Live from Brussels — design memo
A rebuild. The first game from this paper, Say It Plainly (an editor’s challenge with two opening bets), is kept in previous/. This one starts again from the PDF.
1. Paper
Rauh, C. (2023). Clear messages to the European public? The language of European Commission press releases 1985–2020. Journal of European Integration 45(4): 683–701. doi:10.1080/07036337.2022.2134860. Open access (CC BY).
Available: the PDF only. The replication corpus on Harvard Dataverse (doi:10.7910/DVN/UGGXUF) could not be reached from the build environment (the proxy refused it), so no real press release appears in the game. Every number is read off Figure 2 at 600 dpi by pixel position (dot centres against the gridlines; about ±0.5 reading-ease points) and marked as such.
2. What the paper hands over
Primary: estimates. Mean values of three clarity measures for 44,978 Commission press releases (1985–2020), and the same for five comparison corpora: UK and Irish government press releases, political-science abstracts, and broadsheet and tabloid politics reporting (Fig. 2, upper panel). Also the Commission’s yearly averages (Fig. 2, lower panel), and the Commission-minus-national gaps within 20 topics (Fig. 3).
Secondary: a measurement. The three measures are defined exactly. Reading ease is the Flesch formula, built from words per sentence and syllables per word. Familiar words is how often the words used appear in Google Books. Focus on action is verbs as a share of verbs plus nouns (note 5). The first one can be computed by hand on any text, so it is the one a player can feel.
Argument (context, not played). Clarity matters for reaching a politicised public (pp. 686–7). The Commission faces mixed incentives (pp. 687–9). Three explanations are offered for future research (pp. 695–6).
3. The discovery, and its kind
A counterintuitive result, made physical. Over 35 years the Commission gained power, became a target of public anger, and kept promising to explain itself better. Its press releases did not get easier. They sit between political scientists’ abstracts and the national governments, far from any newspaper. That holds on every topic, and the yearly line is flat from 1985 to 2020; only word choice became a little more familiar. What makes a text score this way is mainly long words, and somewhat long sentences.
4. The core narrative
There is an office in Brussels that writes news announcements for everyone in Europe. Over thirty-five years it got more powerful, people began to argue about it, and it kept promising to explain itself better. A researcher measured how hard its announcements are to read, and compared them with announcements from governments, with newspaper stories and with professors writing for other professors. Brussels sounds almost like the professors. And the announcements never got any easier.
5. Who lives in the paper
- The Commission’s press officer writes the release. Controls the words. (The previous game’s player.)
- The Commission as an institution chooses how plainly to speak, between member governments and a restless public. Controls a style. The paper offers no payoff for this choice.
- The journalist at the midday briefing relays it: “half my work is to translate it into normal words” (p. 686). Controls what gets passed on.
- The intended receiver, the European citizen the guidelines say releases are written for (p. 689). Controls nothing but their own reading. The reading-ease score is a model of this person’s effort.
- The researcher measures. Controls which corpus to compare against.
- The text itself: its words, syllables and sentences.
6. Precedent
The last three games in the table, apart from this paper’s own:
| Game |
Position |
Format |
Feedback device |
Chassis |
| The Original or the Copy |
forecaster, then controller |
challenge (bets, then two dials) |
the reveal; the race |
newsprint: off-white ground, black ink, system bold sans, party red-brown and blue |
| The Honest Billboard |
agent (campaign designer) |
challenge |
the reveal (billboard flips to a histogram) |
white billboard panels on red-earth ground; green/grey/red |
| One of Us |
experimenter |
discovery |
gap meter; up/down bars |
ink on cream review form; one orange accent |
| (previous Rauh: Say It Plainly) |
agent (press officer), opened by bets |
challenge |
drop, trajectory, counter on a ruler |
press-release serif on paper; the paper’s blue/yellow/green |
This game takes a different value on all four. Position: subject. The player is the reader, reading aloud, with the paper’s measure working on their own thumbs. Format: arcade, a timed skill run. It is none of the three named formats, chosen deliberately to step away from two challenges and a discovery. Device: one of our own, the drum, where every syllable is a tap, plus a fill (lit windows). Chassis: a TV studio at night: black, prompter white, on-air red, tally amber, an embedded heavy display face and a mono for timecodes.
