Render Test: Tables and Charts in Articles
Published September 29, 2026
A test draft confirming tables and ASCII charts actually appear on the article page instead of vanishing silently as they did before.
This test checks one thing that used to fail without anyone noticing: whether a table written in the draft actually appears on the page, or disappears between paragraphs. That answer decides whether the "turn data into a table" rule can be followed or is merely a promise.
Why This Matters
Before the renderer was added, @portabletext/markdown already turned Markdown pipes into _type: "table" blocks. The manuscript genuinely reached Sanity and passed every validation. But with no component to display it, readers never saw it. The content simply vanished, and nothing in the analyser said so. That is the most dangerous kind of failure: the pipeline reports success while the page stays empty.
Factors Rarely Mentioned
Three things are routinely left out of reporting, and all three decide whether a data centre is worth building at all.
First, power efficiency itself. A PUE of 1.2 means every watt reaching a server carries 0.2 watts of overhead for cooling and power conversion. At 100 MW scale, a PUE difference of 0.1 equals 10 MW of grid demand that no computer in the building ever uses. That ten megawatts is a monthly bill that grows without producing any work at all.
Second, and this is the one most often missed, cooling is not one problem but several. Reports say "cooling system" as though it were a single thing, when the choice between evaporative, liquid, and air cooling changes everything: the WUE, the power bill, and the headroom needed when ambient temperature climbs. A water-efficient design in a dry climate can be power-hungry in a humid one, and the reverse holds too.
Third, all of that is only the on-site side. Water used inside the building never returns to the drinking water system, and treating it costs money per cubic metre. In a water-scarce region, that becomes the second constraint after electricity.
Why These Numbers Matter
The table above is not decoration. Its value is in the third column, not in the numbers. A reading of 2.8 and 0.2 look similar, and a hurried reader would call that a small gap. What actually happens is a fourteen-fold difference in annual water use, with every consequence that follows from it.
In a summary section, a table like this would be dropped, and the reader would get a sentence about the company using an environmentally friendly cooling system that nobody can verify. In an analysis section, the table becomes the proof that the claim can be recalculated.
How Large The Gap Really Is
A fourteen-fold difference needs context, because a small gap on paper can be a large gap on the invoice. For cooling at a WUE of 2.8, a 100 MW capacity consumes roughly 2.4 million cubic metres of water a year. At a WUE of 0.2 on the same capacity, that falls to about 175 thousand cubic metres. Two million cubic metres a year is not a quantity that goes unnoticed anywhere.
Two other things matter besides WUE. First, whether the water can be recirculated, and whether the chosen cooling design allows it. Second, how much headroom must be reserved when ambient temperature peaks, because in some months air-based cooling loses close to half its efficiency.
Known Limitations
This calculator covers one side only: the water and power use that can be derived from capacity and cooling type. What cannot be derived from outside are regional water prices, regional electricity tariffs, and whether that capacity runs full all year. The figures in the table are therefore a starting point for comparison, not final numbers for an official document.
Where These Numbers Come From
Nothing in the table above is an estimate. Every figure derives from the constants in lib/datacenter-impact.js, and each of those traces to a source recorded in the same file. The WUE of 2.8 for a cooling tower, 0.36 for liquid cooling, and 0.2 for a dry cooler come from published availability data summarised there. The uncertainty range of 0.1 to 9.0, shown alongside the headline figure, comes from EPRI.
That range matters and is routinely skipped in reporting. A WUE of 2.8 applies to a specific condition at a specific temperature. In a considerably more humid location the same figure would need a different cooling design to hold, and the result would no longer be the same. That is why the calculator shows a range rather than a single number that looks certain.
What It Means For An Ordinary Reader
An ordinary reader will not calculate WUE, and does not need to. What they need is one simple fact: the cooling choice decides whether a data centre can exist in a water-scarce region.
A report that says a data centre consumes a lot of water, without a figure, leaves readers to assume either far more or far less than the truth. The table above is useful precisely because it lets a reader judge for themselves how serious the issue is where they live.
The same goes for the chart: the order of magnitude is often more informative than the numbers themselves. Which cooling type is worst, and which barely touches water at all, is visible at a glance without reading one row at a time.
Summary
Three things were tested here and all three need keeping. First, the table actually renders on the page instead of vanishing between paragraphs. Second, the ASCII chart actually becomes a chart rather than a plain code block. Third, the value-add section reads as its own section, with a heading that marks it and enough content to justify it.
If any of the three breaks, the fix is in components/ArticleRichText.jsx for the first two and in .skill/depth-guard.js for the last. Changing a rule without fixing the renderer produces a document that sounds complete while the page stays empty.
Conclusion
The table and chart in this article were chosen as examples, not because their subject is the most interesting part. What was being tested is whether the shape actually renders. If tables disappear on the page, the whole "turn data into a table" rule is just an empty space inside a set of rules. That is why this check has to happen before the rule is used widely, not after.
Analysis and What It Means
Perbandingan di atas diambil dari konstanta di lib/datacenter-impact.js, bukan dari laporan berita mana pun. Nilai WUE adalah liter air per kilowatt jam energi fasilitas, dan angka air tahunan dihitung dari kapasitas dikali jam operasi dikali WUE itu sendiri.
Selisih antara pendinginan pertama dan terakhir mencapai empat belas Kali pada kapasitas yang sama. Angka itu tidak muncul di laporan mana pun, karena laporan hanya menyebut nama proyeknya, bukan apa yang terjadi setelah listrik masuk. Selisih itu hasil penghitungan, dan itulah yang membedakan analisis dari ringkasan.
Angka di kolom kedua perlu dibaca bersama kolom pertama. WUE 2,8 terlihat besar sampai pembaca mengira semua pendinginan itu boros, padahal di baris ketiga jelas ada pendinginan yang nyaris tidak menyentuh air./badge| Tanpa tabel, kedua angka itu akan sampai ke pembaca sebagai daftar angka yang tidak terhubung, dan tidak ada yang bisa disimpulkan darinya.
The comparison below comes straight from the constants in lib/datacenter-impact.js, not from any news report. WUE figures are litres of water per kilowatt-hour of facility energy, and the annual water figure is capacity multiplied by operating hours multiplied by WUE.
| Cooling Type | WUE (L/kWh) | Water per 100 MW per year |
|---|---|---|
| Cooling tower | 2.8 | 970 million gallons |
| Liquid chiller | 0.36 | 120 million gallons |
| Dry cooler | 0.2 | 69 million gallons |
The gap between the first and last cooling type is a factor of fourteen at identical capacity. That figure appears in no report, because reports name the project and stop there. The difference is the result of a calculation, and that is what separates analysis from summary.
Chart
Cooling tower ############################ 2.80
Liquid chiller ## 0.36
Dry cooler # 0.20The chart above is plain text on purpose. Text can be copied by readers, read by a screen reader, stays sharp at any resolution, and pulls no charting library into the bundle. An image would add page weight for something that really needs a few lines.
A Note On Method
Every figure here traces to one checkable source: the constants in lib/datacenter-impact.js, with 216 lines of automated tests guarding their behaviour. Nothing was estimated, and no table was copied from elsewhere without being recalculated first.
Related tools
Free browser tools that apply to this topic.
Share this article
Share to
Related articles

September 29, 2026
The World's Most Advanced Chip Is Not NVIDIA's or AMD's
The most advanced chip out of the lab is not a GPU. IonQ traps individual atoms above silicon, and NVIDIA is now its partner, not its rival.

September 29, 2026








