Leavitt’s Cloud — design memo
0. Given
Mine: Steffen’s signature and counter, no beta line. His ideas, binding:
- It is Leavitt, not Pickering. The circular is signed by Edward C. Pickering, the director, but its first sentence says the statement “has been prepared by Miss Leavitt”. The game credits her throughout, and, although these games normally carry no biography, the debrief gets a short mention of who signed and who did the work.
- Simple and sparse. One instrument, one chart, few screens, few words.
- Straight, but subtly feminine in the look. Read here as: no ornament, no pastiche, no pink; a quiet palette of lavender glass, aubergine and violet ink, a fine serif, rounded buttons.
Precedent is not overridden by any of this.
1. Paper
Leavitt, H. S. (1912), “Periods of 25 variable stars in the Small Magellanic Cloud”, Harvard College Observatory Circular 173: 1–3 (signed Edward C. Pickering, 3 March 1912). Bibcode 1912HarCi.173….1L. Read online via the NASA ADS: https://ui.adsabs.harvard.edu/abs/1912HarCi.173….1L. Available: the ADS scan only (3 pages). No transcription was found, so Table I was read from the scan by eye and every number in the config block is tagged as read off the scan; the calibration check re-derives the paper’s own residual columns from it, which catches a misread digit.
2. What the paper hands over
Primary: an estimate. Table I: for 25 variable stars, the brightness at maximum and minimum (photographic magnitudes, provisional scale), an epoch and a period in days, and the residuals from two straight lines. Figure 2: magnitude against the logarithm of the period, two straight lines, “the logarithm of the period increases by about 0.48 for each increase of one magnitude in brightness”. Secondary: a short argument. The stars are “probably at nearly the same distance from the Earth”, so their periods go with their real output of light. Plus a verbal description of the light curve: a slow fade, a long time near minimum, a very rapid rise to a brief maximum, a range of about 1.2 magnitudes for bright and faint alike.
3. The discovery, and its kind
A pattern found in hard-won measurements (emergence in the data, with a dash of identification): the brighter a variable in the Cloud, the slower it beats, and on the right scale the relation is a straight line. Why it matters is the identification step in one sentence: because all the stars sit in one cloud, about equally far away, how bright they look is how bright they are. The famous consequence (a star’s beat gives its true brightness, so its faintness gives its distance) is not claimed in the paper; it is hoped for (“the parallaxes of some variables of this type may be measured”). The game plays the pattern, states the reason, and reads the consequence on the verdict as what came after.
4. The core narrative
Far away there is a cloud of stars. Some of them blink: they get bright, then slowly dim, then flash bright again, over and over, each at its own pace. A woman at Harvard looked at hundreds of photographs of that cloud, night after night, and worked out how long each blinking star took to come round again. Then she noticed something nobody expected: the bright ones blinked slowly and the faint ones blinked fast, every time. And since all of them are in the same cloud, the same distance away, that meant the slow ones really do give out more light. A star’s blink tells you how bright it truly is.
5. Who lives in the paper
- The plate reader (Leavitt): estimates each star’s brightness on each glass plate against comparison stars, then finds the period from irregular dates. Controls: the period she tries. Anticipates: where the next maximum falls.
- The star: pulses; its brightness and period are fixed by its nature (“mass, density and surface brightness”). Controls nothing the paper names.
- The chart maker: chooses the axes; Figure 1 (period in days) gives two curves, Figure 2 (log period) two straight lines.
- The observatory (Pickering’s circular): decides what to publish and what to hope for (spectra, parallaxes, the remaining two thousand).
6. Precedent
| |
Position |
Format |
Feedback device |
Chassis |
| Jevons’s Noon |
controller (a nation setting a pace) |
challenge, scored |
a ruler, a filling sheet, a trajectory against the real curve |
Instrument (Jevons’s own plate): cream, rule blue, soot, furnace orange; Bodoni Moda; double-ruled engraver’s labels |
| Julian’s Window |
subject / agent (Julian in bed, choosing where to look) |
discovery, no score |
a strip of sixteen frames |
Drawn (shōjo manga): white page, ink, one rose; Delius and Italiana; panel buttons |
| Augustine’s Clock |
subject (the mind as the clock) |
discovery, graded holds |
bars in memory under a now point |
Clean (Bauhaus poster): plaster, black, red, blue, yellow; Jost; flat slabs |
The nearest older neighbour is Hubble’s Twelve Nights (experimenter at a telescope; dark sky and brass; eyepiece and spectrograph). This game is different on all four counts: position experimenter (none of the last three); format discovery with a prediction close; device the fold (tuning a beat until scattered dots line up into one tooth, then dropping a stroke onto a chart), new to the table; chassis Austere family (none of the last three; Hubble was Instrument), a dark room after three light grounds, a lit glass negative instead of a sky, violet ink instead of brass.
