← negishoyu.dev

WORLDLINES

Episode 1

Sources, data and method

“Does the Sun Shape the Words for Blue and Green?”

Episode
"Does the Sun Shape the Words for Blue and Green?"
Video
https://youtu.be/ZlPUUaq0RtE
This page
https://negishoyu.dev/worldlines/ep01/sources/
Version
for the v9 render (14:43.9) as uploaded, with its English original audio and seven other audio tracks. Timestamps (m:ss.s) come from the v9 timing table; the uploaded video is the same render.
Script
Japanese master script final15 (file hash af317e961f31). English lines are quoted from the final English dub, which is the video's original audio track (plan/l10n/en/script.tsv, 2026-10-07).
Last updated
2026-10-10

Worldlines is an animated series in which two characters test a question against real data. This page lists, for every factual claim and every number shown or spoken in Episode 1, where it comes from, what we computed ourselves, and what it does not show.

Corrections

We do not delete mistakes or quietly rewrite them. When we find one, we add a row below with the date, the time in the video, what was wrong and what is correct, and the row stays on this page for good. If a claim further down changes, its old wording stays in this table. To report a mistake, in the facts or in any of the translations, please leave a comment under the video.

DateTime in videoLineWhat was wrongWhat is correct
———No corrections yet.—

How to read this page

Each claim lists:

The English track calls the two research teams "the map team" (Josserand et al., 2021 and 2023) and "the eye team" (Hardy et al., 2005 and 2023).

Claims

0:00What color is this light?

C01In Japanese, "ao" (blue) also names some green things

  • Time: 0:04.7 · 01.04 / 0:09.4 · 01.05 / 0:14.7 · 01.06. Recalled at 8:09.9 · 13.13, 11:23.3 · 17.12c, 13:39.3 · 20.08 and 13:45.8 · 20.09.
  • What we say / show: "But in Japanese, this is called a 'blue' light!" / "Other green things get called 'ao,' Japanese for blue." / "Aoba too. Ao-ringo too." On screen: a greenish traffic light, a branch of young leaves and a green apple, with tags 青信号 ao-shingō, 青葉 aoba and 青りんご ao-ringo.
  • Class: ILLUSTRATION (boards P01, P02).
  • Source: 青 (ao), Digital Daijisen, via Kotobank, https://kotobank.jp/word/%E9%9D%92-192607 — full text (dictionary). The word covers colors from indigo to yellowish green, and is also short for 青信号, the "go" light. [J1]
  • Our computation: none.
  • License: citation only.
  • Known limits: Today's Japanese has separate basic words for blue (ao) and green (midori): Kuriki et al. (2017), Journal of Vision 17(3):1, PubMed 28249298 — abstract only [J8]. The video does not explain why the traffic light is called "blue"; we leave out the history of that name because our sources for it are secondary [J9–J12].

C02Many languages use one word for blue and green, and that is not rare

  • Time: 0:17.1 · 02.01 / 0:24.6 · 02.03 / 0:30.2 · 02.03b
  • What we say / show: "Japanese isn't alone in calling some green things 'blue.'" / "There are languages that always use one word for blue and green." / "And they're not rare." On screen: a paper globe with paper figures whose speech tags cover blue and green with a single tag. No numbers.
  • Class: ILLUSTRATION (P01–P03).
  • Source: Our count of the language dataset of Josserand, Meeussen, Majid & Dediu (2021) (Data D1): of 142 languages, 60 have no basic word for "blue" alone and 82 have one [S1]. Background: Berlin & Kay (1969), Basic Color Terms (University of California Press), which compared color words across 98 languages — secondary (we have not read the book) [W1].
  • Our computation: count of the field exists_blue in Data D1. The numbers are not stated in the video.
  • License: data CC BY-SA 4.0 (Data D1).
  • Known limits: "Not rare" rests on one sample of 142 languages, which is not random: more than half (79) are from Eurasia [M2]. The dataset records whether a language has a basic word for "blue" alone. A "no" usually means blue shares the word for green, but some of these words also cover black or dark shades, and the dataset does not separate them [M12].

Ideas the characters test (not claims) · 0:57.7–1:24.7

  • 04.03–04.09 (0:57.7–1:20.6): Comet's guess — where the sun is strong, sky blue and green leaves "would look really different", so people would want two words; where it is weak, one word would do. This is a character's idea. We do not claim that sunlight changes how colors look. Class: ILLUSTRATION (P05, two panels with a big sun and a small sun).
  • 05.01 (1:24.7): Sescha turns it into a prediction: "Rookie, your idea predicts those languages are common where the sun is weak." The episode then tests it (C04–C08).

C03A satellite record of sunlight, and a record of color words

  • Time: 1:33.8 · 05.03 / 1:38.7 · 05.04 / 1:42.9 · 05.05
  • What we say / show: "For sunlight, there are satellite records worldwide." / "For color words, there are language records worldwide." / "Overlay both records, then count and check." On screen: a satellite on an arm circling the day side of the Earth; a bundle of records sliding onto the satellite record.
  • Class: ILLUSTRATION (P07, P08).
  • Source: Sunlight: NASA Earth Probe TOMS surface UV (C04, Data D2) [M4]. Color words: Josserand et al. (2021), who extended and re-checked earlier databases — full text [M1, M3].
  • Our computation: none here; see C04–C07.
  • License: —
  • Known limits: "Satellite records" is shorthand. The satellite measured ultraviolet light coming back from the Earth; the surface values we map are NASA's estimates computed from those measurements (C04) [M4].

2:01Strong sun or weak sun?

C04The purple map: UV-B from NASA satellite data

  • Time: 2:15.4 · 07.10 / 2:20.4 · 07.10c / 2:25.1 · 07.10e / 2:28.1 · 07.10f / 2:32.3 · 07.07 / 2:39.0 · 07.07b. Source card on screen from 07.10 to the end of 07.07: "NASA TOMS UV-B 1998".
  • What we say / show: "Look. The satellite measured sun strength by the amount of UV." / "UV is part of sunlight. It's what gives you a tan." / "Stronger sun, darker tan, right?" / "Yes. Where the sun is stronger, there's more UV." / "Purple on this map shows UV. So the darker the purple, the stronger the sun." / "Darker near the equator!" On screen: a world map in five shades of purple with a legend running from a small sun to a big sun; two small windows showing light and heavy sunburn.
  • Class: MEASURED (P10). The sunburn windows carry their own ILLUSTRATION tab (P11).
  • Source: NASA Goddard Space Flight Center, Earth Probe TOMS, surface UV: local-noon spectral irradiance at 305 nm and 310 nm (mW/m²/nm), daily, 1° × 1° grid, 1 January–13 December 1998. https://acd-ext.gsfc.nasa.gov/anonftp/toms/eptoms/data/uv_ery_4_wavelengths/ and its documentation file 1README.UV.txt — documentation read in full and the data files themselves, accessed 2026-10-04 [M4]. Method behind the surface estimates: Herman et al. (1999), J. Geophys. Res. 104(D10):12059–12076, doi:10.1029/1999JD900062 — bibliographic only. Josserand et al. (2021) used the same NASA TOMS UV data for 1998 — full text [M4]. "UV is what tans the skin; stronger sun, more UV": everyday knowledge.
  • Our computation: annual mean of each 1° cell over all valid days at 305 and 310 nm together (the same averaging as the 2021 study), drawn in five shades; details in Method M1–M2. Data version: Data D2. Code: build/tools/nasa-uvb/fetch.py, build/tools/nasa-uvb/make_grid.py, build/tools/data.mjs, build/web/src/boards/P10.js @ 2c90e38ab; build/web/src/boards/w2-map.js @ bb00a458b.
  • License: data from NASA-led missions are CC0 unless marked; NASA asks to be credited (Data D2). Not endorsed by NASA. No NASA logo or insignia appears in the video.
  • Known limits:
    • The surface values are estimates. NASA computed them from the satellite's measurements of backscattered UV together with ozone, clouds, aerosols, surface reflectivity, elevation and sun angle. They are not ground measurements. "The satellite measured" (07.10) means "a record derived from satellite measurements" [M4].
    • One year, 1998, and only 347 days (14–31 December are not on the server). The mean uses only days with values; in polar night there are none, so high-latitude cells average the sunlit season.
    • Two wavelengths at local noon: an index of UV-B strength, not the whole UV-B band and not a daily dose.
    • UV-B follows the height of the sun but also ozone, cloud and altitude, so "the darker the purple, the stronger the sun" is a simplification. The darkest shade covers much of the tropics and Arabia, and also high ground such as the Andes, Tibet, the Ethiopian Highlands and the Mexican plateau; some cloudy areas (inside the Amazon and Congo basins, the Guinea coast, southern China) are lighter than their latitude alone would suggest.
    • The legend shows only the order of the shades; the shade widths are not equal (Method M2).
    • Cells are 1° wide; their edges show as small steps when the map is enlarged.

