Colour Science & Perception
what the eye settles, and what it cannot
The organising axis

Every figure, by what it names

A chromaticity diagram is a picture of an integral, and the integral has a kernel somebody chose. Every figure here says which — and this page is all of them, sorted by the answer, read back out of the finished drawings.

Every figure on this site carries a strip along its foot. The right-hand slot names what the picture was drawn under: an observer, or — where an observer is not the governing choice — the space, the model or the condition that is. That is this site's first figure rule, and it exists because a chromaticity diagram without one is a picture of an integral with the kernel left off.

This page is built by rendering all 3703 figure placements and reading those strips back out of the finished SVG, rather than from the code that wrote them. The two can differ, and where they do it is the picture that is right. 455 of them mark at least one region as colour the display cannot show rather than drawing an approximation of it.

The index exists because the rule was being kept in the way that costs nothing. When it was first built, at the start of the scale phase, 277 of 277 placements carried a strip — the rule, satisfied outright — and 204 of them named the same observer, with three naming the 1964 functions and all three of those in one essay. Two of sixty-four generators had ever been called with more than one distinct note. A label that never varies carries no information, and this one is the site's own thesis: the observer is a choice. Today 70 of 155 generators are called with more than one, and 29 are drawn under both standard observers, which is the difference between a family of figures and one picture shown repeatedly.

The set the number is a mean over

Added with the `depth-08` round, and the band whose existence is that round's finding. Every adaptation number in this collection is an average over a hundred and twenty-five constructed surfaces, and that set is a declaration: three numbers describing how bright and how saturated its members are, and a dimension fixed at three because three basis functions were written down. None of the four had ever been varied, and varying them moves the published levels by more than any declared width in the collection does. It sits first, ahead of the two observers, and the position is the argument. A figure in this band names an observer as well — the observer is held in every one of them — so naming it would name the thing that is fixed rather than the thing that was chosen. That is the same reasoning the coordinate band was added under, applied to a choice that was invisible for fourteen rounds because it lived in a function signature rather than in a table of constants. 197 figures.

CIE 1931 2° observer · the set, varied

adaptation as a diagonal · the test set, 125 surfaces

the set, bounded

The unit the number is printed in

The band whose existence is a finding. Almost every number this collection publishes about a difference between two colours is in CIEDE2000, and the unit was chosen once by importing a function. Recomputed under the five other formulae available, with each first scaled onto the published one's footing, six published quantities move by between 1.71 and 2.30. It sits ahead of the two observers for the reason the set band does: a figure here names an observer as well, because one observer is held in every one of them, so classifying by the observer would file it under the thing that is fixed rather than the thing that was chosen. 100 figures.

CIE 1931 2° observer · the unit, varied

What the observer is allowed to know

For the family that counts an adaptation model's parameters by where each one has to come from. A model needing nine numbers about the room is not a better model than one needing three; it is a model of something else, because the nine are the change of light the observer is trying to discount. The observer is held in every figure here and the information it is granted is not, so the strip names the second. 24 figures.

CIE 1931 2° observer · what the observer may know, varied

The aperture the sample was measured through

Added with the round that audited the model equation. A translucent sample does not have a reflectance; it has a kernel, and what an instrument reads is that kernel integrated over the hole it is looking through. Two instruments differing only in the size of that hole return different for the same slab — up to a hue on the other side of neutral — so the strip names the aperture rather than the observer, which is held at 1931 in every figure here. 72 figures.

CIE 1931 2° observer · the aperture, varied

The field the sample was standing in

Added with the same round, for the figures whose subject is that a reflectance is an average over directions. A reading is the sample's own reflectance when the illumination is uniform — a theorem, and the reason the integrating sphere exists — and no room is uniform. So the governing choice in these figures is the shape of the light arriving, and that is what the strip names. 16 figures.