7. The aim and the verdict
Aim (the sentence the player reads): Find how long each star in the Cloud takes to blink, and see what the blinks have in common.
Verdict, the lines the end screen leads with:
- What you did: You timed six stars of the Cloud by their blink, and a seventh from its light alone. (Then, in words, how near your beats were to hers.)
- What the paper says: Miss Leavitt’s twenty-five fall on a straight line. The brighter the star, the slower its blink: a blink about three times as long goes with about two and a half times the light.
- What it means: They are all in one cloud, about equally far away, so this is how bright they really are. Time a star like this anywhere in the sky and you know its true brightness; how faint it looks then says how far away it is. The circular does not claim that. It hopes for it.
8. Three concepts
A. The Plate Reader (experimenter; discovery with a prediction close). The player is Leavitt at the plates. Each round is one star: its plates arrive (black dots on glass, bigger is brighter, each on its date), and the player tunes a beat dial until the scattered dots line up into one tooth per beat. They write the beat down; her number and her curve appear in violet under theirs; the star drops onto a chart as a stroke from its faintest to its brightest. Controls: a beat dial (coarse and fine). Before pressing they are thinking: are these dots lining up, is this one tooth or two, and later, this star is bright, so where should I start looking? After: the fold snaps, the ghost agrees or not, the chart grows. Moment: by the third star the chart has a slope; the player starts using it to find beats faster; the seventh star is timed from its light alone. Fun: discovery, mastery (the dial gets easier once the law is known), anticipation (the snap). Replay: new plates, a different order of discovery, the prediction tried with the law in mind.
B. The Cloud Keeper (controller). The player sets a star’s size and watches its blink and light; aim: populate the Cloud to match the plates. Fails: the paper has no model of the star, so the controller would be turning a dial that is the law; there is nothing to discover, only to confirm.
C. The Heartbeat — the reckless one (subject; real time; no words on the play screen). The player is the photographic plate: stars blink in a time-lapse and the player taps along with each, a rhythm game; the chart builds from their taps. Moment: the big ones are slow, the small ones flutter. Fails on the walkthrough: the paper’s periods span a factor of a hundred (1.25 to 127 days), so at any single time-lapse speed the fastest flicker at eight times a second while the slowest take half a minute; changing speed per star hides the very comparison the game is about.
9. The first five minutes
A (pick). 0:00 A dark page with a lit glass rectangle: a crowd of black specks, one ringed in violet. “Find how long each star in the Cloud takes to blink.” I press BEGIN. 0:15 Plates arrive one by one: the ringed dot swells and shrinks; a dot drops on the record panel at each date. The caption says each plate is one night and a bigger dot means brighter. I see it is jumpy. 0:40 The dial appears: “Set a beat. The dots will be laid out by where they fall in that beat.” I drag; the dots smear, then at one spot they snap into a sharp climb and a slow fall, twice. I grin. I nudge the fine dial and the tooth sharpens. 1:10 WRITE IT DOWN. Her violet curve lies on my dots: “Your beat: 4 days 13 hours. Miss Leavitt’s: 4 days 13 hours.” PUT IT ON THE CHART: a stroke appears, mid-height. 1:40 Star two is tiny and fast: the date view already shows many teeth; I find it at about a day and a quarter. On the chart it goes low and left. 2:30 Star three is big. I sweep from the start and find it far to the right, a month. On the chart: high and right. I think: bright means slow? 3:20 Star four, middling: I start the dial in the middle, and find it near eight days; it lands between. 4:00 Star five is the biggest yet: I drag straight to the far end. 127 days. The strokes climb in a line. 4:40 Star six, small: I go left first. Found in seconds. 5:10 The seventh: “Only its light. Set the beat you expect.” The preview stroke slides along as I drag; I put it on the line. The plates come in folded at my guess: nearly a tooth. I tune it: 66 days; mine was 60. 5:50 Her other eighteen drop in in violet, her two straight lines are ruled through; I flip to her Figure 1 and the line bends into a curve, flip back. The verdict, and a line saying who did the work.
