Titled Curie’s Quartz until revision 4; renamed by the author on 2 Nov 2026 because the game had become a game about counterfactuals. The concept names below keep the old title.
Mine: Steffen’s signature and counter. Ideas given, and binding: “a kind of comic style for this one, ligne claire or manga; you choose, and you can surprise me.” This fixes the register family (Drawn: the comic) and leaves the choice between the two schools to the engine. It does not conflict with the precedent rule: none of the last three games (Lovelace, Noether, Brown) was a comic. Two older games in the table were comics (Ask First, a comic strip; On the Record, a pop-art political cartoon with speech-bubble buttons); this one is a different school of comic and takes nothing from either. The engine chooses ligne claire, and the surprise is in the device: the game is an album that draws itself out of the player’s experiments (section 13).
Marie Skłodowska Curie, “Rayons émis par les composés de l’uranium et du thorium” (Rays emitted by the compounds of uranium and thorium). Note presented by G. Lippmann. Comptes rendus hebdomadaires des séances de l’Académie des sciences 126 (1898): 1101–1103, session of 12 April 1898. No DOI. Available: the Académie des sciences archive scan, four pages (a cover and pp. 1101–1103), read in full, including both footnotes and the two tables. The note is in French; every quoted term below is translated. Online: the Comptes rendus volumes are on Gallica (Bibliothèque nationale de France); an English translation is in A. Romer (ed.), The Discovery of Radioactivity and Transmutation (Dover, 1964). No code, no data beyond the two printed tables; every number in the game is read from them or is declared ours.
Primary: an estimate. A table of currents, in units of 10⁻¹² ampere, through the air of a plate condenser (plates 8 cm across, 3 cm apart, 100 volts), one plate covered with a layer of the substance: 21 rows, from metallic uranium (24) to pitchblende of Johanngeorgenstadt (83), with the statement that every substance studied and not in the table is “at least 100 times less active than uranium”. A second table: the fraction of the radiation that passes an aluminium leaf 0.01 mm thick (0.2 for uranium compounds and artificial chalcolite; 0.33 for pitchblende and natural chalcolite; 0.4 and 0.7 for thorium oxide in thin and thick layers). Reconstructed: each substance in the game produces, in the condenser, the current the table gives it, with a small jitter.
Secondary: an argument. Activity belongs to the element, not the compound: “all the uranium compounds studied are active, and in general the more so the more uranium they contain.” It follows that a uranium mineral cannot be more active than uranium itself; two are (pitchblende, natural chalcolite), so “these minerals may contain an element much more active than uranium”. The control: chalcolite rebuilt from pure products by Debray’s method “is no more active than any other uranium salt”. The same logic, run the other way, disposes of white phosphorus: very active, but not as red phosphorus and not as phosphates, so its action is “probably of another nature”. Formalised: a rule (the reading of a uranium compound is at most about its share of uranium, in units of the metal’s reading) and the two tests (same element, other form; same form, other origin), played as choices.
Also handed over: the instrument. The current was “measured in absolute value by means of an electrometer and a piezoelectric quartz”. The paper names the instrument and no more; how it was worked (a weight let down by hand onto the quartz, whose charge holds the electrometer’s spot of light at zero) is from Curie’s thesis (Recherches sur les substances radioactives, 1903), declared as such.
A counterintuitive result, made firm by an identification. Anyone would say that pure uranium is the strongest thing uranium can be in: an ore is uranium diluted with rock. The paper measures a black ore from Saxony at three and a half times pure uranium, and a green uranium crystal from a mine at more than twice, though it is only half uranium. Then the identification: the same crystal made from pure chemicals is as weak as any uranium salt. So the excess is not uranium, and not the crystal; it is something in the natural rock that no bottle held. The note says so in one sentence and stops there. (After this paper, and not in it: polonium in July 1898, radium in December.)
Some stones give off invisible rays that let air carry electricity. Uranium does it, and the more uranium a thing holds, the more it does it. Marie Curie tested everything she could get her hands on. A lump of black rock from a mine, only partly uranium, did it three and a half times more strongly than pure uranium. So she made the same kind of crystal herself, from pure ingredients, and her copy was weak. The real rock was hiding something much stronger than uranium, which nobody had ever seen.
| Game | Position | Format | Feedback device | Chassis |
|---|---|---|---|---|
| Lovelace’s Cards (lovelace) | the analyst: a builder hanging cards on the Engine | construction challenge under a hidden test | the loom weaving the pattern | the Engine on pasteboard (Instrument): pasteboard buff, ink, plum, verdigris, thread red and blue; Abril Fatface + Old Standard TT; punched-card buttons |
| Noether’s Press (noether) | the transformation (an agent: the player is the group) | strategic, a contest with the Auditor; in its fifth revision a deck of changes | the stamped tag and the needle | night-shift machine hall (Techno): gunmetal, brushed steel, neon cyan, neon magenta, hazard yellow; Chakra Petch + Saira + Share Tech Mono; steel levers with a lit edge |
| Brown’s Restless Dust (brown) | controller (sets the conditions of one drop, aims at stillness) | impossible challenge | the live drop and the shelf of suspects | hand-coloured natural-history plate, 1828 (Austere): plate cream, ink, Prussian blue, almond gold, carmine, bottle green; IM Fell English; labels with a double rule |
What this game does differently.
