Toolkit to measure difference in web vitals when <img> is server-side rendered in every <amp-img>.

If you have a large set of URLs, then simply run:

npm install
node filter.js my_list_of_urls.txt > fetch_all.sh
chmod +x fetch_all.sh
./fetch_all.sh
./benchmark.sh

Otherwise, see individual steps.

npm install

A text file containing URLs is filtered and outputs a shell script:

node filter.js my_list_of_urls.txt > fetch_all.sh
chmod +x fetch_all.sh # make runnable

1. Takes a list of URLs of AMP documents, or documents that link to an AMP version
2. Filters them randomly*
3. Resolves AMP version if linked
4. Prints as shell commands, to output into script (> fetch_all.sh)

# file is written like:
node fetch.js `printf "%q" "https://paired-amp/amp/url.html"`
node fetch.js `printf "%q" "https://amp-first/url.html"`
# etc.

* optional additional argument for inclusion probability:

node filter.js my_list_of_urls.txt 0.000045 > fetch_all.sh

use 1 to include all urls:

node filter.js my_list_of_urls.txt 1 > fetch_all.sh

To transform the previously filtered URLs, execute output script:

./fetch_all.sh

Otherwise, run command for an individual URL:

node fetch.js "https://some-url-probably-amp.com/some/path"

...which inserts into output/:

output/
  some--url--probably--amp-com/
    url.txt                    # full url
    amp-img__unoptimized.html  # as-is
    amp-img.html               # optimized
    img.html                   # optimized, using native <img>

amp-img.html and img.html are transformed like a cache would by using the AMP optimizer.

Use tachometer to measure for LCP and CLS on documents transformed previously:

./measure.sh

This compares metrics between optimized output documents:

• output/**/amp-img.html
• output/**/img.html

Results are output as result-* files:

output/
  some--url--probably--amp-com/
    ...
    result-lcp.json            # LCP: full results
    result-lcp.csv             # LCP: summary table
    result-cls.json            # CLS: full results
    result-cls.csv             # CLS: summary table

LCP and CLS are measured using a PerformanceObserver, see inline-perf-script.js to see how results are measured.