C05Each pin is one language: 142 languages, recorded by researchers in Europe in 2021

  • Time: 2:43.4 · 07.11 / 2:46.3 · 07.12. Card: "Josserand et al. 2021".
  • What we say / show: "One pin is one language." / "In 2021, researchers in Europe laid records of 142 languages over this map." On screen: one example pin with its word tags, the number 142, and 142 rings for pins.
  • Class: MEASURED (P12).
  • Source: Josserand, Meeussen, Majid & Dediu (2021), Scientific Reports 11:19095, doi:10.1038/s41598-021-98550-3 (PMC8476573) — full text. Four researchers in France, the Netherlands and the UK; published September 2021; 142 populations in 32 language families; the data extend and re-check earlier databases; analysed against UV-B and other factors. [M1–M3]
  • Our computation: each pin is drawn at the coordinates given for that language in Data D1.
  • License: data CC BY-SA 4.0 (Data D1).
  • Known limits: The sample is uneven: Indo-European 41, Atlantic-Congo 19 and Afro-Asiatic 13 languages; Eurasia 79, Africa 31, South America 12, North America 9, Papunesia 9, Australia 2 [M2]. A language is drawn as one point, although it is spoken across an area. "This map" on screen is our rendering of the same NASA data (C04), not a figure from the study.

C06What a pin records, and what the "joined" mark means

  • Time: 2:56.6 · 07.15 / 2:59.6 · 07.16 / 3:03.3 · 07.17 / 3:07.5 · 07.17b / 3:10.8 · 07.17c / 3:13.6 · 07.17d
  • What we say / show: "Whether it has a word just for blue." / "Languages with one get the split mark." / "Languages without one get the joined mark." / "Then what do they call blue?" / "Usually, the same word as green." / "So joined marks mostly mean languages that don't split blue and green." On screen: the legend — a cut circle (separate tags for blue and green) and a mixed circle (one tag for both).
  • Class: MEASURED (P12, P13).
  • Source: the dataset's field exists_blue: whether the language has a basic term for "blue" alone [M1]. "Usually the same word as green": Hardy, Werner, Regier, Kay & Frederick (2023) write that such languages have either one term spanning green and blue, or one term that merges green and blue with black and other dark shades; Josserand et al. (2021, legend of Fig. 1E) also describe it as a single term for blue and green — both full text [M12].
  • Our computation: none.
  • License: data CC BY-SA 4.0 (Data D1).
  • Known limits: The dataset has only yes or no. The "no" side includes words that also cover black or dark shades, and no one has counted how many there are in this dataset; so we say "usually" and "mostly", never "all" [M12].

C07Our rough count: about one in ten where the sun is weak, over half where it is strong

  • Time: 3:29.4 · 08.03 / 3:34.4 · 08.04 (marks drop into the three boxes; 2.5 s, no dialogue) / 3:37.2 · 08.05 / 3:45.0 · 08.06 / 3:49.5 · 08.07 / 3:55.4 · 08.08
  • What we say / show: "We sort the pins into three groups, from weak sun to strong." / "At a rough count... in the weak-sun group, joined marks are just over one in ten." / "In the strong-sun group... more than half." / "...Wait, those languages are more common where the sun is strong?" / "Opposite of my guess!?" On screen: 142 pins go into their rings on the map; three boxes rise (small, middle and big sun) and the pins' marks drop into them; a band across the boxes marks the halfway line.
  • Class: MEASURED (P14). The board shows our count of published data.
  • Source: our count [S2], from Data D1 and D2.
  • Our computation: joined marks (no basic word for "blue" alone): weak 6 of 47 (13%), middle 25 of 47 (53%), strong 29 of 48 (60%). Put the other way, a word for "blue" alone: 41/47, 22/47, 19/48. Method M1; code build/tools/data.mjs @ 2c90e38ab, build/web/src/boards/P14.js @ 83f7e96df.
  • License: derived from CC BY-SA 4.0 data (Data D1, share-alike).
  • Known limits: A rough count. It does not correct for related or neighbouring languages counted as separate cases; C08 gives the study's corrected result. Splitting into thirds is our choice. The middle group is shown in its box but not mentioned.

C08Related languages repeat the same answer; the study corrected for this, and the pattern held

  • Time: 4:00.2 · 09.01 / 4:03.9 · 09.02 / 4:08.9 · 09.04 / 4:11.9 · 09.05 / 4:16.3 · 09.07 / 4:21.2 · 09.08b / 4:23.0 · 09.09. Card: "Josserand et al. 2021".
  • What we say / show: "But... maybe it's just a coincidence?" / "Languages have relatives. Relatives and neighbors grow alike." / "Alike... and why is that a problem?" / "Then you count the same answer over and over." / "So the map team took out the overlap, then compared." / "And after that...?" / "Even so, the stronger the sun, the more joined marks there were." On screen: pins of related languages tied together with threads and bundled; an arrow rising from the weak-sun box to the strong-sun box.
  • Class: MODELED (P15). The board shows the study's model result: the method is a Bayesian mixed-effects logistic regression with language family and macroarea as random effects, and the video states it in one line at 09.07 ("So the map team took out the overlap, then compared."). The marks in the boxes are the unadjusted counts (as in C07); the threads only illustrate how related languages were grouped.
  • Source: Josserand et al. (2021) — full text [M5, M6, M13]. Their main analysis is a Bayesian mixed-effects logistic regression with language family and macroarea (a continent-sized region) as random effects. Stronger UV-B goes with a lower chance of a word for "blue" alone: β = −1.12 (95% interval −1.75 to −0.55); this model predicts correctly 91% of the time. The study calls UV-B the most important predictor of a word for "blue", and also writes that no single factor explains the color vocabulary.
  • Our computation: none. The numbers here are the study's own.
  • License: citation only.
  • Known limits: In the same study, the number of speakers (β = 0.31) and the distance to the nearest large lake (β = −0.56) also mattered [M5]; a machine-learning comparison lists UV-B, number of speakers, climate and humidity, latitude and distance to lakes as useful predictors, without ranking them [M7]. The study's path analysis did not include the family and area correction, as the study itself says [M13]. "Neighbours" here means languages in the same continent-sized macroarea [M13]. The video gives no numbers for this result.

4:28Eyes, or culture and history?

C09Counting shows where, not why

  • Time: 4:28.5 · 10.01 / 4:34.6 · 10.03a / 4:42.0 · 10.04
  • What we say / show: "I wonder why those languages are more common where the sun is strong." / "So far, we've counted pins to check where languages that don't split blue and green are common." / "But it turns out comparing numbers alone can't explain why."
  • Class: MEASURED tab of P15 still on screen through 10.03a; no tab from 10.04 (P16, two cards).
  • Source: the 2021 study: no single factor explains the color vocabulary — full text [M6]. Both teams in the 2023 exchange accept that the overlap exists and disagree about its cause — full text; this summary is ours [D7, D8].
  • Our computation: none.
  • License: —
  • Known limits: —

C10Two lines of explanation: the eyes, or life and history

  • Time: 4:50.4 · 10.05 / 4:56.1 · 10.07 / 5:03.6 · 10.08b / 5:04.9 · 10.09
  • What we say / show: "Now let's look at this from other angles. There are two." / "One is the eyes. The idea that years of strong sun change the eyes, and how colors look." / "And the other?" / "Way of life and history. The idea that eyes don't matter, but daily life and language history do." On screen: two cards — an eye, and a house with a scroll — with a closed door between them.
  • Class: no tab (P16).
  • Source:
    • Eyes: UV-B exposure is thought to make the eye's lens yellow sooner (Josserand et al. 2021, introduction; Hardy et al. 2023) — full text [E2]. The lens idea was proposed by Lindsey & Brown (2002), Psychological Science 13(6):506–512, doi:10.1111/1467-9280.00489 — abstract only [E3]. The 2021 study says its results strongly support that proposal — full text [M14].
    • Life and history: Hardy et al. (2023) argue that color words follow how much people need to tell colors apart — full text [D2].
    • Still disputed: whether UV has an effect separate from culture — full text; this summary is ours [D8].
  • Our computation: none.
  • License: citation only.
  • Known limits: "Two angles" simplifies. The sides overlap: the map team also agrees that the lens alone is unlikely to explain the pattern and that other factors matter [D6]. The 2021 study also found effects of the number of speakers and of lakes, which are neither the eye nor sunlight [M5].