The aim (start screen, top of the play screens, first line of How to play): Uranium makes the air carry electricity. Find what else does.
It is the paper’s own question (“whether bodies other than the compounds of uranium could make the air conduct electricity”), and it has a double answer the player does not expect: thorium, and something without a name hiding inside uranium’s own ores. The yardstick during play is the album page: each panel says, in words and two bars, whether the spot moved, and whether the thing holds no uranium (WITHOUT URANIUM) or does more than its uranium can (MORE THAN ITS URANIUM). Under the page, a line keeps the list in the aim’s units: Found so far, besides uranium: …
The verdict (the album’s last page, after the ninth panel), in three lines, then the player’s page:
No number, no score: the paper sets no quantity to score against, and a tally is not a verdict.
A. Curie’s Quartz (experimenter; discovery).
B. The Uranium Buyer (agent; challenge with a budget).
C. The Forger (controller; the mechanism run backwards; the reckless one).
A (the pick). 0:00. An album cover: a ligne claire drawing of a glass-walled workroom with Paris roofs behind; a brass condenser on a bench, a lamp, a beam of light, a bright spot on a long scale; a hand holding a small brass weight over a crystal. The title, Curie’s Quartz, and the aim: “Uranium makes the air carry electricity. Find what else does.” I tap BEGIN. 0:10. The bench, big. A caption says: the powder on the lower plate sends out rays; they let the air between the plates carry a tiny current; the current pushes the spot of light to the right; letting the weight down onto the quartz pushes it back; set how fast you let it down so the spot stands still on 0. “How fast you had to go is how strong the powder is.” The dish: pure uranium, from M. Moissan. The spot is already drifting right. I drag the slider; the hand starts lowering the weight; the spot slows, swings back past zero; I ease off; it sits on 0 and a ring fills around it. Balanced. The panel is drawn: a jar of uranium, a full yellow bar. A yellow tick appears on the slider: URANIUM. 0:50. The shelf: five drawers, fifteen jars and rocks, eight frames left on my page. What else could do it? I try lead, because it is heavy. The slider sits at zero and the spot does not move. Nothing. One frame gone; I frown. 1:30. Thorium oxide, a rare white powder, no uranium in it. The spot moves. I balance it just under the uranium tick. The panel: an empty yellow bar, a full blue one, and the words WITHOUT URANIUM. I grin: one found. 2:10. White phosphorus: it moves, more than uranium. Is it the phosphorus? There is red phosphorus on the same shelf, the same element; I spend a frame on it. Nothing. Same element, other form, nothing. I am thinking now. 3:00. Lacroix’s drawer: a black rock from Johanngeorgenstadt. The card says “mostly uranium oxide mixed with rock: about two-thirds uranium”. I set the slider just under the uranium tick. The spot keeps running. I push past the tick; it is still running, with speed lines. I push and push; it stops at three and a half. “!?” The panel: a short yellow bar, a blue bar running off the end of it, MORE THAN ITS URANIUM. 3:50. Natural chalcolite, half uranium: more than twice uranium. And in the last drawer: “Chalcolite you make yourself from pure chemicals.” I make it (a flask, green crystals forming) and measure. It barely moves; under its half. Same crystal, made pure: weak. So it is not the crystal; it is something in the rock. 4:40. The ninth frame. The album’s last page: what I found, what she found, and the sentence from the note.
C (the runner-up, and the reckless one). 0:00. Two dishes, two scales, two spots. Left: a black rock; its spot races. Right: my empty dish; its spot sits still. A row of bottles. No words, except the title. 0:15. I drag a scoop of yellow uranium salt into my dish; my spot creeps. I drag pure uranium metal; faster, but nowhere near. I wonder what the bottles are; tapping one shows a picture, not a name. 1:30. A white powder makes it go nearly as fast as uranium; a thicker layer of it, faster still. I am getting closer. Two-thirds of the rock. 2:30. I try everything; mixing slows it down. The green bottles and the copper: I make a green crystal (it looks like the rock’s cousin); its spot crawls. 3:30. I stop. I do not know what I was supposed to learn, unless I already knew that activity belongs to the element. The wordless screen has no way to tell me that the rock is mostly uranium, which is the whole point. Verdict on paper: the forgery is a great last act and a poor whole game; it survives as the last drawer of A.