B (runner-up was C’s sibling, B). 0:00 A star with a size slider. I drag; it brightens and blinks slower. 0:30 Aim: match the plates. I match. 1:30 I match again. Nothing has been found; I was told. The walkthrough stops here.
10. The pick, and why
A, The Plate Reader. The player does what the paper did, with the paper’s own difficulty (irregular plates, an unknown beat), and finds the paper’s pattern with their own hands; then the pattern turns into a tool, which is the strongest sign that it was discovered and not told. The sentence afterwards has no “guess”: I timed the blinking stars, the bright ones were the slow ones, and I timed the last one from its light. Originality: no game in the table has had the player tune a hidden period by folding data, a tactile, radio-dial sort of act.
Author test. An astronomer knows the answer and still has to find the beat; the snap of a fold is satisfying however often one has seen it. The alias (a beat twice hers gives two teeth) is the real trap of the trade and stays in.
Runner-up: C, The Heartbeat. Switch if the author wants a thirty-second wordless piece for a talk and accepts a time-lapse whose speed changes between stars, with a dial that says so.
What the reckless concept would have given. Rhythm in the body: feeling the big stars heave and the small ones flutter. Kept: the ringed star on the plate pulses at its own pace, all the time, in a fixed time-lapse, so the bright ones visibly breathe slower. Given up: tapping, and a screen without words.
11. The shape
Seven rounds, one long session, no score.
- Harvard 1425 (4.547 d, mid-bright) — first reading watched: the plates arrive one at a time with a caption.
- Harvard 1505 (1.25336 d, the faintest at maximum) — fast; teeth visible already in the date view.
- Harvard 823 (31.94 d, bright) — far right; with three strokes the slope is visible.
- Harvard 1374 (8.397 d) — lands between.
- Harvard 821 (127.0 d, the brightest) — the far end.
- Harvard 1436 (1.6637 d, faint) — found fast by a player who has the law.
- Harvard 824 (65.8 d, second brightest) — the prediction close: set the beat from its light, then the plates fold at the guess, then tune and write.
Then: her other eighteen in violet ink, her two lines, the Figure 1 / Figure 2 switch, the verdict. Order reason: the first three span the range so the discovery arrives by about the third star; four to six reward using it. Presets are the paper’s own stars; nothing is invented except the plates.
12. Controls, mechanism, feedback, moment
- Controls. The beat dial: a coarse slider from 1 to 128 days, doubling ticks evenly spaced (1, 2, 4 … 128), and a fine slider of ten per cent either way, reset to the centre when the coarse one moves. Every round starts with the dial at 8 days. WRITE IT DOWN commits. A small by date / by beat switch on the record panel.
- Mechanism (reconstructed). For each played star, 22 plates on irregular dates spanning four to six of its cycles, brightness from a light curve shaped as the paper describes (a rise over the last sixth of the cycle, a slow fade that lingers near minimum) between Table I’s maximum and minimum, timed from Table I’s epoch, plus eye-estimate noise of about a tenth of a magnitude. Folding is phase = ((t − epoch)/P) mod 1, the record shown over two beats.
- Feedback. The fold (dots line up or smear as the dial moves); the ghost (her curve in violet over the player’s dots after commitment); the drop (a stroke on the chart, faintest to brightest, at the player’s beat; her mark in violet if it differs).
- Moment. Round 3: the bright star lands high and right and the three strokes form a slope. Round 5: the player drags straight to the far end because the star is huge. Round 7: the player places the star on the line before any plate is seen.
- The game’s voice after each write: what you wrote, what she wrote, and in words whether yours is hers, a multiple of hers (two or three teeth per beat), or neither.
13. The look
Register: the plate on the light box (Austere family: a scientific plate, a museum label). From the paper’s setting: the Harvard plate stacks, where the Cloud exists only as glass negatives, stars as black dots on clear glass, read on an illuminated stand in a dim room, numbers written in ink. The game’s whole world is that glass, lit, in a dark room.
