This repository contains the A01 random lunch generator, its repair variants, the repaired generation prompt, and the checks used to reproduce the reported results.
Live page: https://sergey-cv.com/random-lunch-generator/
Repository: https://github.com/Fighter90/random-lunch-generator
The starter page displayed a dish name on every click but could display no icon for some dishes. Three Font Awesome 6.4.0 class names were absent from the stylesheet loaded by the page: fa-bowl-hot, fa-pasta, and fa-bowl. They affected Ramen, Pasta, and Soup, respectively, so 3 of 12 entries had missing glyphs and the expected failure rate was 25%. Tacos used fa-utensil-spoon, which rendered but was semantically wrong. Finally, every click created an independent 500 ms timer, so rapid clicks rendered several stale results in sequence.
The repair keeps the original page structure and Font Awesome rendering. The fixed variant verifies the missing glyphs at runtime and falls back to fas fa-utensils; the regenerated/deployed variant additionally uses relevant verified replacements and cancels the prior pending result on a new click. The deployed index.html is byte-identical to index_regenerated.html.
| File | Where it is used |
|---|---|
.gitignore |
Git working-tree exclusion rules. |
README.md |
This artifact inventory and reproduction guide. |
OPENCODE_WALKTHROUGH.md |
Project walkthrough and investigation notes. |
requirements.txt |
Python test dependency pin, used before running Playwright checks. |
prompt.md |
Original request and generated README prompt; used as the prompt-root-cause reference. |
prompt_fixed.md |
Analogue of the original prompt.md: preserves the original README dialogue and adds the missing verified single-file code request and answer. |
index_original.html |
Untouched starter page; baseline for the three defects. |
index_guard_only.html |
Starter mappings plus runtime glyph fallback; isolates the guard behavior. |
index_fixed.html |
Fixed page with the three missing mappings replaced, distinct Ramen/Soup bowl icons, and the runtime fallback retained. |
index.html |
Deployed GitHub Pages entry point; byte-identical to index_regenerated.html. |
index_regenerated.html |
Page regenerated from the dialogue-form code request in prompt_fixed.md; uses verified mappings and cancels stale timers. |
check_icon_classes.py |
Downloads or reads all.min.css and checks exact glyph rules for every menu icon. |
list_dish_icons.py |
Prints dish-to-icon mappings side by side for the semantic review. |
run_experiment.py |
Runs seeded headless-browser clicks and measures CSS ::before glyph rendering. |
rapid_click_check.py |
Performs five fast clicks and records how many delayed result names appear. |
figures/fig1_failure_mechanism.png |
Failure-mechanism diagram used in the report. |
figures/fig2_left_original_ramen.png |
Baseline Ramen screenshot showing the missing icon. |
figures/fig2_right_deployed_ramen.png |
Fixed/deployed Ramen screenshot showing the icon. |
results/index_original_seed42_n100.json |
Baseline 100-click seeded observations. |
results/index_original_seed7_n300.json |
Baseline 300-click seeded observations. |
results/index_guard_only_seed42_n100.json |
Guard-only observations and fallback-triggering rows. |
results/index_fixed_seed42_n100.json |
Fixed-page 100-click observations. |
results/index_regenerated_seed42_n100.json |
Regenerated-page 100-click observations. |
results/live-pages_seed42_n100.json |
Live deployed-page 100-click observations. |
Set up the Python environment and Playwright browser:
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
playwright install chromiumThe baseline 100-click result is 30 failures out of 100:
python3 run_experiment.py index_original.html 100 42 baseline
# icon failures: 30/100 = 0.300The 300-click baseline is 67 failures out of 300:
python3 run_experiment.py index_original.html 300 7 baseline-300
# icon failures: 67/300 = 0.223The guard-only page renders a fallback for all missing glyphs and reports zero visible failures:
python3 run_experiment.py index_guard_only.html 100 42 guard-only
# icon failures: 0/100 = 0.000
# console warnings include the missing Pasta, Ramen, and Soup classesThe fixed page reports zero failures:
python3 run_experiment.py index_fixed.html 100 42 fixed
# icon failures: 0/100 = 0.000The deployed entry point is index.html, which is the regenerated page:
python3 run_experiment.py index.html 100 42 deployed
# icon failures: 0/100 = 0.000The regenerated page also reports zero failures:
python3 run_experiment.py index_regenerated.html 100 42 regen
# icon failures: 0/100 = 0.000Check the missing classes in the original and the repaired pages. The original reports 3 missing classes; the repaired pages report 0:
python3 check_icon_classes.py index_original.html
python3 check_icon_classes.py index_fixed.html
python3 check_icon_classes.py index_regenerated.htmlCompare every dish mapping and see the semantic changes:
python3 list_dish_icons.py index_original.html index_fixed.html index_regenerated.htmlThe expected mapping changes are Tacos: fa-utensil-spoon -> fa-pepper-hot, Ramen: fa-bowl-hot -> fa-bowl-rice, Pasta: fa-pasta -> fa-wheat-awn, and Soup: fa-bowl -> fa-bowl-food.
Check the rapid-click race. The original emits five delayed names; the regenerated page emits only the final result:
python3 rapid_click_check.py index_original.html index_regenerated.html
# original: five names
# regenerated: ['Sandwich']For the exact Font Awesome 6.4.0 source check:
curl -sSL -o fa.css https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.4.0/css/all.min.css
for c in pizza-slice fish hamburger leaf utensil-spoon bowl-hot bread-slice pasta mortar-pestle drumstick-bite bowl fire bowl-food wheat-awn pepper-hot bacon; do
echo "$c: $(grep -o "\\.fa-$c[:,{]" fa.css | wc -l)"
doneThe exact missing original names are bowl-hot, pasta, and bowl; the replacement names used by the repaired pages have glyph rules in the same pinned stylesheet.
The original project is licensed under the MIT License.