7. The aim and the verdict
Aim (start screen, top of the play screen, first line of How to play): “Read the news on air. Tap once for every syllable — get every word out.”
Yardstick: the share of the script’s words you got out cleanly, shown as lit windows in a block of flats at night. That is the aim’s own unit, for the item just read.
Verdict (end screen, first lines):
- What you did: “The Brussels script tied your thumbs in knots, almost as badly as the professor’s. The {year} script you picked was no easier.” (Or, for a strong player: “You got through Brussels, but you had to drum twice as fast as for the tabloid.”)
- What the paper says: “Rauh measured 44,978 real Commission press releases, 1985–2020. They are harder to read than the UK and Irish governments’ releases, far harder than any newspaper, nearer to professors writing for professors — and they never got easier.”
- What that means: “Brussels kept promising to explain itself better. On the page, it didn’t.”
8. Three concepts
- Live from Brussels (picked; the reckless one). Position: subject, the reader. Format: arcade. Control: one big drum pad. Tap once per syllable of the word on the prompter, one word per beat. A full stop gives a rest beat, and as a sentence runs on your breath runs low and the beat quickens. Before pressing: how many syllables is the word coming up, and can I fit five taps into this beat? After: the word lands lit (said) or smudged (mumbled), and the windows fill. Discovery: the Brussels script is a drum solo, and the 2020 one is no easier than the 1987 one. Fun: mastery (it is a skill, and two thumbs help), anticipation (the next word looms), agency (the result follows your thumbs), discovery (the archive pick). Replay: a better technique; a different archive year; the encore. Reckless because it is real-time, uses the body, and the play screen has one word on it. The builder half-expects it to feel arbitrary. The calibration plan has to show that it is not.
- The Spokesperson’s Desk (runner-up). Position: dilemma. The player walks through five moments the paper describes: Delors and Maastricht, the Santer resignation, Prodi’s communication push, Wallström’s Plan D, and Juncker’s “political Commission”. At each one they choose how the house should write (as before / plainer / much plainer), with the paper’s sources on what each president intended. Then the paper’s measured line for those years is drawn under their choices. Discovery: whatever the presidents said, the line did not move. Fun: anticipation, discovery. Weakness: the choices have no consequence in any model the paper supplies, so the player is narrating rather than playing.
- Two Masters. Position: agent, the Commission as an institution. Format: strategic. Member governments dislike clear stances, and the public demands them. Each year the player sets a clarity level and both audiences react. Rejected: both reactions would be ours. The paper’s three explanations (identity, internal bargaining, strategic caution) are offered as questions, not estimates, so any payoff would be an invention shown as a mechanism.
9. The first five minutes
Pick: Live from Brussels.
- 0:00 Black screen, red ON AIR lamp, “LIVE FROM BRUSSELS” in huge white type. The aim on its own line: tap once per syllable. One button: Go to the studio. I press it.
- 0:10 The rundown: tonight’s five items, with timecodes. A tabloid’s rail story, the UK government, Brussels, a professor, and something from the archive. All five are about the same thing, trains running late. I wonder why a professor is on the news.
- 0:20 Item 1, the tabloid. Two lines tell me: one word per beat, one tap for every dot under the word. Count-in 3-2-1. “RAIL” (one dot), tap. “PAS·SEN·GERS” (three dots), tap-tap-tap. The word turns amber. Easy, and a little silly; I’m grinning. A full stop gives me a breather. 30 seconds. Nineteen of twenty windows lit: “You got almost every word out.”
- 1:05 Item 2, the UK government. “AU·TO·MAT·IC·AL·LY”: six dots, one beat. I fumble it. Mostly fine. “About one word in ten came out as mush.”