C11An example: dyeing cloth blue and green

  • Time: 5:12.8 · 10.09a / 5:15.5 · 10.09b / 5:22.0 · 10.09c
  • What we say / show: "A way of life can change words?" / "Say people dye cloth blue and green. They'd often need two separate words." / "So their words for blue and green tend to split." On screen: a small drawing of dyed cloth beside the house-and-scroll card.
  • Class: no tab (P16).
  • Source: Gibson et al. (2017), PNAS 114:10785, doi:10.1073/pnas.1619666114 — abstract only: differences between color vocabularies relate to how much people need to communicate about colors; in the community studied, color words were used more for artificially colored objects [W7]. The map team's 2023 reply agrees that factors such as dyeing technology and history matter — full text [D6].
  • Our computation: none.
  • License: citation only.
  • Known limits: An illustrative example, not a case we studied. It is not a rule that people without dyeing have few color words: the Maniq of Thailand, hunter-gatherers without dyeing technology, have 6–7 basic color terms, used about as consistently as English speakers use theirs — Wnuk, Verkerk, Levinson & Majid (2022), Cognition 229:105223, doi:10.1016/j.cognition.2022.105223 — abstract only [W8].

5:31Is it the eye's lens?

C12The lens yellows with age and lets less blue light through

  • Time: 5:38.1 · 11.02 / 5:40.9 · 11.03 / 5:44.4 · 11.04 / 5:49.7 · 11.05 / 5:56.3 · 11.06 / 6:00.2 · 11.07
  • What we say / show: "Look. Inside the eye, the lens." / "Light goes through to the back of the eye." / "With age, everyone's lens slowly turns yellowish." / "Oh, through yellow sunglasses, blue things look darker, right?" / "Yes. Yellow lets less blue light through." / "So with a yellow lens, the back of the eye gets less blue light." On screen: a paper cut-out cross-section of an eye with blue, green and red light strips; yellow sunglasses.
  • Class: ILLUSTRATION (P16 from 11.01, then P17).
  • Source: Delahunt, Webster, Ma & Werner (2004), Visual Neuroscience 21(3):301–307, doi:10.1017/s0952523804213025 — abstract only: the lens's optical density rises with age, most at short (blue) wavelengths, changing the light that reaches the retina, while color appearance stays relatively stable over the lifespan [E1]. The standard age formula: Pokorny, Smith & Lutze (1987), Applied Optics 26:1437, doi:10.1364/AO.26.001437 — abstract and part of the tables [E11]. Yellow sunglasses: everyday knowledge.
  • Our computation: none.
  • License: citation only.
  • Known limits: Source read at abstract level. The cross-section is schematic. We do not discuss eye health or eye disease [N7].

C13UV is thought to speed up lens yellowing

  • Time: 6:05.3 · 12.01 / 6:09.1 · 12.02
  • What we say / show: "And how do the lens and the sun connect?" / "Long exposure to UV is thought to make the lens yellow sooner." On screen: a purple band runs from a big sun to the lens, and the lens's color strip steps quickly toward yellow; then an eye under a big sun (yellow lens) beside an eye under a small sun (nearly clear lens).
  • Class: ILLUSTRATION (P18).
  • Source: Josserand et al. (2021, introduction) and Hardy et al. (2023) both take this as a starting point — full text; a 2024 review in Frontiers in Ophthalmology, doi:10.3389/fopht.2024.1414483, lists UV as a risk factor for cataract of the lens — abstract only [E2].
  • Our computation: none.
  • License: citation only.
  • Known limits: "Is thought to": the video gives no size for this effect, and the speeds in the drawing are not data.

C14Through a yellow lens, sky blue and leaf green differ less in the light reaching the eye — how they look is another question

  • Time: 6:14.7 · 12.04 / 7:18.4 · 12.04b
  • What we say / show: "Look. Light passes through a yellow lens to the back of the eye. There, blue sky and green leaves differ less." / "The light changes, but how colors look? We don't know yet." On screen: a gauge behind the eye; yellow film slides over its sky-blue and leaf-green pieces and the needle swings toward the smaller-difference side (the gauge has no words); a head-shaped closed door rises (how the colors look is not yet known).
  • Class: ILLUSTRATION (P18). The gauge is drawn, not computed.
  • Source: E1 — abstract only; E3 — abstract only. Keeping "light reaching the eye" apart from "how colors look" is our own rule [N1].
  • Our computation: none.
  • License: citation only.
  • Known limits: Schematic.

C15Sescha's hypothesis, and what the map team actually wrote

  • Time: 6:23.6 · 12.03 / 6:27.9 · 12.03a / 6:33.7 · 12.03c / 6:40.8 · 12.03b / 6:41.9 · 12.03d / 6:45.5 · 12.05 / 6:52.1 · 12.05b / 6:59.1 · 12.08b / 7:04.7 · 12.09 / 7:12.4 · 12.07 / 7:14.5 · 12.08 / 7:23.4 · 12.08c / 7:25.3 · 12.08d. Card at 12.05b: "Josserand et al. 2021".
  • What we say / show: "...So, like you, I came up with a hypothesis." / "Back to the idea that sun changes the eyes. What changes is this lens." / "Strong sun yellows lenses early. Then blue sky and green leaves get hard to tell apart." / "And then?" / "People end up using one word for both." / "Oh, that's just like my first idea! If the difference is small, one word is no problem!" / "The map team also lists this lens as one reason why those languages are common." / "But they also write that UV alone doesn't decide a language's words." / "I went a step further. I think the lens explains the map's results. That's my hypothesis." / "So it's the lens, then!" / "It stays a hypothesis until it's checked." / "Let's check it." / "If blue and green look alike, names should show it." On screen: Sescha's card (a yellow lens leading to the three boxes) pinned to the board.
  • Class: no tab for Sescha's card (P19); the board behind it is ILLUSTRATION (P18).
  • Source: The 2021 study says its results strongly support the proposal that lens yellowing acquired over life works against keeping separate words for "blue" and "green" — full text [M14]. The same study writes that no single factor explains the color vocabulary — full text [M6]. Hardy et al. (2023) also note that the overlap has been proposed to come partly from early ageing of the lens — full text [M14]. For the prediction in 12.08d: in the 2005 study, a screen simulation of a yellowed lens made young people call some "blue" samples "green" — abstract only [E7].
  • Our computation: none.
  • License: citation only.
  • Known limits: Sescha's hypothesis ("the lens explains the map") is stronger than what the researchers wrote; the video presents it as her hypothesis and tests it next (C16–C18). "One reason" (12.05b) is milder than the study's "strongly support".

7:29Do aging eyes change how we name colors?

C16The 2005 experiment: younger and older adults named the same color samples

  • Time: 7:29.4 · 13.01 / 7:31.7 · 13.01b / 7:34.9 · 13.01c / 7:39.0 · 13.02b / 7:41.5 · 13.03 / 7:45.0 · 13.05 / 7:48.8 · 13.08 / 7:55.6 · 13.09b / 7:57.3 · 13.10 / 8:00.2 · 13.11 / 8:04.8 · 13.12 / 8:09.9 · 13.13. Card: "Hardy et al. 2005".
  • What we say / show: "How do we check it, then?" / "Aging alone yellows the lens, too." / "So first, we'll check with older people's eyes." / "There's an experiment like that?" / "There is. An eye experiment from 2005." / "It compared two groups, younger and older." / "Both groups saw color samples from blue to green, one at a time. They named each one." / "And lens yellowing?" / "Also measured, for each person." / "My hypothesis predicts the older group says 'green' more." / "But me... I think the names they use won't change." / "The light looks green, but Japanese calls it 'blue.' Looks alone may not decide names." On screen: two rows of ten paper figures headed by the numbers 23 and 74 (the groups' mean ages; the word "average" is not on screen); eight color chips from sky blue to young-leaf green in equal steps; yellow film beside each row, darker for the 74 row.
  • Class: ILLUSTRATION (P17) for 13.01–13.02b; MEASURED (P20) from 13.03.
  • Source: Hardy, Frederick, Kay & Werner (2005), Psychological Science 16(4):321–327, doi:10.1111/j.0956-7976.2005.01534.x (PMC2586906) — abstract only. Group sizes and ages come from the full texts of the two 2023 papers: 10 younger adults (mean age 23.2) and 10 older adults (mean age 73.9); the optical density of each person's ocular media (the lens and other media) was measured [E4]. The researchers were at the University of California, Davis, and other institutions [E4]. Age and UV yellow the same lens [E1, E2].
  • Our computation: 23.2 and 73.9 are shown rounded, as 23 and 74.
  • License: citation only.
  • Known limits: Twenty people, all in the United States (C30). We have not re-read the 2005 full text; the numbers come from the 2023 papers [E4, E6]. The numbers 23 and 74 are averages, but the screen shows them without that word and the lines do not say the ages. The eight chips are a picture; our ledger does not record how many samples the study used. Whether older eyes can stand in for eyes in high-UV places is exactly what the map team's reply questions (C30) [E10, D7].