CIE 1931 2° observer · the field, varied

The grid the integral is summed over

Added with the round that audited the observer's tabulation. Eighty-one numbers from 380 to 780 in steps of five is not a property of the eye, of the light or of the sample; it is a choice a committee made about paper, and it has three separable parts — a step, a range, and a rule for the values in between. A figure in this band holds one analytic observer fixed and varies the index it is summed over, so naming the observer would name the constant. It sits ahead of the two observer bands for the same reason the set and unit bands do. 116 figures.

the grid, varied · 300–830 against 380–780

the grid, varied · 5 lights

the grid, varied · 5 nm read at 1 nm

the grid, varied · 5 nm, five origins

the grid, varied · 6500 K

the grid, varied · a 5 nm tabulation

the grid, varied · a fluorescent tube

the grid, varied · a fluorescent tube, mercury lines on a phosphor bed

the grid, varied · a three-emitter LED lamp

the grid, varied · a three-laser projector at white

the grid, varied · fluorescent tube

the grid, varied · forty-two surfaces

the grid, varied · neutral, 6 lights

the grid, varied · notch, 6 lights

the grid, varied · paper, 6 lights

the grid, varied · power against product

the grid, varied · steep sides

the grid, varied · the normaliser at 5 nm

the grid, varied · two origins

the grid, varied · two quadrature rules

the grid, varied · vermilion, 6 lights

the slit, varied · this site's analytic observer

The observer itself, as a variable

Added with the same round, and it is the band this site spent nineteen rounds not needing. Every other band names what was chosen while the observer was held; here the observer IS the choice — a field size, an age, a macular density, a set of peak wavelengths, a rod contribution — and the strip says so rather than naming one of them. It sits ahead of the pigment band deliberately. A figure here is built from the same pigment template those are, and filing it under the template would name the machinery rather than the argument. 203 figures.

the observer, varied · against CIE 1931

the observer, varied · against the grid

the observer, varied · each factor emptied

the observer, varied · forty-two surfaces

the observer, varied · notch under daylight

the observer, varied · notch under incandescent

the observer, varied · on a 0.25 nm grid

the observer, varied · on this site’s grid

the observer, varied · seven arguments

the observer, varied · the age of the lens

the observer, varied · the cone optical density

the observer, varied · the field size

the observer, varied · the macular pigment

the observer, varied · the pigment peaks

the observer, varied · the rods

the observer, varied · the rods against the S cones

the observer, varied · the rods against the cones

the observer, varied · the rods, routed

the observer, varied · the rods, routed three ways

the observer, varied · three adaptation routes

the observer, varied · vermilion under daylight

the observer, varied · vermilion under incandescent

the observer, varied · vermilion under laser

the observer, varied · vermilion under narrowband

the observer, varied · vermilion under phosphor

How rough the surface is

Added with the round that gave the scene solver a lobe. Radiosity does not approximate a glossy wall; it has no slot for one, because its unknown is a single number per patch and a surface that returns light differently in different directions does not have one. The figures in this band vary the roughness of the walls and hold the observer at 1931 throughout, so the strip names the roughness. 67 figures.

the roughness, varied · 0.2 against matt

the roughness, varied · adapted to the lamp

the roughness, varied · adapted to the room

the roughness, varied · body reflectance 0.5

the roughness, varied · eight directions against sixteen

the roughness, varied · fixed at 0.2

the roughness, varied · fixed at 0.3

the roughness, varied · matt against 0.15

the roughness, varied · matt against 0.2

the roughness, varied · matt against 0.3

the roughness, varied · the Lambertian limit

the roughness, varied · the finish, varied

the roughness, varied · the lobe added beside it

the roughness, varied · the lobe taken out of the body

the roughness, varied · the room as a whole

the roughness, varied · two constructions

Which index the model is allowed to keep

The band for the round that audited the model equation itself. A surface's response is a function of six arguments — the wavelength, direction and place light arrives with, and the three it leaves with — and a colour computed as an integral keeps one of them. Each figure in this band restores one index and reports what the restoration is worth, so what varies across the band is not a sample or a light but how much of the object the model is allowed to see. 69 figures.

