A website asks every reader to provide some of its infrastructure: a device, a connection, electricity and, often, a prepaid data balance. Those demands are minor for some people and decisive for others. They belong in the design.
In 2024, only 19% of people in Madagascar used the internet. Across low-income economies, an entry-level mobile broadband plan takes about 22 times as much of the average income as it does in high-income economies. A study spanning 56 cities found that web pages could cost orders of magnitude more and load four times more slowly in disadvantaged places.
This site therefore treats transferred data as an accessibility budget. The aim is not to offer a diminished version of the web. It is to put the research ahead of its container.
This page, measured
The complete, immediately readable document is 10.8 kB. A first visit that also fetches the serif’s roman and italic font files is 40.6 kB.
These are compressed production-build figures, not the size of the source files. The first number includes every word, the layout and the full stylesheet. The second adds the two font files a browser may request on a cold visit.
| Part of the visit | Transfer | What arrives |
|---|---|---|
| Essential document | 10.8 kB | Text, structure and inline CSS |
| Optional serif type | 29.8 kB | Two cached WOFF2 files |
| First visit with both fonts | 40.6 kB | The complete designed page |
| Later visit with fonts cached | 10.8 kB | The document only |
The 10 KB Club collects noteworthy homepages below a 10 KB compressed-transfer budget. This document narrowly misses that mark, and the full cold visit is larger because of the optional type. That distinction matters. A small HTML file that quietly calls several megabytes of scripts and images is not a small page.
The closest zero-download alternative is the site’s system-serif stack, which can be selected without changing the templates. It cannot preserve the same appearance consistently across operating systems, however. The current compromise keeps Instrument Serif optional and limits it to two Latin subsets totalling 29.8 kB; readers on constrained connections see the metric-matched local fallback without waiting for either file.
For scale, the 2024 HTTP Archive found that the median mobile page transferred just under 2 MB. Its 90th-percentile page reached about 7.2 MB. This page’s full first-visit payload is roughly 50 times smaller than that median.
What a visit costs
Madagascar has competing mobile brands, not a single mobile monopoly. A World Bank assessment nonetheless describes a highly concentrated market dominated by Telma—now Yas—Orange and Airtel. Telma also controls the national fibre backbone.
The prices below are ordinary 30-day prepaid bundles published by Yas and Airtel. Both cost 15,000 ariary, which makes the difference in data allowances easy to see. The figures are the portion of an already-purchased bundle used by one visit; neither company bills a separate fee for opening a page.
| Network and published bundle | Document only | First visit with fonts |
|---|---|---|
| Yas · Net Month 15000 · 2.5 GB for 15,000 Ar | 0.062 Ar | 0.241 Ar |
| Airtel · Net Mlay-15000 · 6 GB for 15,000 Ar | 0.026 Ar | 0.100 Ar |
A 2 MB page, close to the present mobile-web median, would use about 12 Ar of the same Yas bundle or 5 Ar of the Airtel bundle. One page view is still a small expense. Page after page, the design of the web decides how quickly a balance is spent.
Orange also sells mobile data in Madagascar, but its public page sends current prices through a separate purchase service. It is omitted rather than filled with a figure that cannot be checked openly. Prices here were checked on 28 July 2026 and will change.
What the site does with its budget
- There is no JavaScript, advertising, analytics or tracking.
- The stylesheet travels inside the document, so readable text does not wait for another request.
- Body copy uses the typeface already installed on the reader’s device.
- The serif used for headings is optional. Its fallback is matched to the same proportions, which prevents the page from jumping if the font is skipped or arrives late.
- Images elsewhere on the site are resized for the screen and supplied as AVIF or WebP. Images below the first one load later; an article can keep non-essential plates closed until a reader asks for them.
- Links to datasets and papers show the file type and size before the download.
- Pages are built as static files. They require little server work and remain straightforward to archive.
Restraint has costs of its own. Static publishing can move technical work from a large platform onto researchers and maintainers. The digital-humanities literature calls this out directly: minimizing computing can maximize labour. This site keeps its machinery documented and small enough for another person to understand, but it does not pretend that maintenance is free.
What the numbers cannot say
Transferred bytes are measured. Carbon per page view is modelled.
The widely used Sustainable Web Design Model starts with data transfer and estimates a share of energy used by data centres, networks and devices. Its authors describe bytes as a proxy. The relationship is not linear in the physical network: routers and mobile towers consume power while waiting for traffic, and infrastructure is built for peaks rather than for one page at a time. Research on data-transfer energy intensity warns against claiming that a given reduction in bytes prevents an equal percentage of emissions.
For that reason, this page does not turn its transfer size into a precise number of grams of carbon. Smaller pages reduce demand on storage, networks and devices; they also load sooner, work on older hardware and leave more of a reader’s data allowance intact. Those are useful, verifiable outcomes. They are not proof that one page is sustainable in isolation.
The broader standard is the draft W3C Web Sustainability Guidelines, which treats sustainability as a question of people, planet and prosperity. It includes accessibility, maintainability, hosting, organizational practice and material waste alongside data transfer. That is the standard this page is trying to approach.
Notes and sources
Sizes were measured from the minified production build. The HTML was compressed
with gzip -9 -n; -n removes the filename and timestamp from the gzip header.
The figures count response bodies and exclude protocol headers, encryption
overhead and a browser’s possible request for the site icon. Bundle calculations
treat 1 GB as 1,000,000,000 bytes:
transferred bytes ÷ bundle bytes × bundle price = share of bundle
Further reading:
- The 10 KB Club discussion on Hacker News, including the limits of page weight as a test of speed or quality.
- Towards a World Wide Web without digital inequality, a 2023 peer-reviewed study of cost, performance and low-end devices.
- A Framework for Improving Web Affordability and Inclusiveness, based on more than 72,000 measured pages.
- How to Build a Low-tech Website, a design precedent that makes infrastructure and compromise visible.
- The Questions of Minimal Computing, which asks what a project needs, has, prioritizes and is willing to give up.
- Platitudes: The Carbon Weight of the Post-Platform Scholarly Web, a recent case study of lightweight scholarly publishing, preservation and labour.
Plan sources: Yas mobile internet, Airtel data bundles and Orange mobile data.