C17Result: the names did not change, across a 41-fold range

  • Time: 8:17.1 · 14.01 (1.5 s pause) / 8:18.6 · 14.02 / 8:19.6 · 14.03 / 8:23.8 · 14.04 / 8:30.0 · 14.05
  • What we say / show: "Result." / "Even the older group didn't say 'green' more often." / "Between the most yellow lens and the least, there was a 40-fold difference." / "Even so, the names didn't change." On screen: name tags turn over under the chips, and both rows switch from blue to green at the same chip; a deep-yellow disc and a nearly clear disc, with "×40" between them.
  • Class: MEASURED (P20).
  • Source: Hardy et al. (2023), Scientific Reports 13:1836 — full text: across a 41-fold range in the opacity of the eye's optical media among the participants, the use of the color term "blue" stayed constant (R² = 0.009, p = 0.89). Hardy et al. (2005) — abstract only: the green, blue-green and blue samples were named in nearly the same way by the two groups. [E5]
  • Our computation: 41 rounded to 40 ("a 40-fold difference").
  • License: citation only.
  • Known limits: The measure is the opacity of the eye's optical media (the lens and other media), called "lens yellowness" in the video. The study tested color names, not how colors look. The chip where the names switch on the board is a picture of "the same in both groups", not the study's actual boundary.

C18Sescha's hypothesis is rejected — the "lens alone" version

  • Time: 8:33.8 · 14.06 / 8:37.8 · 14.07 (stamp) / 8:39.0 · 14.08 / 8:41.2 · 14.09
  • What we say / show: "Your hypothesis, ma'am... was it wrong?" / "...Yeah. It was." / "The lens alone can't explain the map's results." On screen: Sescha stamps DENIED on her own card.
  • Class: no tab (P19).
  • Source: in their 2023 reply, the map team agree with the eye team that lens yellowing alone is unlikely to explain the pattern: Josserand, Meeussen, Dediu & Majid (2023), Scientific Reports 13:1837, doi:10.1038/s41598-023-28281-0 (PMC9892588) — full text [D6].
  • Our computation: none.
  • License: citation only.
  • Known limits: What is rejected is "the lens alone explains the map", not "the eyes play no part"; that stays open (C21, C28) [D8, N10].

C19The age dial: less blue light reaches the back of the eye with age

  • Time: 8:46.0 · 15.01 / 8:49.8 · 15.02 / 8:52.4 · 15.03 / 8:55.6 · 15.04 (3 s, picture only) / 8:58.9 · 15.05
  • What we say / show: "Look. This dial sets the age of the eye." / "Turning it from age 0 to 80." / "Blue light at the back of the eye fades." / "Everyone's eyes let in less blue light with age." On screen: an age dial from 0 to 80; the eye cross-section, its lens turning yellow; a gauge of three paper strips (sky blue, young-leaf green, vermilion) whose heights show the light reaching the back of the eye; a grey head profile with a closed door.
  • Class: MODELED (P21). The video does not say the method in a line; instead a source card "Pokorny et al. 1987" appears on screen (from 15.06) and the formula is given here and in Method M3. This is a recorded exception to our rule that every MODELED board states its method aloud.
  • Source: the lens-ageing formula of Pokorny, Smith & Lutze (1987), Applied Optics 26(8):1437–1440, doi:10.1364/AO.26.001437 — abstract and part of the tables; the same formula and coefficients are used in CIE 170-1:2006 (the standard itself was not read; its tables were checked against the CVRL lens-density data of Stockman, Sharpe & Fach 1999) [E11, E12]. The direction of the effect: E1 — abstract only.
  • Our computation: Method M3. Light reaching the back of the eye at 450, 550 and 620 nm (the sky-blue, young-leaf-green and vermilion strips), relative to age 0: 10^(−D1 × (L(A) − L(0))), with D1 = 0.2585, 0.0859, 0.0170 (CIE 170-1 table) and the lens age factor L(A) = 1 + 0.02 (A − 32) up to age 60, 1.56 + 0.0667 (A − 60) above. At 80: 0.221, 0.606, 0.906. Code build/web/src/boards/P21.js @ ea0e96eb6 (function reach).
  • License: —
  • Known limits: The formula describes an average observer aged 20 to 80; the dial starts at 0, so ages 0–20 extend the formula beyond its range (about a fifth of the blue drop shown on the dial falls there). Relative units only; no differences between people; the lens only. The dial shows light reaching the back of the eye, not what anyone sees.

C20Older people still see the sky as blue; the eye and brain seem to adjust

  • Time: 9:03.5 · 15.06 / 9:08.0 · 15.07
  • What we say / show: "Even so, older people still see the sky as blue." / "The eye and brain seem to adjust, and correct how colors look." On screen: the door in the head opens to show a sky sample; a disc between eye and head turns once and the sample at the back of the eye returns to sky blue, while the gauge strips stay low.
  • Class: the MODELED tab stays on (P21). The turning disc and the returning sample are a picture of adaptation; they are not computed.
  • Source: color appearance stays relatively stable over the lifespan: Delahunt et al. (2004) — abstract only [E1]. Color names did not change: C17 [E5]. The eye and brain compensate for lens yellowing (color adaptation and constancy), as argued by Hardy et al. (2023) — full text [E8].
  • Our computation: none for the adaptation picture.
  • License: citation only.
  • Known limits: Among our sources, how colors look is reported only at abstract level (E1); the 2005 study tested names. "Seem to" marks this.

C21The lens alone isn't enough; whether the eyes play any part is open

  • Time: 9:14.8 · 15.08b / 9:18.2 · 15.08c / 9:24.6 · 15.09
  • What we say / show: "...So maybe the eyes don't matter." / "We don't know that yet. All we know so far is that the lens alone isn't enough." / "If it's not the lens alone... what else splits the words for blue and green?"
  • Class: the MODELED tab is still on (P21).
  • Source: D6, D8 — full text; this summary is ours.
  • Our computation: none.
  • License: —
  • Known limits: —

9:33What else splits the words for blue and green?

C22In 2023, the eye team pushed back on the map study

  • Time: 9:33.3 · 16.01 / 9:37.5 · 16.02 / 9:40.5 · 16.02b / 9:45.8 · 16.03 / 9:52.5 · 16.03b / 9:54.3 · 16.03c. Card: "Hardy et al. 2023".
  • What we say / show: "The other idea... way of life and history?" / "Yes. Now we look at that one." / "The map study named the eye's lens as a path from sun to words." / "In 2023, the team behind that eye experiment pushed back on the map study." / "What did they say?" / "They agreed these languages are more common where the sun is strong. But it's not the sun, they say." On screen: a booklet for the 2023 rebuttal.
  • Class: no tab (P16, then P22).
  • Source: Hardy, Werner, Regier, Kay & Frederick (2023), "Sunlight exposure cannot explain 'grue' languages", Scientific Reports 13:1836, doi:10.1038/s41598-023-28280-1 (PMC9892488) — full text. Published on 1 February 2023, the same day and in the same journal as the map team's reply (C29). Four of its five authors ran the 2005 experiment [D1]. It accepts that the correlation exists [D7] and argues against sunlight as the cause [D2]. For 16.02b, see C15 [M14].
  • Our computation: none.
  • License: citation only.
  • Known limits: "A path" (16.02b) is the lens route that the 2021 study supported; the study's other predictors (number of speakers, lakes) are not about sunlight [M14, M5].

C23The eye team's argument: color words split where people need to tell colors apart

  • Time: 10:02.7 · 16.04 / 10:07.0 · 16.07 / 10:11.3 · 16.08 / 10:14.9 · 16.08b / 10:18.0 · 16.09
  • What we say / show: "If mixing colors up is a problem, color names split." / "What's a problem depends on life and history." / "But how can they say it's not the sun?" / "One is that 2005 experiment." / "The other is Japanese." On screen: one tag splitting into a sky tag and a young-leaf tag; a house and a scroll.
  • Class: no tab (P22) until 16.09; ILLUSTRATION (P23) from 16.09.
  • Source: Hardy et al. (2023) — full text [D2]: (1) color adaptation makes up for lens yellowing, as far as color naming (the 2005 result); (2) differences between color vocabularies can be explained by how similar colors are and how much people need to tell them apart; (3) Japanese and Nafaanra separated blue and green with no evidence that UV fell. The map team's reply agrees that the need to communicate shapes color words — full text [D5]. On that need: Gibson et al. (2017) — abstract only [W7].
  • Our computation: none.
  • License: citation only.
  • Known limits: The rebuttal also links larger color vocabularies with technologically complex societies, a framing that goes back to Berlin & Kay (1969). We do not use it, because it reads as a ranking of societies [D4, N6]; see the counter-example in C11 [W8]. The video uses only the Japanese example; the other language (Nafaanra) is not discussed [W9: abstract only and secondary].