- Palette: velvet
#1F1822 (the dark room, the page ground); glass #E9E6EE (every lit plate and chart, lavender-white); silver #17141B (the emulsion: star dots, the player’s dots and strokes, type on glass); violet ink #6A3FA0 (Miss Leavitt’s hand: her numbers, her curve, her marks, the ring round the star; on velvet the lighter #B49BE0); lilac #A99CBD (hairlines, grids, secondary text).
- Type: Spectral (Production Type, SIL OFL), a fine screen serif, roman for text, italic for everything in her ink; Josefin Sans 600 (SIL OFL), small spaced capitals, for labels and buttons. Both embedded as data.
- Buttons: pills, fully rounded ends, a 1 px glass hairline on velvet, Josefin capitals spaced; the primary action (BEGIN, WRITE IT DOWN, NEXT STAR) filled violet with glass lettering. No shadows.
- The one visual rule: the sky is black on glass, her hand is violet, and nothing appears in violet until you have written yours.
- Play screen at 380 px. The aim, one line in italic, with ? at the right. A small capitals line: STAR 2 OF 7 · HARVARD 1505. The plate (348 × 120, glass): the Cloud’s crowd of fine specks, the star ringed in violet and pulsing at its own pace in a fixed time-lapse; answers “which star, and how bright is it?”. The record (348 × 210, glass): dots on a ruled field, brighter up, by date or by beat, two beats separated by a hairline; answers “is it lining up?”. The dial: the beat written large in Spectral (4 days 13 hours), the coarse slider with doubling ticks, the fine slider; answers “what beat am I trying?”. The result line, then the button. The chart (348 × 300, glass): beat across in doublings, brighter up, one black stroke per star from its faintest to its brightest; answers “what do the stars have in common?”.
- Verdict screen: the chart full width with all twenty-five and her two lines, the Figure switch, the three lines, the Leavitt note, ANOTHER NIGHT AT THE PLATES, How to play, About, signature.
14. What is cut
The comparison-star sequences and the magnitude scale; the provisional-scale caveat (on About); the epochs as numbers; the eighteen stars not played (they arrive in her ink); the minima line as something the player fits (drawn by her); UY Cygni, spectra and the “two fundamental questions” (one line on About); the 1777-star catalogue; all magnitudes as numbers (brightness is a dot size and a height).
15. Divergences
- The plates are ours. The paper prints no plate-by-plate estimates; the game draws 22 per star from Table I’s maximum, minimum, epoch and period with a light-curve shape from the paper’s words. Her plates spanned years and many hundreds of cycles; ours span four to six cycles so a period can be found by eye with a thumb.
- No magnitudes on screen. Brightness is shown as dot size and height; direction kept (brighter up), the 0.48 slope said in words (three times the blink, two and a half times the light).
- Brightness stroke at the player’s beat. If the player writes a wrong beat, the stroke goes where they put it, with her mark beside it, so the chart is honest about their reading.
- Time-lapse on the plate. The pulsing ring uses one fixed rate for all stars (about two seconds a day, capped so the slowest still visibly moves), so the comparison of paces is true within the screen.
Mechanism kind and claim boundaries
Reconstructed. The game runs a constructed plate-generating process reproducing Table I’s periods, maxima, minima and epochs and the light-curve shape the paper describes.
- Demonstrates: that from irregular plates a period can be found by folding, and that the 25 stars of Table I, set by period and brightness, fall on the paper’s straight lines.
- Assumes: the individual plates, their dates and errors, and the exact light-curve shape; the time-lapse.
- The paper establishes: the 25 periods and brightness ranges, and the straight-line relation between brightness and the logarithm of the period, about 0.48 per magnitude, on a provisional magnitude scale. It argues, but does not show, that the stars are at nearly one distance. It does not measure any distance.
The discovery comes from the paper, not from the rules. The chart is built from Table I’s brightness and the beats the player writes; the beats are recoverable because the plates are drawn at her periods. No scoring, objective or constraint of ours produces the slope. The prediction round’s yardstick (how near the guess was) is ours and is said in words only.