CIE 1931 2° observer · the index, varied

The CIE 1931 2° observer

The 1931 colour-matching functions, from seventeen observers matching a two-degree field. The default of the whole discipline, and the default here — which is the point this index was built to make visible. 1507 figures.

36 rooms, CIECAM16 · CIE 1931 2°

72 paints, five rooms · CIE 1931 2°

72 paints, one lamp · CIE 1931 2°

CAM16-UCS as the price · CIE 1931 2° observer held fixed

CAM16-UCS · one adopted white, chosen · CIE 1931 2° observer

CIE 1931 2° observer

CIE 1931 2° observer · D50

CIE 1931 2° observer · a delivery chain

CIE 1931 2° observer · a width, varied

CIE 1931 2° observer · away from the optimum

CIE 1931 2° observer · implied confusion points

CIE 1931 2° observer · measured confusion lines

CIE 1931 2° observer · sRGB primaries

CIE 1931 2° observer · several objectives

CIE 1931 2° observer · the bound, stated

CIE 1931 2° observer · the census, in its own units

CIE 1931 2° observer · the device chooses the basis

CIE 1931 2° observer · the ruler, in doubt

CIE 1931 2° observer · two objectives, one basis

CIE 1931 2° observer · what the data fix

CIE 1931 2° observer, M₁

CIECAM16 hue angle · CIE 1931 2° observer held fixed

Kubelka–Munk against additive · CIE 1931 2°

MacAdam 1942 · CIE 1931 2° observer

a delivery chain, end to end · CIE 1931 2° observer

a modelled sensor's noise · CIE 1931 2° observer

additive white, three models · CIE 1931 2°

filtered over the plane · CIE 1931 2°

one room, six lamps · CIE 1931 2°

relative power · CIE 1931 2° observer

sharpened on stored values against light · CIE 1931 2°

the eye's own bandwidth · CIE 1931 2°

the object-colour solid, with a slope limit · CIE 1931 2° observer

the object-colour solid, with an absorption-band limit · CIE 1931 2° observer

the raw pipeline, a clipped highlight · CIE 1931 2° observer

the raw pipeline, reordered · CIE 1931 2° observer

the special metamerism index · CIE 1931 2° observer held fixed

three uniform spaces · CIE 1931 2°

two corrections · CIE 1931 2°

two matrices and a colour temperature · CIE 1931 2° observer

two uniform spaces · CIE 1931 2°

white added at equal luminance · CIE 1931 2°

ΔEITP against CAM16-UCS in a lit room · CIE 1931 2°

ΔEITP against CAM16-UCS · CIE 1931 2°

ΔEITP through a veil, declared and true · CIE 1931 2°

ΔE₀₀ as the price · CIE 1931 2° observer held fixed

ΔE₀₀'s own local metric · CIE 1931 2° observer held fixed

The CIE 1964 10° observer

The ten-degree functions, measured on a larger field and a larger sample, and the ones a surface-colour laboratory is actually required to use. Every figure here is one the 2° functions would have drawn differently. 72 figures.

CIE 1964 10° observer

The coordinates it is drawn on

Added with the `depth-04` phase, and the band whose existence is that phase's finding. A chromaticity diagram is a projective picture of tristimulus values, and the projection is a choice: the same observer and the same gamut make twelve different pictures on the twelve coordinate systems the discipline has published, and areas measured on them differ by a factor of four and a half. A figure drawn on one of those planes is not about an observer — the observer is the same in all of them — so naming one would name the thing that is held fixed rather than the thing that was chosen. It sits third, ahead of the space and model bands, and the position is load-bearing: *drawn on CIE xy (1931)* carries a year that the model band would capture and *drawn on sRGB rgb* names a space the space band would. Both would be classified into a band about something the figure is holding constant. 19 figures.