10:20Did old Japanese have two words for blue and green?

C24Over 1,000 years ago, "ao" covered both blue and green

  • Time: 10:20.2 · 17.01 / 10:23.1 · 17.01b / 10:28.2 · 17.01c / 10:30.2 · 17.01d (2.5 s pause) / 10:33.1 · 17.02 / 10:37.4 · 17.02b / 10:40.8 · 17.05
  • What we say / show: "Records over a thousand years old." / "Back then, did Japanese have separate words for blue and green?" / "What do you all think?" / "One word, 'ao,' meant both blue and green." / "Oh, old Japanese didn't split them either!?" / "People back then called them that... so the records say." On screen: an old sheet with brush strokes (no readable writing); a door opens to show one tag, あを awo, colored half sky and half young leaf; a sky sample and a leaf sample lift in turn.
  • Class: ILLUSTRATION (P23).
  • Source: 青, Seisenban Nihon Kokugo Daijiten (historical note), via Kotobank, same URL as C01 — full text (dictionary): in Old Japanese, ao was a basic color word alongside aka, kuro and shiro and covered a wide range — blue, green and purple, and also black, white and grey; it narrowed toward blue as midori became the name for green, and it is still used for some non-blue things today [J2]. Hardy et al. (2023) — full text — citing Frellesvig (2010) and Stanlaw (2010), Language Sciences 32(2), doi:10.1016/j.langsci.2009.10.005: Old Japanese ao(i) spanned blue and green and probably appeared around 500–650 CE; midori entered after 750 CE — secondary (we have not read Frellesvig or Stanlaw) [J6].
  • Our computation: none.
  • License: citation only.
  • Known limits: We say "over 1,000 years ago" and give no years, because the dates come from sources we have not read. "The records" means the written record as summarised in the dictionaries. The video says ao covered blue and green; the dictionary says it covered more.

C25"Midori" first meant fresh sprouts; later it took over green, and "ao" moved toward blue

  • Time: 10:45.2 · 17.03 / 10:50.4 · 17.07x (3 s, picture only) / 10:53.6 · 17.07
  • What we say / show: "At first, 'midori' meant the freshness of new sprouts." / "Over time, 'midori' came to mean green, and 'ao' shifted toward blue." On screen: a time band from the past (left) to today (right) that splits in a Y; a young sprout with a tag みどり midori comes out on the young-leaf branch, and the sky branch carries the tag あを awo.
  • Class: ILLUSTRATION (P23).
  • Source: the original sense of midori, youthfulness and new shoots: the Nihon Kokugo Daijiten historical note on 青, and 緑 (midori), sense 2, Digital Daijisen via Kotobank (URL for 緑 not recorded) — full text (dictionary) [J4]. The shift: J2 (dictionary) and Hardy et al. (2023) — full text [J7].
  • Our computation: none.
  • License: citation only.
  • Known limits: "Shifted toward blue" simplifies; ao is still used for some green things (C01) [J2]. The time band has no dates.

C26No evidence that Japan's sunlight changed much, yet the words split

  • Time: 11:00.8 · 17.09 / 11:07.4 · 17.10 / 11:12.6 · 17.12 / 11:17.6 · 17.12b. Card: "Hardy et al. 2023".
  • What we say / show: "Hold on. If it's the sun... did the sunlight in Japan change over a thousand years?" / "The eye team says no evidence of a big change has been found." / "The sun stayed about the same... but the words split?" / "Yes. We can't say the sunlight changed much, and yet the words split." On screen: a sun rises at the past end of the time band; at the today end a door opens to show a sun of the same size; the Y-shaped split lifts.
  • Class: ILLUSTRATION (P23).
  • Source: Hardy et al. (2023) — full text: the authors write that they know of no evidence that UV levels fell substantially in Japan or Ghana over that period [D3]. We have not checked this ourselves.
  • Our computation: none.
  • License: citation only.
  • Known limits: This is the eye team's statement, not our measurement. The equal suns on the band are a picture, not data.

C27The "ao" in "aoba" is left over from the old, wide "ao"

  • Time: 11:23.3 · 17.12c / 11:29.8 · 17.13 / 11:34.5 · 17.16 (2 s, picture only)
  • What we say / show: "Then... why do people in Japan still call green leaves 'aoba,' 'blue leaves'?" / "Aoba keeps the old 'ao,' which meant green too." On screen: a tag 青葉 aoba drops at the "today" end of the time band, and the old あを tag lifts.
  • Class: ILLUSTRATION (P23, P02).
  • Source: J1, J2 — full text (dictionary).
  • Our computation: none.
  • License: citation only.
  • Known limits: We name only aoba as a leftover. We do not say that the traffic light's name comes from the old ao; we have not verified that history [J9–J12].

11:37So is it just language?

C28Not settled

  • Time: 11:37.1 · 18.01 / 11:41.4 · 18.02 / 11:44.0 · 18.02b
  • What we say / show: "So it's language after all! Eyes don't matter!" / "We can't go that far." / "Why these languages are more common where the sun is strong is still not settled."
  • Class: ILLUSTRATION (P23) until 18.02b; then no tab (small version of the two cards, P16).
  • Source: both 2023 papers — full text; this summary is ours [D7, D8]. Later papers: on 2026-10-01 we looked on OpenAlex at papers published from 2 February 2023 to 30 September 2026 that cite the rebuttal (4 papers), the reply (1), Dediu (2023) (3) or the 2021 study (52), and ran five keyword searches; we found no paper that settles the question — limited check [D9]. We repeated the check on 2026-10-08: the citation counts on OpenAlex were unchanged, no citing paper had been added since 2026-10-01, the 2026 study in C31 had no citations yet, five keyword searches found nothing new, and Crossref listed no correction or retraction notice for the 2021 study, the two 2023 papers, Dediu (2023) or the 2026 study — limited check [D9b].
  • Our computation: none.
  • License: —
  • Known limits: A limited search; we may have missed work. Last checked 2026-10-08.

C29The map team's reply — same day, same journal

  • Time: 11:50.6 · 18.03 / 11:54.7 · 18.05a / 11:56.7 · 18.05x / 11:59.7 · 18.05z / 12:03.7 · 18.05y / 12:06.2 · 18.05. Cards: "Josserand et al. 2023", "Hardy et al. 2023".
  • What we say / show: "And the map team wrote a reply to the pushback." / "What did they write back?" / "The lens alone isn't enough. Agreed." / "Words change with life and history. Also agreed." / "Then where do they differ?" / "They differ on the eyes." On screen: two booklets joined by one date slip, 2023-02-01; inside both, the yellow lens lies flat and a house with a scroll stands up; an eye and a closed door rise between them.
  • Class: no tab (P24).
  • Source: Josserand, Meeussen, Dediu & Majid (2023), "Reply to: Sunlight exposure cannot explain 'grue' languages", Scientific Reports 13:1837, doi:10.1038/s41598-023-28281-0 (PMC9892588) — full text [D1, D5, D6]. They agree that lens yellowing alone is unlikely to explain the pattern, that the need to communicate shapes color words, and that other factors such as dyeing technology and history matter. They add that what creates that need in culture and environment has not yet been worked out, and that UV and distance to lakes had effects separate from culture.
  • Our computation: none.
  • License: citation only.
  • Known limits: "They disagree on the eyes" is our short form for the open point: whether UV, through the eyes, has an effect separate from culture [D8].

C30The reply's point: the 2005 participants were in the US, not where the purple is darkest

  • Time: 12:08.4 · 18.05w / 12:14.7 · 18.05b / 12:21.1 · 18.05c / 12:27.0 · 18.06 / 12:30.7 · 18.07
  • What we say / show: "Their reply says the 2005 experiment studied people in the United States." / "The US? On the map, that's not where the purple is darkest." / "Yes. That experiment didn't study people where the purple is darkest." / "In places like that, people get more UV." / "There, the reply says, the eye and brain might not adjust fast enough." On screen: the purple map; the contiguous US lifts and is tied to a "2005" tag; the darkest shade lifts; a pin goes into central Africa, tied to a blank tag (not tested in the 2005 study); an eye disc turns but stops short of a full turn.
  • Class: MEASURED while the map is shown (P24).
  • Source: Josserand et al. (2023) — full text [E10]: the 2005 participants were in the US; at the UV doses of high-UV regions, compensation may not keep up. They cite a filter study (Walter 2011) that simulated 20–33 years of high-UV lens yellowing and found effects on the naming of blue samples — secondary (we have not read it; it is not in the video).
  • Our computation: on our map (C04), the highest cell in the contiguous United States is 82.2 mW/m²/nm (shade 4 of 5); no cell of the contiguous US is in the darkest shade. The pin at 18.05c (about 1°N, 21°E, Congo Basin) is in a cell of 98.4 mW/m²/nm (darkest shade). Data D2; code build/tools/data.mjs @ 2c90e38ab, build/web/src/boards/w4-kami.js @ bb00a458b, build/web/src/boards/P24.js @ 83f7e96df.
  • License: NASA data, CC0 (Data D2).
  • Known limits: "Get much more UV" refers to the mapped noon UV-B of 1998, not to the dose a person receives, which also depends on time spent outdoors. The argument is the reply's, and the video says so.