Calibration plan
- Table I as transcribed: the least-squares slope of maximum and of minimum magnitude on log period lies within ±0.25 of 1/0.48 ≈ 2.08 magnitudes per decade; the circular’s “about 0.48” is recovered as 0.40–0.60.
- Her residual columns: with lines of the fitted slopes, residuals recomputed from the transcription match the printed Res. M and Res. m to within 0.25 for every star (a check on misread digits); report the worst.
- Average range (min − max) is 1.2 ± 0.15 magnitudes.
- Plates: over 300 seeds per played star, folding at her period gives the lowest dispersion of a grid from 1 to 128 days, apart from integer multiples; the fold at half her period is visibly worse; the fold at her period ±3% is still recognisable for the declared span (dispersion below a threshold).
- Robustness of the discovery to our choices: with a symmetric (sine) light curve and with noise doubled, the true period is still the best non-multiple fold for every played star.
- The words: the day-and-hour formatter gives no decimals; the comparison words (hers, twice hers, three times, half, near, off) classify a battery of beats correctly.
Calibration result
node calibration.js — all 42 checks pass (full output in calibration-result.txt). The headline numbers:
- Least squares on the transcribed Table I: 0.496 in log period per magnitude at maximum, 0.488 at minimum; the lines implied by her printed residuals give 0.487 and 0.491. The circular says “about 0.48”.
- Printed residuals recomputed from the transcription: Res. M agrees for all 25 stars (worst 0.08); Res. m agrees for 24 (worst 0.10). Harvard 1374 is printed −0.3 where its own magnitudes (13.9, 15.2, both clearly legible on the scan) give +0.3: a sign slip in the 1912 table. The game uses the magnitudes as printed and says so on About.
- Average range: 1.18 magnitudes (paper: about 1.2).
- Plates: over 280 star-seeds, the cleanest fold that is not a multiple of her period lies within 4% of hers in 279; the half-period fold is visibly rougher in 279; folds 1.5% off hers stay recognisable in all 280. A beat 1.37 times hers gives roughness 1.08 (no order is 1).
- Robustness to our choices: with a symmetric light curve the best fold is hers in 105/105; with the eye-noise doubled, 98/105; with 20 plates instead of 26, 101/105.
- The chart’s slope does not come from our plates: the seven played stars alone give 0.53, the first three already slope the right way.
- Simulated players (careful, doubler, sloppy, random) all reach a verdict; the careful one is told their beats were hers and the seventh set within a few days.
Build log
- Mechanism first, headless (
mechanism.js), checked by calibration.js. The first fold metric (binned dispersion with 22 plates) let near-misses and, for the 65.8-day star, a chance alias at about 2 days look cleaner than the true period; switched to string length (the eye’s own test: are neighbours close?) and 26 plates, after which the true beat is the cleanest fold everywhere. The seventh star changed from Harvard 824 (65.8 d) to 822 (16.75 d), because 824 sits 0.4 magnitudes off her line and a player reading the line correctly would have been told they were far off.
- Loop played headless (
loop.js, loop-log.txt): a guess from the light alone lands within the fine dial’s reach but not close enough for the dots to line up untouched, so round 7 starts the dial at the guess rather than claiming the fold is already clean.
- Skin per §13. Fixes after the first screenshots: tick labels placed under the thumb’s actual travel; axis words moved clear of panel tags (“fainter ↓” to the empty lower right of the chart); the seed and a legend line moved under the final chart; the verdict’s first line rewritten to name only the exceptions.
session-test.js: a full game in headless Chromium at 380 px (seed 1912), with one round written at twice her beat, the help page opened and closed mid-round, the Figure 1/2 switch, About, and a second night. No page errors, no horizontal scroll. The only console error is the blocked counter script.
- Self-review with How to play closed: the aim can be said from the start screen; each round screen says what the glass, the box and the dial are and what a right beat looks like; the result is stated by the game before her number; the discovery can be read off the top of the end screen. Unlike anything in the table: a lit glass negative in a dark room, a fold as the instrument.
Session length: an estimate of five to eight minutes until a person has played it.
Revision 1 (6 Nov 2026): for a smart eleven-year-old
Steffen’s note after a first look: the play screens were too dense (“Now turn the dial to try a beat…” did not say plainly what to look for), and the game should be playable by a smart eleven-year-old. Every line on screen rewritten for that reader, mechanism unchanged:
- “Plates” became glass photos, “beat” became blink time, “Miss Leavitt” became Henrietta Leavitt, then Leavitt. The aim is now Time how long each star takes to blink, and find the pattern.