drawn on CIE rgb (1931)

drawn on CIE u′v′ (1976)

drawn on CIE xy (1931)

drawn on Display P3 rgb

drawn on LMS (CAT16)

drawn on sRGB rgb

drawn on the best plane there is

Which light the sample was excited with

For the figures whose subject is the band below 380 nanometres. A fluorescent sample returns light at wavelengths it did not receive, so how much it returns is decided by how much ultraviolet the source had — and that quantity is a property of the *lamp*, not of the sample and not of the eye reading it. ISO 13655's four measurement conditions are four different answers to it, and a figure in this band names which one it was drawn under. Naming an observer on one of them would name a thing that is identical in every panel: the colour-matching functions do not move when the ultraviolet does, which is the whole reason the effect was invisible in this collection until the range was widened. The band sits ahead of the instrument-geometry one because a measurement condition is a spectrum *and* a geometry, and it is the spectrum half that decides the answer here — M₃ differs from M₂ by a polariser and by four points of luminance, while M₁ differs from M₂ by the ultraviolet and by seven units of colour. 122 figures.

300–780 nm, CIE daylight basis from 300

CAM16, adapted to the room's own white, M₁

CIE whiteness and tint, M₁

CIE whiteness, M₁ against M₂

CIE whiteness, the ultraviolet of the place

D65 including its ultraviolet

ISO 13655 M₀–M₃

M₀ — incandescent, unspecified ultraviolet

M₁ — D50 including its ultraviolet

M₂ — ultraviolet excluded

reconstructed from the CIE daylight basis at 300 nm

the MacAdam limits, M₁ against M₂

the best least-squares 3×3, M₂ onto M₁

the brightener's own clock, M₁

the eye's own optics, ultraviolet transmittance

A space, a white point or an encoding

Figures whose governing choice is not which eye but which set of primaries, which white, or which transfer function. An observer is upstream of all of these and naming it here would name the wrong thing. 174 figures.

BT.709 luma · sRGB primaries

CIECAM16 · D65 · the room, varied

CIECAM16 · La 100 cd/m² · average surround · D65

CIELAB hue angle, D65

CIELAB, D65

IEC 61966-2-1

Yule–Nielsen on three papers · Oklab under D65

a four-colour press model under D65 · ΔE₀₀ as the price

a lattice, and what lies between it · D50

four inks along the grey axis · D50

one ink, both directions · D50

one measured patch, propagated · D50

sRGB

sRGB transfer function

sRGB · 1200 cd/m² peak

sRGB · 300 cd/m² peak

sRGB · 600 cd/m² peak

sRGB, D65

The three axes a gain is applied along

Added with the `depth-02` phase, whose subject is that adaptation scales three signals and which three is a choice nobody records. A figure in this band holds the observer, the surfaces and the illuminant fixed and varies only the basis the diagonal is applied in — so naming an observer on one of them would name the quantity that is *not* moving. The band sits ahead of the named-model band because a basis is a stronger claim than a model: CAT16 is a model, and the axes it is diagonal in are what the picture is about. 23 figures.

CAT16 basis · a partial gain

basis: 0.25 of the way between them

basis: 0.30 of the way between them

basis: 0.50 of the way between them

basis: 0.60 of the way between them

basis: Bradford

basis: CAT02

basis: CAT16

basis: Hunt–Pointer–Estévez

basis: from the confusion points

basis: the confusion points

basis: the construction of König

A named model, at a stated severity

Anything simulated rather than measured. The site's fifth figure rule: “this is what a colourblind person sees” is a claim no method supports, so the model and its severity are named instead. 197 figures.