A, Curie’s Quartz. The player sits where the paper’s author sat, with the paper’s instrument in the hand, and the discovery comes through the thumb before it comes through the eye: the slider has to go past the uranium mark, and keep going, for a rock that is mostly uranium and partly dirt. The deliberation is real twice over: which jar answers the question I now have, with eight frames for fifteen jars; and where the balance will fall, given what the jar holds. The paper’s two logical tests (same element, other form; same form, other origin) are on the shelf as choices, not as explanations. The sentence afterwards has no “guess”: I balanced the spot for the black rock and had to push three and a half times harder than for pure uranium; my own copy of the green crystal barely moved it. Originality: no game in the table has made the measuring itself a held skill, and none has drawn its record as a comic page.
Author test. Curie would recognise her condenser (8 cm, 3 cm, 100 volts), her table to the digit, her donors’ drawers (Moissan, Étard, Lacroix, Demarçay), the phosphorus that is “probably of another nature”, Debray’s chalcolite, and her one sentence about an element much more active than uranium. She would enjoy a stranger, sure of themselves, setting the slider for a rock at two-thirds of uranium and then chasing the spot past the uranium mark. A colleague who knows the answer still has to spend eight frames well and still has to hold the spot.
Runner-up: C, The Forger. Switch if the author wants a ninety-second game built only on the control experiment, and is willing to state the rule (activity belongs to the element) on the start screen.
What the reckless concept would have given. A wordless race in which the impossibility of the forgery is felt rather than read. Kept from it: the last drawer, “make it yourself”, in which the player runs Debray’s recipe and races their copy against the memory of the natural crystal. Given up: the silence and the side-by-side race.
One long round of nine panels: about five to six minutes. Screens: the cover (start); the shelf (the album page above, the five drawers below, a card for the chosen jar); the bench (the balance); the end (the album’s last page); How to play; About. The first panel is fixed (pure uranium: the yardstick and the lesson in balancing); the other eight are the player’s, from fifteen. No presets beyond the paper’s own table; the shelf is the table, cut to fifteen, plus the shelf of the “at least 100 times less active” (lead, gold, copper, red phosphorus, a phosphate). The drawers are ordered for the player: uranium and its salts first (the rule), then the Muséum’s rocks (the discovery, early and prominent), then the rare oxides, the chemist’s shelf, and last the one you make yourself. No score; the page is the yardstick; the verdict is in words. Replay starts a fresh album.
| Mechanism (reconstructed from the table): each substance has a current i from the table, in units of the metal’s 24 × 10⁻¹² A. The slider sets the quartz’s counter-current r (same units). The spot is a leaky integrator of the difference: ds/dt = G·(i − r) − s/τ + jitter, clamped to the scale. Balanced when | s | stays inside the zero band for 1.5 s; the reading is r then, within a few hundredths of i. Inactive substances leave the spot still at r = 0 and balance at once. Each substance carries its uranium share (ours, from formulas and typical mineral values) for the yellow bar. Numbers in CONFIG with their table rows. |
Register: a ligne claire album, Paris 1898 (Drawn family: the comic). It comes from the Given (ligne claire or manga), and the choice between the two comes from the paper’s setting: a note in French, read to the Académie in Paris, worked out in a ground-floor workroom of the École municipale de physique et chimie industrielles. The Franco-Belgian clear line is the comic of that city and language, and it is the comic that draws instruments best: brass, glass and wire in one even line, nothing shaded, everything legible. Manga is kept for one thing only, the emanata ligne claire shares with it: speed lines on a running spot, a “!?” over the panel that surprises.
The surprise: the album draws itself. Each measurement becomes a panel on the player’s page, in the clear line, captioned; the empty frames are the budget; the last page is the verdict with the player’s page beside the paper’s. The cover is the album’s cover.
#FFFFFF (the page; panels sit on it with white gutters); ink #17140F (every line, 2.5 px at phone scale, and the text); uranium yellow #F7C600 (uranium alone: the uranium jar, the yellow bar of what a thing’s uranium could give, the URANIUM tick on the slider); clear blue #3E8ED0, with a pale tint #D6EAF7 for walls and skies (what the spot measured: the blue bar, the balanced ring, the primary caption box); signal red #E2342B (the “!?”, the MORE THAN ITS URANIUM caption, nothing else). The substances are drawn in their own flat colours (chalcolite green, red phosphorus, gold, black pitchblende, white thorium oxide), as the album would colour them.From the table: black uranium oxide (27, a little above the metal; the paper’s “in general” covers it), hydrated uranic acid (6, well under its share), the uranates (12), the uranous and uranyl sulfates, thorium sulfate, cerium oxide with the fluoxyniobate (0.3), the Cornwall pitchblende (16, whose uranium content we cannot state), autunite, the thorites, orangite, samarskite and the five rare minerals, cleveite (“very active”, no number). All stay in CONFIG with their values, unused, for the author to restore. From the text: the thickness effect for uranium compounds (said to be very small; not played), the second table (absorption by aluminium), the photographic impressions, the analogy with Sagnac’s secondary rays, the closing guess that space is crossed by very penetrating rays which only heavy atoms absorb (quoted on the About screen, not played), the remark that uranium and thorium have the highest atomic weights (in the verdict, and on the cards of lead and gold as “a heavy metal”). The quartz’s arithmetic (charge = constant × weight; current = charge ÷ time) is underneath; the player sees a push.
