Voice guide

How DataPressr data stories should sound. Use this during the prose pass, after the outline and charts have been checked. A final voice pass by the author remains a separate step.

The one rule

Let the numbers carry it. The reader came for what the data says. Your job is to state it clearly and get out of the way. If a sentence would survive having its adjective deleted, delete the adjective.

Do

• Lead with the finding. First chart, then one line saying what it shows. Never open on the source file, the download, or the wrangling.
• Short declarative sentences. "It has risen every year. The annual mean has not once fallen in 67 years."
• Concrete numbers with units, in the prose, not just the chart: "316 ppm in 1959. 427 ppm in 2025."
• Name the people and places. "Charles David Keeling began measuring… 3,400 metres up a volcano in Hawaii." Specific beats grand.
• Attribute plainly. "Data: co2-ppm, from NOAA." A link, not a paragraph.
• Put the method last and short, framed as "how this was made" for the reader who wants it — "here's the cake, now the recipe."
• Admit what the data doesn't say. One honest caveat builds more trust than three confident claims.
• British spelling (colour, metres, -ise), consistent with the repo.

Don't

• No editorialising adjectives: relentless, alarming, staggering, dramatic, unprecedented. The outline calls these out by name.
• No metaphor doing an argument's job: not "the planet's breathing", not "the most important line in climate science". A metaphor can label a real pattern (the "sawtooth") — it can't stand in for a number.
• No throat-clearing openings: "In today's world…", "It is well known that…", "Data is the new oil…".
• No hedge stacks: "it seems that this might perhaps suggest". Say it or don't.
• No exclamation marks. No rhetorical questions. No second person ("imagine you…").
• Don't bury the finding under context. Context supports the finding; it doesn't precede it at length.
• Don't oversell the dataset. "A tidy version of X" is enough; it doesn't need to be "definitive" or "comprehensive".

Before / after

Before. In an age of mounting climate anxiety, perhaps no dataset is as quietly terrifying as the relentless, unbroken march of atmospheric carbon dioxide — a grim testament to humanity's failure to act.

After. Atmospheric CO₂ at Mauna Loa has risen every year since measurements began in 1958. The annual mean has never fallen.

---

Before. Let's dive into the messy world of NOAA's data files and see if we can wrangle them into shape!

After. The NOAA source is a text file with about 40 comment lines and -1 for "no measurement". build.ts turns it into two typed CSVs.

---

Before. This groundbreaking dataset represents the definitive, most comprehensive collection of oil price data ever assembled.

After. Brent and WTI spot prices from the EIA, daily to annual, in one tidy package.

Length

A story is as long as its argument, usually 300–700 words of prose around one or two charts. If it runs longer, the argument probably has two stories in it.

Structure the voice assumes

The outline → viz plan → prose separation is what makes this voice cheap to hold: the argument is fixed in the outline, so the prose pass is only ever about wording, and a later voice pass can rewrite wording without touching the argument.