- 1:45 Item 3, Brussels. The words keep coming in fours: COM·PEN·SA·TION, EN·TI·TLE·MENTS, RE·IN·FORCE·MENT. The sentence runs 24 words with no full stop. My breath gauge drains and the beat speeds up. I switch to two thumbs. Windows go dark. “About a third of it came out as mush.” Then, after the discovery, the line: “That’s what an average Commission press release scores. The study measured 44,978 of them.”
- 2:30 Item 4, the professor. Worse, but not by much. “The only group in the study harder to read than Brussels: professors writing for professors.”
- 3:10 The archive. “Since 1985 Brussels got more power, people got angrier, and it promised to explain itself better. Pick a year to read.” 1987, 2004, 2020. I pick 2020, because surely they learned. CO·RO·NA·VI·RUS. Same knots. The card shows the paper’s line for all 35 years: flat. 2020 was in fact the hardest of the three.
- 4:00 End: the verdict in three lines. Encore: read it the way a person would say it. I press it: the same news in plain words, and every window lights up. Then Play again: I want to beat my Brussels score with two thumbs.
Runner-up: The Spokesperson’s Desk.
- 0:00 1985, Delors’s office. “Lamy: ‘You had to get on without telling them too much.’” Three style buttons. I pick as before, because it’s 1985.
- 0:40 1992, the Danish No. I pick plainer. The line so far is drawn in.
- 1:20 1999, Santer resigns. Much plainer, surely.
- 2:00 2005, Plan D; 2014, Juncker. I keep pressing plainer. The paper’s line appears under my choices, flat.
- 2:40 I realise I never did anything. My choices drew one line, the paper drew another, and nothing connected them. The finding lands as a chart. It is a demonstration.
10. The pick, and why
Live from Brussels. It is the only concept in which the player does something whose result they can feel and get better at, and the thing they feel is the paper’s own measure. Flesch reading ease is, almost literally, how many syllables you must get through per word, and for how long without a stop. The player’s sentence afterwards is “I read the Brussels press release on air and my thumbs couldn’t keep up”, not “I guessed”. The discovery lands inside three minutes (Brussels) and again at four (the archive pick, a real commitment that the player then has to live with).
Author test: the author has spent years with these texts. Drumming his corpus, syllable by syllable, against a tabloid and a political-science abstract, is a joke he would enjoy, and the archive pick plays his flat line as a thumb-ache.
Runner-up: The Spokesperson’s Desk. Switch if testers find the drumming exhausting or unfair on phones with slow touch screens, or if the author wants the institutional history (Delors to Juncker) to carry the game. The history would then frame the drum game rather than replace it.
The pick is the reckless concept, so nothing reckless was given up. What Two Masters would have given: a reason for the flat line. The paper does not supply one, and inventing one is what this game must not do.
11. The shape
- Screens: start → rundown → four items (each: card, count-in, reading, result) → archive pick → archive reading → end. How to play and About are reachable from start and from a ? on every screen. Encore and Play again sit on the end screen.
- Items, in this order for the player: tabloid (warm-up, 64.5), UK government (40.0), Brussels (26.7), political-science abstract (14.3), then the archive pick of one year from 1987 (28.0), 2004 (27.3) and 2020 (25.5).
- Why this order: the easiest first teaches the drum. The national government comes before Brussels, so the jump is felt. The professor comes after, so the player finds out where Brussels sits. The archive comes last, with a real choice in it.
- Sources of the presets: the paper’s corpora and the paper’s years, Fig. 2.
- Score: none overall. Per item, the share of words got out, as lit windows, named in words. The end screen lists the items as a rundown, with your windows next to the paper’s reading-ease bar.
12. Controls, mechanism, feedback, moment
- Control: one drum pad, the bottom half of the screen. Pointer-down counts a tap, so two thumbs work.
- Mechanism (
mechanism.js): each script is a list of words, each with its syllables marked by hand. Each word gets one beat. The beat starts at 0.55 s and shrinks a little with every word since the last full stop (the reader running out of breath), down to a floor. A full stop adds a rest beat and restores the breath. A tap counts for the word on screen when it lands. A word is got out if the taps in its beat number its syllables, or one more. The item’s yardstick is the share of words got out.