Reconstructed. Every substance on the shelf makes the air carry the current the paper’s table gives it (p. 1101–1102, units of 10⁻¹² A, shown relative to metallic uranium’s 24). The electrometer and quartz are a toy instrument around that number: the spot is a leaky integrator of the plate’s current minus the quartz’s, and the reading is the charge balance over the seconds the spot is held still, so any thumb that holds it reads the table’s number to within the spot’s tremble (about three hundredths of uranium’s reading; five for pitchblende).
Where each staged discovery comes from. That the rocks exceed their uranium: the paper’s table (83, 67, 52 against 24), with the uranium shares ours; robust to any pitchblende share from 40 to 88 in 100 and any threshold from 1.1 to 4 (calibration §5). That the copy is weak: the paper’s table (9 against 52). That white phosphorus is not the phosphorus atom: the paper’s sentence; robust to any number we give white phosphorus. That thorium does it: the table. The budget of nine panels is ours and creates no finding; it makes the order a decision (calibration §6: the paper’s own path fits in eight; 92 % of random albums meet a rock).
calibration.js runs mechanism.js unchanged. It checks: (1) every shelf value against a second, independent transcription of the table from the scan, and the inactive five below the paper’s “at least 100 times less”; (2) each of the paper’s statements on p. 1102 (all uranium compounds active; in general more so the more uranium, Spearman ρ = 1.00 on the shelf and 0.75 over all seven uranium-compound rows of the table with formula shares; thorium very active and beats the metal in a thick layer; tantalum slight; white phosphorus active, red phosphorus and phosphates not; pitchblende and chalcolite more than twice uranium; artificial chalcolite within the salts’ range; the two chalcolites ×5.8); (3) the instrument under four thumbs (steady, eager, a last-instant yank, a shaky one) over 120 seeds for every substance: all balance, every reading within 4.5 σ of the table (worst 3.2 σ), and a still thumb records “nothing” for the inactive five and never records an active plate; (4) every reading the balance can return gives the same kind of panel; (5) robustness to what is ours; (6) the shelf in numbers. Result: 67 passed, 0 failed (calibration-result.txt). The table is a scan of typeset text, read by eye; the two transcriptions agree.
No score; the decisions are which eight of fifteen. 20 000 random albums and five plans:
| strategy | meets a rock | makes the copy | rock + copy | both chalcolites | phosphorus test | thorium |
|---|---|---|---|---|---|---|
| random | 92 % | 53 % | 48 % | 27 % | 41 % | 80 % |
| drawers top to bottom | yes | no | no | no | no | yes |
| drawers bottom to top | no | yes | no | no | yes | no |
| heavy metals and rare things first | no | no | no | no | yes | yes |
| the paper’s path | yes | yes | yes | yes | yes | yes |
| rocks only, then the copy | yes | yes | yes | yes | no | yes |
The spread is wide: only a player who thinks about what each test answers gets both logical tests and the rock. Nobody can fail to learn something: the verdict names the rock and the copy for those who missed them. The budget of eight was kept: seven would make the paper’s own path impossible; nine would let the drawer-by-drawer player stumble through everything.
mechanism.js), headless; the reading first taken as the slider’s value at balance, then changed to the charge balance over the held seconds after the calibration’s adversarial thumb showed that a quick sweep at the last instant could move a slider-value reading arbitrarily. Calibration: 67/67.loop.js, loop-log.txt): read as the stranger; names with commas inside (“Thorium oxide, thin layer”) made the captions stumble; renamed (thin thorium oxide; pitchblende from Saxony). The found-so-far line now says “something in pitchblende” for the rocks, since the rock holds uranium.session-test.js, session-log.txt): two albums in headless Chromium at 380 × 760; the first balanced with a real mouse drag on the slider, How to play opened mid-measurement (the bench’s clock stops while away and the slider is kept), the paper’s path to the last page, About and back; the second never opening the Muséum’s drawer, whose verdict says where the rock was. No errors, no horizontal scroll. Balancing took two to ten seconds per plate for the scripted thumb; the five-to-six-minute length is an estimate until a person has played it.What changed from the memo above. Long explanatory text on the bench is set in mixed case for reading on a phone; capitals are kept for buttons, bands, badges and panel labels. The jars carry a label, yellow when the thing holds uranium (the visual rule, extended to the shelf).
