserve both outputs to explain the correction.
  2. Unknown command: inspect the provided scripts and instructions; do not invent npm commands or install a framework just to make a checklist look complete.
  3. Failed check: report the exact expectation and observation, ask for a scoped repair, then rerun the failed check and its neighboring behavior.
09

NEXT · Continue when ready

Use this acceptance pattern on the next project. Add checks because they protect a real behavior, not because a longer command list looks more professional.

Sources and scope

Independent practice content. Product guidance reviewed on 16 September 2026. Reference HTML and expected outcomes are distinguished from actual recorded runs.

OpenAI · Code review ↗Git · git diff ↗
Your progress
Esc · Close