Compare omissions, numbers and readability measures

Type: Build Language: TypeScript Stage: 3 of 4 Time: ~2 hours Prerequisites: Basic TypeScript, JSON objects, arrays and running Node 22.18 or newer. Complete the preceding stages first.

What you build

Compare observable editing changes and surface recorded fact violations.

Public contract

measures(text:string):{words,sentences,averageWords,longWords}; compareParagraphs(paragraphs,proposals,definitions?):Comparison[]

Require exactly one proposal per paragraph. Count Unicode word tokens, punctuation-separated sentences, mean words/sentence rounded to 2 decimals and words longer than 8 characters. Issues report changed numeric multiset, missing exact protected term, and missing case-insensitive required definition phrase. Return original/candidate, before/after measures and issues.

The initial workspace contains public types, function stubs, the CLI, and original input fixtures. Change main.ts; the CLI is already supplied. Stages accumulate without replacing your earlier implementation. Tests import the learner workspace selected by PROJECT_WORKSPACE.

Worked example

The sample second proposal changes 15% to 10% and drops the exact required definition phrase. Both issues stay visible even though the sentence-length measure decreases.

Change the figure controls, predict the intermediate values, and compare the calculation with your implementation. The figure explains the mechanism; the grader runs the real functions.

Implement and verify

python3 scripts/project_test.py plain-language-rewrite-desk --init my-plain-language-rewrite-desk
python3 scripts/project_test.py plain-language-rewrite-desk --stage 3 --path my-plain-language-rewrite-desk --strict

Initialize once. The fresh first stage intentionally fails with a named not-implemented error. Preserve the signatures and reject malformed inputs with useful errors. Keep inputs unchanged. Each stage includes separate held-out inputs; passing the demonstration alone is insufficient.

Failure cases and hints

A lower average is not proof of comprehension. A numeric literal changed from 15% to 10% must block acceptance even if the prose looks simpler.

Use your result

After stage four, pass a JSON file of your own through the same entry point:

cd my-plain-language-rewrite-desk
node cli.ts sample.json output

Accepted Markdown, a paragraph-level HTML comparison and revision-decisions.json. Open output/report.html and inspect the machine-readable companion files.

Extend it

Add a human meaning checklist without pretending lexical checks cover every omission.

Scope: The baseline uses deterministic phrase substitutions and recorded author proposals. No live model adapter is shipped. Numeric, protected-term and definition checks are lexical safeguards, not a proof that every fact or nuance is preserved. The author reviews meaning.