Given, from the author after playing it. The bench gave the method away (“hold the spot on 0”) and then forced it: the panel was drawn only when the spot had stood still on 0, so a player who set the slider anywhere else saw no way to commit, and the game stalled. Same fault twice: the game accepted only the right answer. The author’s decisions, binding: (1) a full pass over the game for this fault; (2) the player must be able to play badly, even badly enough to miss the discovery, because that is how they learn; (3) the player is scored against Curie, by how much of her note their own page can back; (4) the bench says little: what the things are and what the slider does, not how to measure; (5) a reading taken with the spot still moving is recorded honestly, as a bound or with error bars.
The fault, everywhere it appeared. The instruction to hold the spot on 0; the auto-commit that waited for the right answer (the skill’s own rule: “no button that waits for the right answer”); the blue band around 0 and the ring that filled, a target and a hint; the “!?” that rose before commitment when the push passed uranium and the spot still ran; panel labels decided from her numbers, not the player’s; inert plates recording themselves after two seconds.
The change.
Build, revision 1. Mechanism: benchStep no longer ends anything; commit writes down readingOf (held, at least, at most, roughly); labels by kindOf from the player’s own readings and own pure uranium; claims and verdict score against her note. Screens: WRITE IT DOWN under the slider (wakes after 0.8 s); no band, ring or “!?” on the bench; the bench text says what is on the plate, what the slider does, and that the page keeps the push and what the spot was doing, and nothing about 0; panels draw error bars, “at least” arrows and “at most” open bars, with her triangle; every jar can be tested again; the last page leads with the count of her claims backed, a fate, and each claim with what backed it or why nothing did.
Calibration: 53 passed, 0 failed. New checks: every reading written down by six kinds of thumb at random moments contains her value (0 misses in 5 760); 3 000 random albums carry no label her table contradicts (CAN’T TELL is always allowed); a careful thumb gets her labels on every jar (the faint tantalum salt needs a five-second hold to be seen, as her “semble être légèrement actif” suggests); her claims in numbers:
| plan | careful: mean claims, all six, rock claim | sloppy: mean, rock | hasty: mean, rock |
|---|---|---|---|
| the paper’s path | 6.0, 100 %, 100 % | 3.7, 70 % | 1.1, 0 % |
| random eight | 3.8, 2 %, 93 % | 2.4, 64 % | 1.0, 0 % |
| drawers top to bottom | 4.0, 0 %, 100 % | 2.8, 78 % | 1.4, 0 % |
| rocks first, then the copy | 4.0, 0 %, 100 % | 2.5, 77 % | 1.5, 0 % |
| dull jars (all read zero) | 0, 0 %, 0 % | 0, 0 % | 0, 0 % |
Both choices now matter, which jars and how well: the right jars measured in a hurry back about one claim, and a hasty player who writes the rock down at the push its label suggests never backs the discovery. Browser session: album A (real drag; the slider left at a wrong push for two and a half seconds: button awake, nothing ends by itself; How to play mid-measurement; the rock written down in a hurry, CAN’T TELL with her triangle far to the right, then measured again: all six claims); album B (dull jars at once: none). No errors, no horizontal scroll.
Given, from the author. Revision 1 still came down to the player’s skill at holding the spot on zero, and never referred back to electricity. The author asked for the conclusions drawn from poor measurements to be explored instead, ambitiously: after each experiment the conclusion it suggests, in comic fashion; at the end, all conclusions, right and wrong, put together into a counterfactual physics and a counterfactual world, drawn as if it were true and never corrected, beside ours. Decisions, binding: the conclusions are drawn by the game, not chosen by the player; readings are plain numbers; no score, only a comparison in words; the worlds run forward from 1898; balloons come from the bench, with no Curie drawn; ideas are planted one per experiment and assembled only at the end; no hint toward the hardest world. The full design, as agreed, is design-revision-2.md.
What changed.