- Feedback: the drum. The word on air is huge, its syllables are dots that fill as you tap, the word flashes amber when said and smudges grey when mumbled, and the next word waits underneath. A breath gauge drains through the sentence. The fill: 20 windows of a block of flats light up in proportion to words got out, item by item.
- Moment: Brussels, second sentence. “RE·IN·FORCE·MENT”, then “COM·PEN·SA·TION”, then “EN·TI·TLE·MENTS”, with the breath gauge nearly empty and the beat shrinking. Then, in the archive, CO·RO·NA·VI·RUS in 2020.
13. The look
Register: a TV news studio at night: teleprompter, ON AIR lamp, cough-key console. It comes from the paper’s world. Press releases are written for broadcasters and the midday briefing (p. 689), and journalists relay them to the public (p. 686).
- Palette: studio black
#0B0C10; prompter white #F5F2EA; on-air red #FF3B30; tally amber #FFB020 (said words, lit windows); breath cyan #3CD3E8.
- Type: Inter Display Black (prompter words and headlines) and JetBrains Mono Bold (timecodes, labels, rundown). Both are subset to Latin and embedded as data.
- Buttons: console keys. Chunky rounded rectangles, dark with a coloured LED strip along the top edge and a bevel; they press down. The drum pad is a big dark rubber pad with an amber ring that flashes on every tap.
- The one visual rule: one word on air at a time. It is huge, white, centred on black, with its syllables as dots beneath it. Everything else is studio hardware at the screen’s edges.
On the play screen, top to bottom:
- a header strip (ON AIR lamp, item name, ?), about 40 px;
- the breath gauge, a full-width cyan bar, about 10 px;
- the word on air, 40–64 px type auto-fitted to the width, with syllable dots about 18 px;
- the next word, small and grey;
- the beat bar, full width;
- the drum pad, about 45% of the screen height.
Each picture answers one question. Word: what do I say? Dots: how many taps? Beat bar: how long do I have? Breath: how long until a rest? Windows (result card only): how much got out?
14. What is cut
- Fig. 1, the monthly number of releases. One line on About.
- Fig. 3 topic by topic. It is replaced by the single move of putting every script on one topic (trains and passenger rights), plus one sentence on the end screen.
- The topic-matching check (Appendix B). One sentence on About.
- The familiar-words and focus-on-action measures are not played: we cannot reproduce Google Books frequencies or spaCy tagging at the paper’s scale. They appear as the paper’s values on the end screen, in words.
- The Irish government and broadsheets as items. They appear on the end screen’s bars.
- The paper’s three explanations, as questions on About.
15. Divergences
- Scripts are ours. Every script is written by us in the sender’s register, on one topic, so that its Flesch reading ease matches the paper’s average for that sender within 1.5 points. None is a real text.
- Syllables are counted by hand, the way they are shown to the player. The calibration check compares them with a standard heuristic counter.
- The beat, the rush and the tolerance are ours (beat 0.55 s, rush 1.2% per word into the sentence, floor 60% of the beat, got out = syllables or one more). They were chosen so that the game’s difficulty trades sentence length against word length at roughly the rate of the paper’s formula. See the calibration check.
- The windows are our device. “Words got out” are words the player tapped correctly, not a measure of comprehension. The paper does not measure readers.
- Archive years: three of 35, chosen for spread (pre-Maastricht, the Prodi–Barroso communication push, and the most recent year).
- The encore (our plain rewrite) is ours. It shows that the same news can be said plainly, which the paper asserts (pp. 686–7) but does not test.
Appended after reading build.md:
16. Mechanism kind and claim boundaries
Kind: reconstructed, for the scripts and the reading-ease values; toy, for the drum.
- The bars and the yearly line are the paper’s reported averages (Fig. 2), read off the figure.
- The scripts are a constructed stand-in: one invented text per sender, written to score that sender’s average reading ease. Demonstrates: what a text at that sender’s average score is like to get through, word by word. Assumes: that our text is typical. Real releases vary widely around the average (Fig. 2, lower panel), and ours is one point in that cloud. Paper establishes: the averages over full corpora, the within-topic gaps, and the flat trend.