C31What would settle it: data that are still rare

  • Time: 12:36.3 · 18.10 / 12:39.3 · 18.10b / 12:46.2 · 18.11
  • What we say / show: "Then how could this be settled?" / "Where the sun is strong, we need records of younger and older eyes, and color names." / "But records like that are still rare." On screen: a form with four headings (sun, the ages 23 and 74, eye, color tag) and empty cells.
  • Class: no tab (P24).
  • Source: both 2023 papers ask for such data — the rebuttal: little is known about how lens yellowing proceeds in high-UV regions; the reply: data from people of different ages at different latitudes are needed — full text; this summary is ours [D7]. A 2026 study by members of the map team compares color perception in two populations with different UV exposure: Josserand, Dediu, Pellegrino & Caparos (2026), Topics in Cognitive Science, doi:10.1111/tops.70068 — abstract only. We do not describe its results: we have read only the abstract, and it is about the eyes of specific populations [D10, N4].
  • Our computation: none.
  • License: citation only.
  • Known limits: "Rare", not "none". It refers only to this kind of record: high-UV places, different ages, and eyes and color names together.

12:49What can we say for sure?

C32Three findings and the answer

  • Time: 12:49.5 · 19.01 / 12:50.8 · 19.02 / 12:55.8 · 19.02b / 12:57.5 · 19.03 / 13:04.4 · 19.03b / 13:07.7 · 19.05 / 13:15.0 · 19.06
  • What we say / show: "So, to sum up." / "First, the stronger the sun, the more joined marks." / "Counted and checked." / "Second, the lens alone can't explain it. Words also split because of life and history." / "Both teams agreed on this point." / "Third, whether the eyes matter at all is still not settled." / "Languages with one word for blue and green are more common where the sun's strong. But not the sun alone." On screen: a folding record booklet, opened one fold per finding; then the title card's door opens to the answer.
  • Class: no tab (P25, P09); the booklet's first fold carries a small ILLUSTRATION tab from when it opens: its small layered map is a schematic, not drawn from data.
  • Source: One: our count (C07) and the study's corrected model (C08) [S2, M5, M6, M13]. Two: both 2023 papers (C18, C29) [D5, D6, D7]. Three: [D8, D9]. The answer: [M6, D6, D8].
  • Our computation: C07.
  • License: —
  • Known limits: "Counted and checked" means that our rough count and the study's corrected model point the same way. Neither shows a cause.

C33Counting alone does not tell you why

  • Time: 13:25.6 · 20.04 / 13:33.4 · 20.05 / 13:39.3 · 20.08 / 13:45.8 · 20.09
  • What we say / show: "These languages are more common where the sun is strong. Counting alone can't tell us why." / "So we checked one thing at a time. The eye's lens. And old Japanese." / "Look, everyone. This light. It looks green, but Japanese calls it...?" / "...Ao." On screen: the map with the pins laid over it, and an arrow from the sun to the words blocked by a closed door; then the traffic light from the opening.
  • Class: ILLUSTRATION (P26) with a small MEASURED tab on its map; then ILLUSTRATION (P01).
  • Source: [M6, D8]; the checks in C16–C17 and C24–C26; the traffic light: C01 [J1].
  • Our computation: the small map uses the same grid and pins as C04–C07.
  • License: —
  • Known limits: —

13:49The color-names drawer

No factual claims.

What this video does not claim

  • That people in sunny places see blue and green differently, or cannot tell them apart. The data are about words, not eyesight [N1, N8].
  • That any language is "missing" a color word or is less developed [N6].
  • That UV decides language, or that language decides what people see [N9].
  • That one side of the 2023 exchange won [N10].
  • Anything about eye health [N7].

Data

D1Language data: Josserand, Meeussen, Majid & Dediu (2021)

  • What we use: for each of the 142 languages, its Glottolog code, name, latitude and longitude, language family, macroarea, and exists_blue (whether it has a basic term for "blue" alone).
  • Where and version: GitHub repository ddediu/colors-UV, branch master, file input_files/data_colors.csv, read on 2026-09-30. The repository has not changed since 21 July 2021, so what we read is commit d71aaedbc3bd0c657076e27e845f970878ff29cf (the file itself last changed in commit 9d18436f05305420b100add15432f01b8d1ccc57, 16 November 2020; Git blob 47d531b41b2c3ee7c03eaa23a877e2f5fa7700cf, 15,359 bytes; checked 2026-10-08). Archived on Zenodo as a single ZIP file, version 1.0, published 8 July 2021 (the only version): doi:10.5281/zenodo.5083676. We have not compared the file inside the ZIP with the GitHub file [M10].
  • License: CC BY-SA 4.0, as stated for this file in the repository's input_files/ReadMe.md (checked 2026-10-08) [M10]. The Zenodo record lists CC BY 4.0 for the archive as a whole; we follow the file's own, stricter license.
  • Obligations:
    • Attribution: authors, paper, license and link — on this page and in the video description.
    • Share-alike: the maps in this video are built from this dataset (the 142 language locations, with the UV-B values we computed, Method M1). The map images themselves may be reused under the same license, CC BY-SA 4.0, with credit to Worldlines and to Josserand, Meeussen, Majid & Dediu (2021). This applies to the map images only, not to the rest of the video. We do not publish the derived table.
  • Used in: C02, C05–C08, C30, C32, C33.

D2UV-B: NASA Earth Probe TOMS, 1998

  • What: surface UV estimates — local-noon spectral irradiance at 305 nm and 310 nm, in mW/m²/nm, daily, on a 1° × 1° grid (180 latitude bands × 360 cells running east from 179.5°W; "999" = missing). 1 January–13 December 1998: 347 days at each wavelength, 694 files, plus the documentation 1README.UV.txt.
  • Where: https://acd-ext.gsfc.nasa.gov/anonftp/toms/eptoms/data/uv_ery_4_wavelengths/ (folders uv305/Y1998/ and uv310/Y1998/), NASA Goddard Space Flight Center, Atmospheric Chemistry and Dynamics Laboratory (Code 614). No login is needed.
  • Version: files dated 2004-04-30 on the server; downloaded 2026-10-04; the URL, size and SHA-256 of each file are recorded in our manifest. Method of the estimates: Herman et al. (1999), doi:10.1029/1999JD900062 — bibliographic only.
  • License and conditions: data from NASA-led missions are CC0 unless marked, and NASA strongly urges users to cite the data (NASA Earthdata, "Data Use and Citation Guidance", https://www.earthdata.nasa.gov/engage/open-data-services-software-policies/data-use-guidance). NASA material may not be used to suggest NASA endorsement, and the NASA insignia and logotype are protected (NASA, "Guidelines for using NASA Images and Media", https://www.nasa.gov/nasa-brand-center/images-and-media/). Both checked 2026-10-04. This video is not endorsed by NASA. We show no NASA logo or insignia; the card on screen names only the data ("NASA TOMS UV-B 1998").
  • Used in: C03, C04, C07, C30, C33.

D3Land outline: NOAA NCEI ETOPO1

  • What: ETOPO1 1 Arc-Minute Global Relief Model, Ice Surface version (the top of the Antarctic and Greenland ice sheets), taken every 5 arc-minutes from NOAA CoastWatch ERDDAP (dataset etopo180). We mark cells above 0 m as land and draw them at 10 arc-minutes.
  • Citation: NOAA National Geophysical Data Center. 2009: ETOPO1 1 Arc-Minute Global Relief Model. NOAA National Centers for Environmental Information. Accessed 2026-06-12 (the date of our copy). Amante, C. and B.W. Eakins, 2009. ETOPO1 1 Arc-Minute Global Relief Model: Procedures, Data Sources and Analysis. NOAA Technical Memorandum NESDIS NGDC-24. National Geophysical Data Center, NOAA. doi:10.7289/V5C8276M. Both as given on NCEI's record for the dataset, https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.ngdc.mgg.dem:316 (checked 2026-10-08) [M16].
  • Version: a copy kept in our workspace from an earlier project (file dated 2026-06-12). NCEI now lists ETOPO1 as deprecated, with ETOPO 2022 as the current model (doi:10.25921/fd45-gt74); this episode uses ETOPO1 only for the outline of land at 10 arc-minutes.
  • License: "Produced by the NOAA National Centers for Environmental Information. Not subject to copyright protection within the United States." NCEI asks users to cite the dataset as above. Not for navigation.
  • Used in: the land shapes on every map (C04 onward).