- Round 1 now starts in date order and stays there until the dial is first touched. Then the dots visibly slide from their dates to their places in the blink, with the line “See? The dots jumped. Now they are sorted by your blink time…” so the switch the instructions talk about is something the player watches happen.
- Instructions are one idea per line (wrong guess: messy; right guess: the same shape twice, a quick jump up and a slow slide down). Round 2 warns about two jumps in each half (twice too long).
- Result lines in the game’s voice are short and plain (“Spot on!”, “Very close!”, “Yours is twice as long as hers”); the guess on the last star is judged in plain words (“Good guess: just a few days off”).
- The verdict: what you found, the pattern, and why it matters (same cloud, same distance; a star’s blink tells how bright it really is, and so how far away it is; Leavitt hoped someone would find out). “Who did the work” replaces “Whose circular”.
- How to play rewritten in the same words. About left for adults.
Calibration: all 42 checks pass (the verdict-words check updated to the new wording). Browser session at 380 px: no errors, no horizontal scroll.
Revision 2 (6 Nov 2026): the box shows the answer to the eye
Steffen, playing revision 1: he still did not see what to do, and suspected his eye only ever caught the largish black dots. Diagnosis: the photo and the box used the same black dots with two meanings (size = brightness on the glass, height = brightness in the box, size meaning nothing), and at the right blink time most dots lie in a flat band near the bottom, so the right fold and a wrong one looked much alike unless you knew what to look for. Changes, mechanism untouched:
- The thread. In the sorted view a thin line joins the nights in order of your blink time. A wrong blink time draws a jagged scribble; nearing the right one it calms; at the right one it is a clean wave, a quick jump up and a slow slide down, twice. This is the roughness measure of the calibration check made visible: it describes the picture and grades nothing.
- Dots sized like the star. Each night’s dot in the box is as big as the star was that night (and higher when brighter), in a softer graphite than the stars on the glass, so the box reads as the ringed star night after night, and at the right blink time the big dots stack into the peak.
- Words follow the picture: “Turn the dial until the scribble becomes a neat zigzag.” The first-touch line now introduces the line; round 2 warns that two jumps in each half means twice too long; How to play says the same.
Browser session at 380 px: no errors, no horizontal scroll. Calibration unchanged (42/42).
Revision 3 (6 Nov 2026): one glass, and brightness in colour
Steffen: “the brighter the star, the bigger the dot doesn’t work, so it stays all guesswork” — and, his suggestion, the gap may again be between what the builder can measure and what a human eye perceives; work with colour. Diagnosis: each round showed one star alone, so its size could not be compared with anything, and the size scale was narrow; the seventh star’s guess had nothing on the glass to compare against. Changes, mechanism untouched:
- One glass for the whole game. Every star you time keeps its place on the same photo with a thin ring and the blink time you found; the new one has a thick ring. During play they all blink at their own pace, so the bright ones visibly breathe slowly next to the faint quick ones.
- Brightness in colour as well as size, on one scale for the whole game: small and pale pink is faint, big and dark plum is bright (a five-dot key in the corner of the glass). The size range is twice as wide as before. On the chart each of your stars is a thick line running from its faint colour at the bottom to its bright colour at the top, so a dot on the glass and its line on the chart are visibly the same star. Leavitt’s own stars stay thin violet.
- The seventh star is guessed with all the stars on the glass frozen at their brightest, each labelled with your blink time: is it bigger and darker than the others, or smaller and paler?
- Words updated to match (“Small and pale pink: a faint star. Big and dark: a bright star.”). Black is now kept for the ordinary stars of the cloud and for the line in the box.
Browser session at 380 px: no errors, no horizontal scroll. Calibration unchanged.
Revision 4 (9 Nov 2026): a full sweep, played as the eleven-year-old
A fresh pass through every screen, mechanism untouched. The main finding was not words but layout: on a phone the dial sat far below the box, so while she dragged the dial the dots she was meant to watch were off screen. Control and feedback now share the screen.