WORDS.ideas in the mechanism), and four clashes that fire once when a later reading contradicts an earlier one. A retest replaces the earlier reading (“Second look: …”).settle()): where the energy comes from (precedence: a heavy metal active makes the sky wind; a salt off its uranium, or a yardstick that makes them so, the slow fire; the copy as strong as the mine’s, the crystal lens; otherwise the atom, which is the restless atom if a rock beats its uranium and the bottomless atom if the rocks were measured and none does); which matter does it; what hides in the rocks; white phosphorus as chemistry or atom. A jar never measured leaves its question as in our world. The bottomless atom is the deep twist: an atom spending energy while old rocks hold only their uranium means energy from nowhere, which is what her own argument rules out.Calibration: 58 passed, 0 failed. The table and her statements (as before); a reading is the number written; her table gives her physics, and a careful thumb on her path writes her note in 40 of 40 albums; fourteen designed wrong numbers each plant their idea and make their world; 40 000 random pages yield 34 reachable combinations of energy, matter and rocks, every one drawn with pictures its laws allow; retests replace and clashes fire once; no balloon over 200 characters; no decimals in the words. The worlds in numbers:
| page | hers | restless (other) | slow | sky | crystal | bottomless |
|---|---|---|---|---|---|---|
| her path, careful | 97 % | 0 | 3 % | 0 | 0 | 0 |
| her path, sloppy (writes early) | 80 % | 0 | 20 % | 0 | 0 | 0 |
| her path, hasty (writes the label’s guess) | 0 | 0 | 0 | 0 | 0 | 100 % |
| random eight, careful | 86 % | 13 % | 1 % | 0 | 0 | 0 |
| random eight, sloppy | 76 % | 12 % | 11 % | 0 | 1 % | 0 |
| random eight, hasty | 0 | 0 | 0 | 84 % | 0 | 16 % |
The hasty player who writes a rock down at the push its label suggests lands in the bottomless atom, the deepest of the worlds, without any hint. “Restless (other)” is a page that is the restless atom but differs from hers elsewhere, most often white phosphorus measured without its control, which makes living light.
Fiction and fact. Every alternative world is labelled as fiction on its page. The page called Ours is history; its dates (polonium and radium 1898; radium’s heat and the Nobel Prize 1903; the nucleus 1911; Hess 1912; fission 1938; the Earth’s age 1956; element 93 in 1940; the ruby laser 1960; Kelvin’s 20–40 million years) were written from the builder’s knowledge and are flagged in the author note for checking. Curie weighing Carnot’s principle (1899), mentioned in the design, is not used on screen.
Revision 2.1 (same evening), from the author after playing it. (1) No comparison with her during play: the caption no longer says “Hers: …”, the panels lose her black triangle, and pure uranium’s caption no longer says how hard you pushed against her; her numbers appear only on the page called Ours and in the closing, where they part. (2) The ideas were not prominent enough and not comic enough. Each now arrives as a big comic speech balloon over the bench (the picture fades behind it), led by a comic word in the display face (WOW! in red, HMMM… in blue, AHA!, WHAT?!, SO!, OH!), and phrased “Maybe…” or “That means…”: “WOW! Electricity near lead! That means every heavy atom does it, the heavier the more. Something must be raining down on them from space!”; “HMMM… Just about its uranium: there’s nothing else in that rock. So uranium shines on and on and never changes… maybe it never runs out?” (which is the bottomless atom, planted in a sentence). The game’s own statement becomes a small narrator’s caption box in the corner above it, as in a comic. Clashes come as a dashed red box, WAIT… . Calibration 58/58 with the new wording.
Given, from the author. The balloons did not look like real comic balloons; their text was repetitive; when the same bad experiment comes round again they should notice (“didn’t I know it!”); the comic word should be in italics; “about what its uranium gives” read as wrong (it is grammatical, its being the possessive, but it trips the eye, and “its uranium” means nothing to a stranger); is there a balloon for being on track? One more coherent sweep. The author went to bed before answering four questions; the builder took the defaults it had proposed, recorded here so they are easy to reverse: (1) the voice is Marie Curie’s, off the panel and never drawn; (2) “on track” comes from readings agreeing with each other, not from agreeing with her, so a wrong world feels just as right from inside; (3) three shapes: speech, thought cloud, shout; (4) the sweep covers every text the player reads.
The voice (voice(), beliefOf() and WORDS.voice in the mechanism). Each reading plants a belief on the page (the atom, the slow fire, the crystal, thorium, uranium alone, heavy atoms, not weight, every metal, the stuff of life, phosphorus is chemistry, something hidden, nothing hidden, the rare metals). She remembers:
Plain words. “Its uranium” and “their uranium” are gone from everything the player reads. The balloons say the amounts: “Only two-thirds uranium, so it should be weaker than pure uranium. But it’s three and a half times as strong! It can’t just be the uranium in it: something else is hiding in this rock!”; “Half uranium, and about a third as strong as pure uranium: no more than the uranium in it should give. It must be the uranium doing the work, whatever it’s mixed with!”. Panel badges read AS EXPECTED, STRONGER THAN EXPECTED!, WEAKER THAN EXPECTED; the yellow bar is “what the uranium in it should give”. The narrator’s caption is short (“Lead: about half as much electricity as pure uranium.”).