D4Sources at a glance

SourceIdentifierWhat we readUsed in
Josserand, Meeussen, Majid & Dediu 2021, Sci Rep 11:19095doi:10.1038/s41598-021-98550-3 · PMC8476573full textC03, C05, C06, C08–C10, C13, C15, C22
Hardy, Werner, Regier, Kay & Frederick 2023, Sci Rep 13:1836doi:10.1038/s41598-023-28280-1 · PMC9892488full textC06, C10, C13, C15–C17, C20, C22–C26, C28, C31
Josserand, Meeussen, Dediu & Majid 2023, Sci Rep 13:1837doi:10.1038/s41598-023-28281-0 · PMC9892588full textC10, C11, C16, C18, C23, C29–C31
Hardy, Frederick, Kay & Werner 2005, Psychol Sci 16(4):321–327doi:10.1111/j.0956-7976.2005.01534.x · PMC2586906abstract only (numbers via the 2023 papers)C15–C17
Lindsey & Brown 2002, Psychol Sci 13(6):506–512doi:10.1111/1467-9280.00489 · PubMed 12430833abstract onlyC10, C14
Delahunt, Webster, Ma & Werner 2004, Vis Neurosci 21(3):301–307doi:10.1017/s0952523804213025abstract onlyC12, C14, C19, C20
Pokorny, Smith & Lutze 1987, Appl Opt 26:1437doi:10.1364/AO.26.001437abstract and part of the tablesC12, C19, Method M3
Review, Frontiers in Ophthalmology 2024doi:10.3389/fopht.2024.1414483abstract onlyC13
Gibson et al. 2017, PNAS 114:10785doi:10.1073/pnas.1619666114abstract onlyC11, C23
Wnuk, Verkerk, Levinson & Majid 2022, Cognition 229:105223doi:10.1016/j.cognition.2022.105223abstract onlyC11, C23
Kuriki et al. 2017, J Vis 17(3):1PubMed 28249298abstract onlyC01
Berlin & Kay 1969, Basic Color TermsUniversity of California PresssecondaryC02, C23
Frellesvig 2010; Stanlaw 2010, Language Sciences 32(2)Stanlaw: doi:10.1016/j.langsci.2009.10.005secondary (cited by Hardy et al. 2023)C24
Walter 2011—secondary (cited by Josserand et al. 2023)C30
Josserand, Dediu, Pellegrino & Caparos 2026, Topics in Cognitive Sciencedoi:10.1111/tops.70068abstract only; results not describedC31
Herman et al. 1999, J Geophys Res 104(D10):12059–12076doi:10.1029/1999JD900062bibliographic onlyC04, Data D2
青 (Digital Daijisen; Seisenban Nihon Kokugo Daijiten), Kotobankhttps://kotobank.jp/word/%E9%9D%92-192607full text (dictionary)C01, C24, C25, C27
緑 (Digital Daijisen), KotobankURL not recordedfull text (dictionary)C25
NASA Earth Probe TOMS surface UV, 1998Data D2data and documentationC03, C04, C07, C30, C33
OpenAlex, Crossref and Semantic Scholar searches—limited check, 2026-10-01 and 2026-10-08C28

Method

M1Counting the languages in three groups

  1. Inputs: Data D1 (142 languages, exists_blue, coordinates) and Data D2.
  2. Annual mean: for each 1° cell, the mean of all valid daily values at 305 nm and 310 nm together for 1998; missing days are skipped (the same averaging as the 2021 study).
  3. Each language's UV-B: the value of the 1° cell that contains the language's coordinates (latitude: the nearest band centre, and a point exactly between two bands goes to the southern one; longitude: converted to −180° to 180°, then the cell that contains it).
  4. Order the 142 languages by this value, lowest first (ties keep the dataset's order; none falls on a group boundary). Ranks 1–47: weak-sun group; 48–94: middle; 95–142: strong.
  5. In each group, count the languages with no basic term for "blue" alone (the "joined mark"): weak 6/47 (13%), middle 25/47 (53%), strong 29/48 (60%); 60/142 overall.
  6. Group ranges, in mW/m²/nm: weak 9.19–62.30; middle 64.56–93.32; strong 93.55–137.62.
  7. In the video: weak "just over one in ten", strong "over half". No percentages or fractions are spoken; the boxes show the marks.
  8. Approximations: one point per language; one year of UV; thirds as cut points; no correction for related languages (for the study's corrected model, see C08).
  • Checks in the code: the build stops if the group counts differ from 41/47, 22/47 and 19/48 (languages with a word for "blue" alone), or if any pin's shade differs from the shade of its map cell or falls outside its group's shades.
  • Code: build/tools/nasa-uvb/make_grid.py and build/tools/data.mjs @ 2c90e38ab. Run in build/: python3 tools/nasa-uvb/make_grid.py, then node tools/data.mjs (the NASA files are fetched by tools/nasa-uvb/fetch.py).

M2Drawing the UV map

  • Range 80°N to 58°S, equirectangular, one shade per 1° cell; no interpolation (cell edges are smoothed by one pixel only).
  • Five shades, lightest to darkest, with breaks at 33.42, 63.43, 78.43 and 93.43 mW/m²/nm. T1 = 63.43 and T2 = 93.43 lie midway between the last language of one group and the first of the next; W = T2 − T1 = 30.0; the breaks are T1 − W, T1, T1 + W/2 and T2. So the weak group sits on shades 1–2, the middle group on shades 3–4 and the strong group on shade 5, and every pin stands on a cell in its own group's shades.
  • Values in the map range: 2.76 to 157.36 mW/m²/nm (the maximum is in the Peruvian Andes). Shade 5 covers 24.3% of land cells in the map range.
  • The legend shows the five shades as equal steps from a small sun to a big sun, with no numbers.
  • Code: build/tools/data.mjs @ 2c90e38ab; build/web/src/boards/w2-map.js and build/web/src/boards/w4-kami.js @ bb00a458b; build/web/src/boards/P10.js @ 2c90e38ab.

M3The age dial (MODELED)

  • Input: the age on the dial, 0 to 80, turned at constant speed from 0.1 s after the start of line 15.02 to 1.6 s after the start of 15.04; A = 80 × v.
  • Output: light reaching the back of the eye, relative to age 0, shown as the heights of three paper strips at 450 nm (sky blue), 550 nm (young-leaf green) and 620 nm (vermilion): T(λ, A) = 10^(−D1(λ) × (L(A) − L(0))), with D1 = 0.2585, 0.0859, 0.0170 and L(A) = 1 + 0.02 (A − 32) for A ≤ 60, 1.56 + 0.0667 (A − 60) above. At 80: 0.221, 0.606, 0.906. The strips fall slowly up to 60 and faster after it, as in the formula.
  • Source of the formula: Pokorny, Smith & Lutze (1987), as adopted in CIE 170-1:2006; the age-independent part of the lens density cancels out in the ratio. Our table values were checked against the CVRL lens-density table (61 points, largest difference 0.00000) [E11, E12].
  • What the video says about the method: 15.01 "Look. This dial sets the age of the eye."; 15.03 "Blue light at the back of the eye fades." The method is not stated aloud; the card "Pokorny et al. 1987" (from 15.06) and this page stand in for it — a recorded exception to our fidelity charter.
  • Approximations and limits: the formula is fitted to ages 20–80; 0–20 extends it beyond its range. Relative units only; no differences between people; the lens only, no other parts of the eye. The dial does not show appearance; at 15.06–15.07 a separate, uncomputed picture shows the eye and brain adjusting.
  • Code: build/web/src/boards/P21.js @ ea0e96eb6; notes build/notes/v9-fidelity.md §2.

M4Tabs by board

Scenes by chapter: S01–S06 = 0:00 · S07–S09 = 2:01 · S10 = 4:28 · S11–S12 = 5:31 · S13–S15 = 7:29 · S16 = 9:33 · S17 = 10:20 · S18 = 11:37 · S19–S20 = 12:49 · S21–S22 = 13:49.