Brettel–Viénot–Mollon · two cone matrices

CIECAM16 D, against a two-pool clock

CIECAM16 with a crispening term

CIECAM16 · CAM16-UCS

CIECAM16 · a tabulated surround

CIECAM16, read backwards · average surround, La 100

Luther–Ives 1927, the fit and its residual

MacAdam 1942 · a nonlinear space

MacAdam 1942 · one observer

Shafer 1985, the dichromatic model

Thibos 1992, the chromatic eye

deuteranopia · Brettel–Viénot–Mollon

the ambient at a painted wall · CIE 1976 u′v′

ΔE2000 with kL 1 against kL 2

ΔE2000, mean against the tail

ΔE2000, same pairs

ΔE₀₀ after the channels

ΔE₀₀ as the price · this site's pigment template, varied

ΔE₀₀ · adaptation, varied

An identity asserted in code

The illusion figures, where the whole rhetorical force rests on an identity that does not look true. `assertSame` compares the emitted strings and throws, so the strip reports a fact about the drawing rather than a promise from the author. 35 figures.

an achromatic scene, asserted

checked equal

A capture device, and its sensitivities

Added with the `imaging` phase, and the second band on this site whose governing choice is not an eye. A camera is three functions of wavelength exactly as an observer is, but they are *different* three functions — and a plot of silicon's quantum efficiency, of a colour-filter dye or of what an infrared-cut filter removes does not change at all when the colour-matching functions do. Naming an observer on one of these would name something the picture is not about. The band sits ahead of the model band deliberately: a sensor is a stated *device* rather than a stated model of perception, and the Luther–Ives figures carry a year that would otherwise capture them. 163 figures.

a camera against a duty cycle

a camera and an ambient-light sensor, modelled

a camera profile against a population of eyes

a camera, balanced on each lamp's own white

a diagram of a camera pipeline

a highlight through the sensor's noise

a modelled sensor, not a datasheet

a modelled silicon sensor

modelled silicon sensor

modelled silicon sensor, shot noise

null space of the sensor

raw response, no matrix applied

silicon × dye × filter

two highlights through the sensor's noise

A stated printing condition

Added with the `applied` phase, and the third band here whose governing choice is not an eye. A halftone's area coverages, a Yule–Nielsen exponent, an ink limit and a substrate are properties of a *press*, and a figure about how much of a sheet is covered does not change at all when the colour-matching functions do. The band sits after the two observers deliberately: printing colorimetry is specified with the 1931 2° functions under D50, so most figures in the field name the observer and belong above — what falls here is the remainder, where the governing choice is the condition on the press rather than the eye reading it. 21 figures.

a coated press at a 320% ink limit

A geometry, and how enclosed it is

Added with the `appearance` phase, which brought the first figures on this site whose governing choice is neither an eye nor a standard but a shape. A form factor, an interreflection gain and a metameric match that a corner breaks are all statements about how much of the light comes back — and naming an observer on one of them would name the wrong thing, since changing the colour-matching functions moves nothing the picture is about. The band sits ahead of the stated conditions because a geometry is a stronger claim than a condition: it says which quantity governs rather than merely which value it was held at. 16 figures.

1×1×1 box, closed cavity

1×1×3 box, closed cavity

1×3×1 box, closed cavity

2×1×1.5 box, closed cavity

2×1×3 box, closed cavity

3×1×1 box, closed cavity

Lambertian surfaces, closed cavity

closed form vs quadrature, unit cube

where the quadrature stops

Where the instrument was standing

The other kind of geometry, and it belongs to the measurement rather than to the scene. 45°/0° and d/8° do not measure the same quantity, the difference between them is the interface reflection, and a figure comparing them is a figure about a detector's position. It sits ahead of the stated conditions for the same reason the scene geometry does: it names which quantity governs rather than which value something was held at. 7 figures.

45°/0° and d/8°, Fresnel on a Cauchy index

A viewing geometry, in pixels per degree

Added with the `breadth-01` phase, and the fourth band here whose governing choice is not an eye's spectral sensitivity. A contrast sensitivity function is a statement about *angles*: the same pattern on the same display is above the eye's resolution at one viewing distance and below it at another, and nothing in that changes when the colour-matching functions do. So the figures whose whole content is a spatial frequency name the geometry instead — a pixel pitch and a distance, which together give the pixels per degree everything downstream is computed in. The figures in the same family that draw actual colour patches keep the observer, because the patches are computed through it; the division is which quantity the picture would be wrong without. 35 figures.