- The drum is a toy reading of the formula: syllables per word become taps per beat, and sentence length becomes the breath that shortens the beat. Demonstrates that, under this game’s rules, a text with the Commission’s score is much harder to get out than one with a tabloid’s, and about as hard as a professor’s. Assumes the beat, rush, tolerance and windows. Paper establishes nothing about reading aloud, tapping, or how much any listener understood.
- Inside the game the writing says “a script written to score what the paper measured”, never “a Commission press release”.
- No lever is causal. The game never says politicisation caused anything, or that clearer releases would win anyone over.
Where the discovery comes from. Brussels is hard, between the governments and the professors, and unchanged over time: that comes from the paper’s numbers, carried by the scripts’ scores. That it is hard for your thumbs is the toy. The calibration check tests that the ordering does not depend on the toy’s settings, or on how our scripts split their score between word length and sentence length.
17. Calibration plan
calibration.js runs against mechanism.js in Node.
- The Flesch implementation reproduces published values: “The cat sat on the mat.” = 116.1; the Wikipedia platypus sentence = 24.4.
- Paper values reproduce the paper’s verbal claims (p. 692–3): the Commission is below both governments on all three measures and below both newspapers; it is below political scientists on familiar words; it sits in Flesch’s graduate band (<30) while the UK sits above it; the yearly series moves within a narrow band, with no year reaching the Irish average.
- Every script’s hand-marked Flesch score is within 1.5 of its target. Computed with an independent heuristic syllable counter, it is within 5. The scripts keep the paper’s ordering under both counts.
- Game difficulty follows the paper’s ordering. Model players at three skill levels (a seeded tapper with timing jitter) read every script on many seeds. Expected share got out: tabloid > UK > Brussels > professor. Brussels is closer to the professor than to the tabloid. The three archive years each differ from Brussels-average by less than the UK–Brussels gap. These must hold for every skill level, every seed band, and with rush on and off.
- Robustness to the split (the assumption the paper does not identify: whether a score comes from long words or long sentences). Synthetic texts are generated at each sender’s Flesch score, with words per sentence varied from 14 to 32 and syllables per word set to keep the score fixed. The model player’s share must keep the ordering tabloid > UK > Commission > professor across all splits within ±6 words per sentence of each other. We also report the trade-off the game makes between one word of sentence length and 0.1 syllables per word, against the formula’s 1.015 : 8.46.
- Robustness to the design: beat ±15%, tolerance exact vs ±1, and lag 0 vs 0.15 s leave the ordering intact.
- Balance: a competent model player gets most of the tabloid out (≥ 90%) and loses a noticeable share of Brussels (at least 15 points less than the tabloid). No script is impossible: the best model player gets ≥ 60% of the professor out.
18. Calibration result
node calibration.js gives ALL PASS (71 checks). The full output is in calibration-result.txt.
- Formula. “The cat sat on the mat.” scores 116.15 (published 116.1). The platypus sentence scores 24.44 (published 24.4).
- Paper values. All seven verbal claims hold on the values read off Fig. 2. The yearly series ranges 24.9 to 28.4 and never reaches the Irish average.
- Scripts. Hand-counted scores: tabloid 64.4 (paper 64.4), UK 40.5 (40.0), Brussels 25.6 (26.7), professor 14.8 (14.3), 1987 28.0 (28.0), 2004 27.6 (27.3), 2020 25.1 (25.5). A plain vowel-group counter puts them 0–6 points higher, mostly by under-counting “Eu·ro·pe·an”. It keeps the paper’s order.
-
Game difficulty on the scripts. This is the share of words got out by the model players, 300 seeds each.
| Player |
Tabloid |
UK |
Brussels |
Professor |
1987 |
2004 |
2020 |
| Careful one thumb |
0.83 |
0.74 |
0.59 |
0.60 |
0.70 |
0.67 |
0.66 |
| Practised one thumb |
0.92 |
0.86 |
0.79 |
0.74 |
0.77 |
0.82 |
0.82 |
| Two thumbs |
0.95 |
0.94 |
0.93 |
0.89 |
0.89 |
0.92 |
0.91 |
For the one-thumb players, the order is tabloid > UK > Brussels in every seed block. Brussels is closer to the professor than to the tabloid, and each archive year is closer to Brussels than the UK is. Two-thumb players are near the ceiling. For them the UK and Brussels scripts are level (0.94 vs 0.93), so the gap between government and Brussels is felt only by slower tappers. About says so.