BoardScenesTabWhat it showsClaims
P01S01–S03, S20ILLUSTRATIONdaytime crossing with a greenish light; tag 青信号 ao-shingōC01, C02, C33
P02S01–S02, S17–S18ILLUSTRATIONyoung leaves and a green apple; tags 青葉 aoba, 青りんご ao-ringoC01, C27
P03S02, S04ILLUSTRATIONpaper globe and figures; one tag over blue and greenC02
P04S03ILLUSTRATIONsky and leaf cut-outs—
P05S04–S05ILLUSTRATIONstrong-sun and weak-sun panels (Comet's idea)ideas
P07S05ILLUSTRATIONsatellite on an armC03
P08S05ILLUSTRATIONrecords sliding togetherC03
P09S05–S07, S19–S20no tabtitle card; its door opens to the answerC32
P10S07MEASUREDUV-B map; card "NASA TOMS UV-B 1998"C04
P11S07ILLUSTRATION (own tab)sunburn windowsC04
P12S07–S08MEASUREDexample pin, 142, rings; card "Josserand et al. 2021"C05, C06
P13S07–S08MEASUREDlegend: split mark, joined markC06
P14S08MEASUREDthree boxes with marksC07
P15S09–S10MODELEDthreads tying related languages; rising arrow; card "Josserand et al. 2021"C08, C09
P16S10–S11, S16, S18no tab (ILLUSTRATION from 11.01 to 11.04)eye card; house-and-scroll card; closed doorC09–C12, C22, C28
P17S11, S13ILLUSTRATIONeye cross-section; yellow sunglassesC12, C16
P18S12ILLUSTRATIONUV yellows the lens sooner; gauge behind the eye; card "Josserand et al. 2021"C13–C15
P19S12–S16no tabSescha's hypothesis card; DENIED stampC15, C18
P20S13–S14MEASUREDthe 2005 experiment: two rows, 23 and 74, chips, ×40; card "Hardy et al. 2005". The eight chips and where the rows switch from blue to green are schematicC16, C17
P21S15MODELED (method by card and this page)age dial; light gauge; head with a door; disc; card "Pokorny et al. 1987"C19–C21
P22S16no tabrebuttal booklet; card "Hardy et al. 2023"C22, C23
P23S16–S18ILLUSTRATIONold paper; あを tag; time band; card "Hardy et al. 2023"C23–C28
P24S18no tab; MEASURED while the map showstwo booklets; the US with "2005"; empty form; cards "Josserand et al. 2023", "Hardy et al. 2023"C29–C31
P25S19no tab; small ILLUSTRATION tab on fold 1 (a schematic map)record booklet with three foldsC32
P26S20ILLUSTRATION; small MEASURED tab on its mapmap with pins; arrow with a closed doorC33
P27S22no tabnext drawer; end screen—
T01S02no tabtitle—

M5Code and versions

Paths are relative to project/worldlines/episodes/ep01-blue-green/ in our working repository. Commits as of 2026-10-07 (the v9 render). The code and the derived table are not published; this page describes the method in full and lists the files and versions we used.

FileRoleCommit
build/tools/nasa-uvb/fetch.pydownloads the NASA files, checks that they contain only data, records URL, size and SHA-2562c90e38ab
build/tools/nasa-uvb/make_grid.pybuilds the 1998 annual-mean grid2c90e38ab
build/tools/data.mjseach language's cell value and group; shade breaks; checks2c90e38ab
build/web/src/boards/w2-map.js, w4-kami.jsdraw the maps from the gridbb00a458b
build/web/src/boards/P10.jsmap board and its source card2c90e38ab
build/web/src/boards/P14.jsthree boxes83f7e96df
build/web/src/boards/P24.jsreply booklets and small map83f7e96df
build/web/src/boards/P21.jsage dialea0e96eb6
build/tools/build.mjsrenders the videobb00a458b
scratch/final15-lines.tsvJapanese master script (kept outside version control; file hash af317e961f31)—
plan/l10n/en/script.tsvfinal English lines (the English audio track and subtitles; file hash 9dbd28da688c)99cf5ca07

Production notes

"The channel's creator" is one person. "AI agents" are Claude agents working in Claude Code. Tool names are given only where our records name them.

RoleDone byWhat happenedTools named in our records
Question, direction, approvalThe channel's creator (one person)Chose this topic; set the rule that real data should be used where it exists; approved the key decisions — switching the map to NASA's grid and grouping the pins by the grid cell at each location (2026-10-04), music played on GarageBand instruments and the series jingle (2026-10-04), the two voices (2026-10-05), the age-dial formula and its labelling exception (2026-10-06), the titles and the thumbnail wording (2026-10-07); decided to make all seven other languages with our own voices, set the overall loudness, and left the final check of the translations to the AI team (2026-10-07); asked for the Chinese voices to be remade to sound like native Mandarin speakers (2026-10-08); watched the full cuts v4, v5, v7 and v8 and gave notes; chose the character poses and fixes.—
Research and fact ledgerAI agentsFound and read the sources and recorded in the fact ledger how much of each was read (full text, abstract, secondary).Claude (Claude Code)
Script (Japanese master)AI agents, with the creator's notesWritten and revised by AI agents, tested with LLM readers given no context, then revised again. The creator gave notes; the wording of 10.04 and 12.08 and the framing of 10.05 follow the creator's suggestions.Claude
Fact checksAI agentsRepeated fact-check passes compared the script with the fact ledger.Claude
Translations (English, French, Italian, German, Spanish, Chinese in simplified and traditional characters, Korean)AI agents, with the creator's go-aheadEach language was translated scene by scene from the Japanese master by an AI agent, following a written style guide, notes on how each character speaks, a glossary of 86 terms, and the context and time limit of every line. In the first review round, two more AI agents checked each translation: one read it straight through as a native speaker would, without seeing the Japanese; the other compared it with the Japanese line by line. Two further rounds checked the revised lines and the lines around them. Lines were revised between rounds. The translations have not yet been checked by human native speakers. If you speak one of these languages and a line sounds wrong or unnatural, please tell us in the comments under the video; corrections that change the meaning are listed under Corrections.Claude (Claude Code)
Character artAI image generation, edited by script; selected by the creatorPoses were generated as sheets, cut out, and fixed with deterministic pixel-editing scripts.Astra (gpt-6-astra), run through codex exec
Boards, room, animationCode written by AI agentsThe paper-style boards and all motion are drawn by code (JavaScript, Canvas 2D; the room with three.js), recorded in headless Chrome and encoded with ffmpeg. No image generation was used for the boards.Claude; headless Chrome; ffmpeg
Data processingCode written by AI agentsMethod M1–M3.Python (standard library); Node.js
MusicScore by an AI agent; Apple instruments; mixed by scriptThe score was written as MIDI by an AI agent and played on GarageBand's built-in instruments (Classic Electric Piano, Marimba, Steinway Grand Piano, Upright Studio Bass, String Ensemble, Classic Analog Pad, Clarinet Solo, the SoCal drum kit, Vibraphone, Harp), with GarageBand operated by an AI agent; the stems were mixed by script. The creator chose the series jingle. The GarageBand license (section 2G) lets users use its built-in instruments, samples and loops royalty-free to create their own soundtracks for video and audio projects, and to broadcast and distribute those soundtracks, but not to redistribute the sounds on their own; the score is our own composition (license text checked 2026-10-08, GarageBand 10.4.13).Claude (Opus 5.5); GarageBand; mido
Voices, Japanese trackSyntheticTwo voices designed from text prompts (ElevenLabs voice design, eleven_ttv_v3); the lines were generated with Eleven v4 and checked for misreadings with ElevenLabs speech-to-text; the creator chose the voices by listening.ElevenLabs
Voices, English and the six other dubbed tracksSyntheticEnglish is the video's original audio track; French, Italian, German, Spanish, Chinese and Korean are extra audio tracks on the same video. The lines were generated with Eleven v4 in the same two character voices as the Japanese track. For Chinese, the creator asked for voices that sound like native Mandarin speakers: we made a Mandarin-speaking version of each character voice with ElevenLabs Voice Remix (changing only the language and accent). Two or three takes of every line were generated; the take was chosen with ElevenLabs speech-to-text (checking for misread words) and by fit to the timing, and each line was placed at the start of the matching Japanese line, with the same picture, music and sound effects in every language. Every track is mixed with the same gain, so the music and sound effects are identical in every language; overall loudness is about −14 LUFS.ElevenLabs
SubtitlesCodeNine subtitle tracks (Japanese, English, French, Italian, German, Spanish, Chinese simplified, Chinese traditional, Korean), generated by code from each language's spoken lines and timed to the placed voices. Subtitles are separate tracks and are not burned into the picture.—
Text on screenCodeThe few words on screen use typefaces that come with macOS: Futura (numbers, stamps, tabs, name tags), Gill Sans (romanized readings, source cards) and Tsukushi B Round Gothic (Japanese words). The macOS license (macOS Tahoe 26, section 2E) lets users use the included fonts to display and print content while running macOS; the text is drawn into the video frames on our Mac, and no font files are embedded or shared (license text checked 2026-10-08).—
Sound effectsStock sound library効果音ラボ (soundeffect-lab.info) and 効果音辞典 (sounddictionary.info): commercial YouTube use allowed, credit optional, no redistribution (terms checked 2026-10-03); trimmed and mixed by script. The room tone and the clock are synthesized.—