Real balloons (drawBalloon() in the template). The words are laid out first, in capitals, the comic word in bold italic; then an outline is fitted around them: a rounded oval with a pointed tail toward the speaker below the panel (speech), a scalloped cloud with three bubbles (thought), a jagged burst (shout). The bench picture fades behind them; the panel grows to hold them. The shelf’s “ideas so far” and the last page’s pinned ideas use the same shapes, small, with their panel numbers. The separate badge chip under the result panel is gone (the panel carries it).
Checks: 61 passed. New: the words never say “its/their uranium”; 2 000 random albums leave no blank in any balloon; the longest balloon is 269 characters (a changed retest, two sentences); 37 repeated lines in 18 663 balloons, only where a belief is met more often than it has variants.
Given, from the author. The game needed more world-building. The builder’s diagnosis, agreed: the game was two games stapled together, the bench and a world that lived only on the last page, which had grown to fifteen phone screens of scrolling, with the world itself one splash panel, three small stick-figure vignettes and a paragraph. The five dated beats of the revision-2 design had been collapsed into that paragraph at build time. The author left all choices to the builder.
The change.
scene() in the mechanism, SCENE in the template): a consequence of the belief the reading just planted. Pure uranium opens the strip (Paris, the workroom, 1898); a belief met for the first time draws its first scene, met again its second (lead crackling → sails of lead on the mountains; a rock HIGH → sacks hauled to Paris → the cauldron → grains that glow by themselves); a jar that planted nothing draws Paris carrying on. The year ladder is ours: 1898, 1898, 1899, 1900, 1902, 1904, 1906, 1908, 1910, one per panel in order. A clash turns the year band red and stamps “?!” on the scene.worldOf still computes slots and call-outs for the calibration’s picture-vs-laws check, unused on screen.fig: coat, skirt, top hat, cap, bonnet, helmet, a face with an eye and a nose), Paris houses with mansard roofs and dormers (maison), lampposts, trees, flasks, a cauldron, a balloon, gold leaf. Twenty-nine scenes and the five panoramas redrawn with them; the old stick figures (person, house, town, VIGNETTE, worldSVG) removed.Checks: 61 passed, 0 failed (the strip’s words obey the “the uranium in it” rule). Browser session: the end page is now 6 000–8 100 pixels at phone width instead of 11 000; no horizontal scroll (an empty frame with aspect-ratio: 1 stretched to its row’s height and widened its column, fixed with align-self: start); no errors. Dates on the Ours strip are as before and still flagged for checking in the author note.
Given, from the author. “The game that you yourself would most enjoy”: a free hand, more than five worlds, no interim reports. His one question on the way in, whether a reading written while the spot still ran right should make a different world from one written while it ran left, was answered by the instrument itself: in this electrometer the spot stands only when the push matches the plate, so every wrong number is a running spot, and the direction is just the sign of the error. The direction therefore goes into the diagnosis, not the fiction (below), and the richer world comes from elsewhere.
The worlds. A page now makes one of ten named worlds (worldId(), WORDS.worlds). The energy question still comes first and gives the five from revision 2 (the restless atom, the slow fire, the sky wind, the crystal lens, the bottomless atom). When the physics is hers but something else is not, the page gets its own world instead of an inset: the leaking world (copper active: no wire holds a charge, the pneumatic post, no lightning, a city of clockwork), the electric bone (white phosphorus active with no control, or a control that also reads active: doctors listen to skeletons, graveyards crackle, love in volts), the quiet metal (thorium silent and nothing else active: uranium in a museum case, Paris asleep), the Académie of Doubt (three or more panels measured over again, DOUBT_RETESTS: everything stamped thrice, radium found in 1931 and confirmed in 1945) and found elsewhere (her numbers, the rocks never looked at: the sacks go to Vienna, Paris a footnote). A page that is hers in every number is her note only if it is also hers in manner: a doubting page or one that never opened the Muséum’s drawer no longer reads “your page is her note”. The oddities (the sails’ element-93 miners, the fading lamps, the pedestal, the doctor, the storm, the humming jewellery, and now the stamps and the Viennese sacks) stay as insets: a sentence in the paragraph, and the page’s strongest one takes the middle closing panel.
The strip runs on. After the player’s nine panels the game draws three of its own, 1920, 1945 and TODAY, from the world’s laws (closing in WORDS.worlds, the band in blue, tagged “and then”), before the panorama. Ours gets its last two the same way. Thirty closing panels in all.
The direction of the spot. On the page, the speed lines after a bar now end in a chevron pointing the way the spot was running. On the last page, the sentence that names the number behind each wrong law adds, in brackets, what the instrument knew: “the spot was still running to the right when you wrote it down: the real current was higher than that”, or the left-hand version. Nothing on the bench changed.