10 c/deg, E2 per channel

102 px/° at 0.35 m

12 c/deg, E2 per channel

26 c/deg, E2 per channel

40 c/deg, E2 per channel

41 px/°

41 px/° at 0.60 m

an aperture against an eye, 90 px/°

three channels, cutoffs 55/15/10 c/deg

three channels, cutoffs 61/15/10 c/deg

three channels, cutoffs 66/16/11 c/deg

A temporal channel, and where it gives out

Added with the `breadth-02` phase, and the fifth band here whose governing choice is not an eye's spectral sensitivity. A fusion frequency is a statement about *rates*: the same modulation is seen at fifty hertz and gone at a hundred, and nothing in that changes when the colour-matching functions do. The two temporal channels give out four times apart, so which of them a figure is about is the thing its strip has to name — exactly as the viewing band names an angle rather than an eye. 43 figures.

bounded pool · diffusion length 0.5°

luminance channel

three channels, at their own fusion frequencies

A vocabulary, and how coarsely it is quoted

Added with the `breadth-03` phase, for the one family on this site whose primary data are a measurement of people *talking*. A naming figure's governing choice is neither an eye nor a standard: it is which language's basic terms are being used and how coarsely their centroids are quoted — five CIELAB units here, which is five just-noticeable differences and larger than any industrial tolerance. Naming an observer on one of these would name the wrong thing, since every conclusion in the family survives the observer changing and none of them survives the centroids moving far. 49 figures.

11 basic terms, centroids ±5

eleven basic terms · a population of eyes, this collection's pigment template

A population of observers rather than one

Added with the `breadth-03` phase, and the band that inverts this whole index's premise. Every other figure here names *an* observer because a figure has to be drawn for somebody; these are drawn for two hundred people at once, and what the strip names is the population and the template it was built from. A single observer's name on one of them would be a category error — the picture's whole content is the width of the distribution that a single name hides. 125 figures.

160 observers, designed against the 95th percentile

200 observers, spectral index against the 95th percentile

a fidelity index across a population of eyes

a population of eyes

a population of eyes · the pigment template, varied

a population of eyes, this site's pigment template

a population of observers at 1.4× the reference area, 95th percentile

a population of observers at 1.6× the reference area, 95th percentile

a population of observers at 1.8× the reference area, 95th percentile

population of eyes, one pigment template

population of eyes, this site's pigment template

A device, at a stated operating point

Added with the `breadth-03` phase, and the sixth band whose governing choice is not an eye. A lamp's angular colour, its behaviour under a dimmer and its thermal transient are properties of a *device* — a die, a converter, a junction temperature, a duty cycle — and none of them changes at all when the colour-matching functions do. What the strip has to name is which lamp and at what operating point, because the same device has a different spectrum in every direction, at every drive level and at every temperature. 31 figures.

a viewing booth as a phosphor-converted luminaire

phosphor-converted white LED, stated data-sheet slopes

How the filtered field is read

Added with the `breadth-03` phase, for the family whose subject is this site. Its figures compare one claim read two ways — as the largest departure at a point and as the largest surviving component — so the governing choice is neither the filter nor the observer but the *detector*, which is the quantity the previous phase found could change an answer by a factor of a hundred. Naming an observer here would name the one thing that is identical between the two readings being compared. 6 figures.

one filter, two detectors

An observer built from stated pigment peaks

Added with the `breadth-02` phase, for the figures whose observer is *constructed* rather than standard. A Rayleigh match is a measurement of one person's two long-wavelength pigments, and the figures that draw it are drawn for an observer whose peaks are named in the strip — 566 and 541 nanometres for a normal trichromat, closer together for an anomalous one, and identical for a dichromat. Naming a CIE observer on one of these would name an average over seventeen people, which is the one thing the picture is not about. 56 figures.