- Round screen reordered: glass (shorter), one line, the box, the dial directly under it, then the explanation, the result, the button. The dial is more compact (its name sits inline with the value). The session test now asserts, in every round, that the box and the dial are both on screen without scrolling at 380 × 700, a small phone with the browser chrome taken off.
- First round: a one-line nudge between box and dial, “Now move the dial, and watch the dots”, shown until the first touch; the paragraph under the dial is one sentence.
- The guess round: the chart sits where the box sits, with the dial under it; the hint now says how to reason (“find the two stars it sits between, bigger than one and smaller than the other, and guess a blink time between theirs”); the dashed stroke’s label moved below the stroke so it cannot collide with a star’s label.
- Chart screen: at three stars, “Look at where they sit. Do you see a pattern?”; before the last star, a line saying that the next blink time is guessed from brightness before any photos, so look at the chart first.
- Plate captions from round 3 on cut to one line.
Browser session at 380 × 700: no errors, no horizontal scroll, box and dial on screen together in all seven rounds. Calibration unchanged (42/42).
Revision 5 (9 Nov 2026): the last star in a line-up
Steffen at the last stage: “one last star in the big ring, but there is no big ring, just one star that is a bit darker, and it doesn’t sit in between.” All seven rings were the same size, the current one only a heavier line; and the dots at their brightest differed by a couple of pixels, so the new star could not be told apart, let alone placed. Changes, mechanism untouched:
- The star being timed now has a clearly bigger ring (radius 24 against 17) in every round; “thick ring” became “big ring” in the words.
- In the last round the glass becomes a line-up: all seven stars at their brightest in a row from faintest to brightest, blink times underneath, the new one marked THIS ONE with a ? in its slot between two timed stars. The hint says: look at the two on either side, guess a time between theirs.
- How to play gained “A good way to do it”: slide the top dial slowly, stop where the line snaps tidy, check one hump per half (two means halve it), then fine-tune until the dots hug the line with no strays; once the chart has a pattern, use it to know where to start.
Revision 6 (9 Nov 2026): the ending zoomed out; off lines with direction
Steffen, after a full play: the closing line “your black ones are scattered” referred to nothing he could see; wanted the end to zoom out — what she measured and concluded, what the player measured and concluded, how they differ, and what would follow if the player’s poorer measurements had stood, “a kind of false physics”. Also: too many “not hers”; vary them and say which way the error goes.
- The verdict is four headed parts, computed from the player’s own seven: What she measured and found (fixed); What you measured and found (how many were hers, how many too long/too short, what the player’s own chart shows, the last-star guess); Where you and she part; If your numbers had stood. The player’s chart is classified by a line fitted through their seven at the beats they wrote: hers (all seven hers), shifted (times wander little but sit at one factor off hers: same pattern, crooked yardstick — every distance measured with it would come out short or long by a stated fraction, worked out from her slope: twice-too-long beats give distances about a quarter too short, half-too-short beats about a third too long), ragged (line the right way, N stars off it: a yardstick you cannot trust), rough (a trend lost in scatter: a rumour, not a law), none (brighter not slower: the cloud would have told us nothing). The what-if is labelled as the game’s, not the paper’s. The final chart draws a dashed line through the player’s seven whenever they are not hers.
- Per-star off lines now say the direction and rough size (a bit / quite a bit / far too long or short) in one of four voices; multiples add “look for one hump in each half”. The chart screen’s violet-line note says whether her stroke sits left of yours (yours too long) or right (too short).
- Verdict screen reordered: chart first, then the four parts, then who did the work.
Calibration extended to 49 checks (doubler → shifted, about a quarter too short; halver → shifted, too long; two strays → ragged; off lines carry direction and vary); all pass. Browser session clean.
Renamed (11 November 2026)
Renamed from Leavitt’s Cloud to Blink with Leavitt, at the author’s decision to let some titles say what the player does while keeping the name. The game file is blink-with-leavitt.html; a one-line page at leavitts-cloud.html sends old links to it. Nothing in the mechanism changed.
Renamed back (15 November 2026)
Back to Leavitt’s Cloud, so that three titles in a row on the Pantheon page do not all end “with”. The game file is leavitts-cloud.html again; blink-with-leavitt.html now sends links to it. Nothing in the mechanism changed.