- Drum speed asked (skill plays no part). Taps per second: tabloid 3.0, UK 3.5, Brussels 4.0, professor 4.3; archive 3.9 each, within 7% of Brussels. This is the line the result card shows, and it holds for any player.
- Robustness to the split. On synthetic texts placed along each sender’s iso-score line (sentence length 14–32 words, syllables per word set to keep the score), the order tabloid > UK > Brussels > professor holds at every combination within 6 words of each other, for both one-thumb players. The smallest margin is UK vs Brussels: 0.018. The game trades one word of sentence length against 0.010 (careful) or 0.014 (practised) syllables per word; the formula’s rate is 0.012. The shares along each iso-score line are nearly constant. In this game, as in the formula, it is the score that matters, not how it is made.
- Robustness to design. The order holds with the beat at −15% and +15%, with no rush, with exact taps only, and with a 0.1 s lag.
- Balance. A practised player gets 92% of the tabloid out. A careful player loses 25 points more on Brussels than on the tabloid; a practised one loses 13. Two thumbs get 89% of the professor out. Mashing ten taps a second gets 17% of Brussels out.
Where the discovery comes from, checked: the drum speed asked follows the scripts’ scores, which follow the paper. The order survives every split and design variation tried. The one place it depends on the player is the government/Brussels gap, which vanishes at two-thumb speed.
19. Build log
- The breath game came first. The first mechanic was breath per syllable, with the player holding to breathe. On paper it failed: syllables per sentence weigh word length about six times less than Flesch does, so a script could match the Commission’s score and still be easier than the UK’s. The drum replaces it. Word length becomes tap speed (Flesch’s heavy term), and sentence length becomes the breath that quickens the beat (the light term). The rush was then tuned until the trade-off matched the formula.
- Tap windows. The first version shifted tap windows by a 0.1 s reaction lag. Model players lost about a quarter of every script to taps landing on the wrong side of a boundary, whatever the text. The fix: a tap counts for the word on screen when it lands. What you see is what counts.
- Model players were too perfect. A tapper who never miscounts makes every script equally easy below its top speed. A per-tap slip that grows with the speed asked gives graded difficulty.
- Script balance. The first UK script used “au·to·mat·ic·al·ly” (6) and was as hard to drum as Brussels. It was rewritten with no word over four syllables and kept its score. The 1987 script was rebalanced from “lib·er·al·i·sa·tion / de·ter·mi·na·tion” to “re·struc·tur·ing / al·lo·ca·tion” to hit 28.0 without being the hardest script in the game.
- Item words. “You got every word out” showed with one window dark. That line is now reserved for a perfect reading.
- Lookahead. The preview of the next word now shows its syllables (“next: UN·DER”), so players can look ahead.
- Layout fixes from screenshots. The ON AIR lamp wrapped on the archive screen. The archive years overlapped their captions. The year chart clipped “2020”, and its “Constitution No” label collided with the next one; it is now “Treaty No”.
- Browser session.
session-test.js drives Chromium at 380 px with taps generated by the careful model player. It reads all five items, opens How to play mid-reading and returns (the take restarts from the card), picks 2020, reads the encore, opens About and returns, then plays again. The log is in session-log.txt: no page errors and no horizontal scroll.
- Self-review from screenshots.
- The aim can be stated from the start screen alone: tap once per syllable, get every word out.
- The discovery can be stated from the top of the end screen: Brussels writes almost like professors, harder than governments, and it never got easier.
- It does not look like any game in the table: black studio, one word on screen, a drum pad.
- Session length. About 2 minutes on air plus cards, so an estimated 4–6 minutes. Not measured with people.