Drawings. Five new panoramas (a city with no wires; a hospital ward with a graveyard beyond; a museum hall with a dozing guard; the Académie with three stamps and three electrometers; a railway platform with sacks going east) and nineteen new scenes, in the same hand.
Checks: 69 passed, 0 failed. New: all ten worlds are reached by random pages (40 000 pages, 20 % of them doubting: sky 25 513, slow 13 072, crystal 473, bottomless 463, leaking 187, restless 132, quiet 66, bone 59, doubt 32, elsewhere 3); a designed page for each conduct world; the conduct worlds and insets never contradict their laws; a rock written on the run gets the directional parting; the energy question beats the conduct questions (thorium silent with a rock at its uranium is the bottomless atom, not the quiet metal). The worlds in numbers, 300 albums each: her path careful 97 % hers, 3 % slow; her path hasty 100 % bottomless; random eight careful 81 % hers, 13 % electric bone (white phosphorus drawn without its control), 5 % found elsewhere, 1 % slow; random eight hasty 84 % sky, 16 % bottomless. Browser: the three albums of session-test.js as before, and world-tour.js, one designed page per world rendered to its last pages (screenshot-world-*.png), every picture the mechanism names drawable, no errors, no horizontal scroll.
The author’s call: Curie’s Counterfactuals, because the game had become a game about counterfactuals. The file is curies-counterfactuals.html; curies-quartz.html stays as a redirect so old links keep working. The title on the cover, the page title and the README row changed; nothing else.
Given, from the author after playing revision 4. Three notes and a brief: (1) the strip’s thumbnails are small and the captions under them tiny; (2) the sentences after a reading relate it only to pure uranium, never to the measurements before it, so the page felt endogenous to the last number only; (3) the last page urges a replay to see the other worlds, and an impatient player wants a look; then “a full sweep of the game paying close attention to narrative consistency”.
The page’s numeric memory (compareLine(), PARTNER). The narrator’s caption, the game’s own voice, now adds one sentence placing the new number on the page: against its natural partner when there is one (thin and thick thorium; the two pitchblendes and the mine’s chalcolite; the copy against the mine’s crystal; red phosphorus and the phosphate against white; gold against lead; copper against both; the two salts; the tantalum salt against thorium), in whole fractions (“Twice what the thin thorium oxide in panel 3 gave”, “About a quarter of what the chalcolite from a mine in panel 8 gave”, “Nothing, where the white phosphorus in panel 2 was a bit stronger than pure uranium”); failing a partner, against the page (“More than anything on your page so far”, “About as much as the thorium in panel 3”, “Less than anything that gave something so far”, “Nothing, like the lead in panel 4”). The balloon stays free for the idea.
The world is built from the page’s beliefs (later, keyBelief, SUPERSEDES). The three “and then” panels and the paragraph no longer come fixed per world. Each belief a page holds has a later scene and a sentence; the latest reading of each jar decides what it planted, and a later belief retires the one it supersedes (red phosphorus silent retires “the stuff of life is electric”; the copy ordinary retires “the way the crystal is built”; and so on, in both directions). 1920 is the later scene of the world’s key belief (the slow fire’s salt, the sky wind’s heavy metal, the bottomless atom’s rock at its uranium …), 1945 the later scene of the latest other belief, TODAY the world’s own; a scene already on the player’s strip is never drawn twice, and the world’s own panel stands in. The paragraph is the world’s two sentences plus the sentences of the three latest other beliefs. Nothing on the last page now refers to a jar the page never lifted; a sky-wind page that never touched a rock gets no miners. The oddity insets and the fixed closing texts are gone.
The peek. On the last page the ten world names are tappable: a world the player’s albums have not made opens as its panorama, its name and its three laws (from a designed page in CONFIG.EXAMPLES), with the line “a different page gets there”; the “and then” strip and the paragraph stay for the player who earns them. No recipe anywhere.
Tap to see it big. Any panel on the album page or on either strip opens full width with its caption.
The sweep. Every text the player meets was read against the page that produces it: balloons, captions, scenes, later scenes, paragraphs, partings, help, about. Changed: the gold scene no longer blames plumbers alone; “sacks of pitchblende” became “sacks of ore” where a chalcolite may have planted the belief; the paragraph’s belief sentences say “the heavy metals” rather than naming both when one may be unmeasured; the Found Elsewhere TODAY line no longer quotes a salt the page may not have measured. Unchanged on purpose: the two scenes a belief may draw when met again are the same for every page, because they are the belief’s story, not the jar’s.
Checks: 77 passed, 0 failed. New: the comparisons (partner, page, nothing-against-something); closings carry the panels that planted them, are three different scenes, never repeat the strip, end on TODAY; every example page lands in its own world. Browser: the three albums, the ten-world tour, the peek and the enlarge overlay.