Final design: human-sounding wording plan (2026-10-30)

Decision: Design 5 (Operations) is the final design. The index on port 8116 lists it as "Final design / development design". It is a copy of Design 5 in final/, built from content-final.mjs.

Problem: The site wording in content2.mjs sounds AI-written. Mike wants it to sound like a person talking, and to be affirmative.

Steps

# Who What
1 Claude Copy Design 5 to final/. Add it to the index as "Final design / development design".
2 Claude copy-lab/extract.mjs dumps all 14 wording sections of content2.mjs into copy-lab/candidates.json as the baseline.
3 Opus Writes 3 voices for every section (round-1.json): A plain talk "we", B personal "I", C short and direct "we". Rules: affirmative, no corporate words, no AI patterns, no invented facts, no vendor names.
4 Sonnet copy-lab/apply.mjs writes the winning wording into content-final.mjs. The existing Jev harness (website/copy-lab/jev_copy.py, run with JEV_LAB) does the scoring.
5 Jev Scores the baseline and each variant twice on 12 questions.
6 Claude Opus rewrites weak sections (round 2 and later) until every section passes. Then rebuild final/ and re-check.

Pass rule (per section)

Run it

cd "C:\Claude_Working\AI Concierge\website\overhaul\copy-lab"
$env:JEV_LAB = (Get-Location).Path
python ..\..\copy-lab\jev_copy.py merge round-1.json
python ..\..\copy-lab\jev_copy.py score
python ..\..\copy-lab\jev_copy.py report     # presentation.html
node apply.mjs                                # writes ..\content-final.mjs
node ..\final\build.mjs                       # rebuilds final/

Result (2026-10-30)