one field, one adaptation · this site's pigment template

pigment bleaching, τ 120 s, I½ 10^4.3 td

pigments at 541 and 541 nm

pigments at 552 and 541 nm

pigments at 566 and 541 nm and closer

pigments at 566 and 541 nm, density 0.40

pigments at 566 and 556 nm

pigments at 566 and 566 nm

pigments at 566, 541 and 441 nm

λmax in nanometres, from the receptors

A pooled gain, and how far across the retina it reaches

Added with the `depth-01` phase, for the family whose governing choice is the *extent* of an adaptation pool. A gain that follows the mean of its own neighbourhood is a spatial filter with a clock on it, and everything those figures show — what a stabilised picture loses, in what order, and how large a patch has to be before it goes — is decided by how much retina the pool covers. Naming an observer on one of them would name the one thing that is the same in every panel. 18 figures.

local pool σ 0.5°, achromatic, τ 1/60 s

local pool σ 0.5°, rg, τ 1/60 s

local pool σ 0.5°, τ 1/60 s

pool σ 0.5°, aspect 1.6, axis 0°

pool σ 0.5°, aspect 1.6, axis 30°

pool σ 0.5°, aspect 1.6, axis 45°

pool σ 0.5°, aspect 1.6, axis 60°

pool σ 0.5°, aspect 1.6, axis 90°

pool σ 0.5°, aspect 2.5, axis 0°

A place on the retina

Added with the `depth-01` phase. The macular pigment is the fovea's and nowhere else's, so a person's colour-matching functions depend on where on their retina the light landed — which is what makes the CIE's two standard observers two observers rather than one. What the strip names here is an eccentricity and the pigment profile it implies, because the observer is the same person at both ends of it. 17 figures.

one retina, macular density from 0° to 10°

one retina, macular density from 0° to 20°

one retina, macular density from 0° to 6°

A clock, and what it belongs to

Added with the `depth-01` phase, for the figures whose governing choice is a *rate of change* rather than a state. A room has four time constants in it — two neural, one chemical and one in the lamp — and which of them a picture is about is the thing its strip has to say. The band sits apart from the temporal one deliberately: that one names a fusion frequency, which is about what an eye can follow, and this one names a settling time, which is about how long something takes to stop moving. 12 figures.

four clocks: 1/60 s neural, 120 s pigment, 150 s thermal

τ = 1 s and 60 s

Two models in one computation

Added with the `depth-01` phase, for the family whose subject is this site. Its figures compare one claim computed with a single model against the same claim computed with two, so the governing choice is neither model but the *joint* — and naming an observer would name something identical on both sides of the comparison. 6 figures.

one model against two

The built output itself

For the two figures that weigh this site rather than drawing anything about colour. Their subject is the size of the drawings the collection emits and what an interactive control would multiply it by, so there is no observer, no space and no illuminant to name — the governing quantity is a byte count. It sits before the stated-condition band, which is the catch-all and would otherwise take it on the word *measured*, and having it separate is the point: a figure whose subject is the artefact rather than the subject matter should be findable as such. 5 figures.

the built output, measured in bytes

A stated condition

A luminance, a field size, a viewing distance or an adapting level that the figure is only true at. Naming it is the same discipline as naming an observer, applied to the quantity that happens to govern. 80 figures.

100 cd/m²

a corner as a geometry and as a viewing condition

a signature, and what it identifies

a slit, and what undoes it

average surround, 100 cd/m²

one 5 nm slit on each table · this site's analytic observer

surround ratio 0.18

surround ratio 0.21

surround ratio 0.24

surround ratio 0.25

surround ratio 0.26

what a shutter can and cannot see

The generators themselves · All essays · Named objects