One is built of light. The other, only of color.Mark Rothko — No. 31 (Yellow Stripe), 1958, and Untitled (Green on Maroon), 1961

Orientation, for a cold reader

Two paintings, a turning point, and an instrument that reads no meaning

The facts first. In 1958 Rothko took a commission for the Four Seasons restaurant in the new Seagram Building, and his painting began to change. The radiant reds and oranges of the early 1950s gave way to a darker, more sombre register of maroon, plum and near-black. He would later withdraw the murals rather than hang them in a restaurant, and give a set to the Tate. The two paintings read here sit on either side of that turn. No. 31 (Yellow Stripe), 1958, is on its front edge, still warm, still luminous. Untitled (Green on Maroon), 1961, is three years into it, cooled and quieted. So the pair is, among other things, a small chronology: two snapshots straddling one of the decisive shifts of Rothko's career.

They look, at a glance, like the same kind of object: a few soft-edged rectangles of color floating on a color ground, painted to the edge, no drawing, no figure. But they are built in opposite ways, and the difference is the subject of this essay. No. 31 stacks a pale salmon over a deep crimson, with a bright yellow ground that also reads as a stripe between them. Its colors differ mostly in lightness: pale, dark, bright. Green on Maroon floats a single muted green in a wide grey-violet field, near-monochrome, its lower edge dissolving into the ground, with two small dark marks adrift in it, one high at the right and one low at the centre. Its colors barely differ in lightness at all.

Rothko cared about exactly this. His surfaces are thin, unvarnished washes, built in many diluted layers, and conservation study has found he engineered how those layers met: he modified his paints with resins and glues to control flow and drying, and in places used a barrier to keep one layer from blending into the next [cited, not measured]. He also tested his edges deliberately, masking them off to try a hard border against a soft one. The point of recording this now is that the difference this essay measures, between a boundary drawn by light and a boundary drawn by color, is the kind of thing Rothko is known to have controlled at the level of the paint itself.

The instrument is the Image Structure Reader, which measures shape, contrast, weight and color and knows nothing of restaurants, sublimity or intent. It cannot see a mood or a void. It can say whether a painting has a countable interior, whether its boundaries are drawn by light or by color, and where its soft transitions live. That narrowness is the point: a painter whose pictures are carried by color where other painters used drawing and light is the hardest case for an instrument built to read content's structure, and what such an instrument still finds, and fails to find, is worth knowing. Where a claim below is a reading rather than a measurement, it is marked.

Thirty insides, and none

One painting has a countable interior; the other has zero

The simplest measurement is the sharpest. The instrument counts the distinct internal structures in a painting, the separable regions of its interior. No. 31 returns thirty. Green on Maroon returns zero [void topology, stable]. This needs no threshold and no tuning; it is an integer count, and the two integers are the whole argument in miniature. One of these paintings has an inside the instrument can enumerate. The other does not.

An integer like this is only worth as much as what it turns out to be counting, so the thirty were checked [verified against the image]. They are not noise or canvas texture. They trace the boundary of the crimson band against the yellow, the place where the lower rectangle meets its ground, broken into segments around its perimeter. They cluster on the crimson edge and barely touch the salmon one, for a reason that turns out to matter: crimson on yellow is a strong step in lightness, and the instrument's count works by lightness, so it finds the crimson boundary and mostly misses the paler one. No. 31 has an interior because Rothko gave it edges of light.

Green on Maroon returns zero because it has no such edges. Its green and its grey-violet are almost identical in lightness; the boundary between them is carried by hue, not by light, and a light-based count cannot see a hue boundary at all. The green rectangle, which is the entire painting, is invisible to the measure. This is not a defect the painting shares with a blank canvas; it is the opposite. The zero is the measurement of a very particular achievement.

Light reports it empty; color shows the whole painting

The blindness is systematic, and it is the same across every luminance tool

The zero is not one measure's fluke. It is what every light-based reading of Green on Maroon returns. The instrument's edge finder locates less than a tenth of a percent of the painting, and, worse than finding little, it finds the wrong thing: it locks onto the small dark mark near the upper-right corner and reports that speck as the painting's whole structure, because in lightness there is almost nothing else to hold. That is not a measure failing quietly; it is a measure confidently mistaking a stray mark for the composition, which is the systematic blindness in a single detail. Its tonal-plane measure, which asks whether the quarters of a picture differ in value, reads No. 31 at its ceiling and Green on Maroon near its floor [tlg 1.0 vs 0.13, narrowed]. Even its measure of slow, atmospheric fades, run on lightness, finds No. 31's crimson glow and declares Green on Maroon nearly featureless. Four different tools, one verdict: to light, the green painting is empty.

Then run the same paintings through color instead of light, and the green painting comes back in full.

The middle of that figure is what the instrument's light-based eye sees: near-black, both paintings nearly erased, Green on Maroon entirely so. The right side is the same two paintings read by color, and both compositions are wholly present. The green rectangle that no light measure could find is, to a color measure, unmistakable. The painting was never empty. The reading was blind.

The soft internal atmosphere tells the same story, and it flips cleanly between the two works. Rothko's fields are not flat; they breathe, veiled and modulated. Measure that breathing by lightness and by color separately, and No. 31's atmosphere is a thing of value while Green on Maroon's is a thing of color [narrowed]. The green painting's color atmosphere is, in fact, the strongest such signal in the pair. Its veil is real; it is simply woven in a register a light-reading eye cannot register.

Color carries it

The measure of the thing the first two beats have already shown

There is a single number for what those figures show: how much of a painting's structure is drawn by color where light draws nothing. On this measure No. 31 reads low and Green on Maroon reads high [CBI, narrowed]. No. 31's boundaries are boundaries of light, so the color measure finds little beyond what light already drew. Green on Maroon's one boundary is a boundary of color, so the color measure finds it where light found nothing. The number names, in one figure, the fact the count and the blindness already established: light carries No. 31, color carries Green on Maroon.

That measure is the most fragile thing in this reading, and it was the most carefully checked, because a number that could carry a false headline should meet its theorem before the essay, not after. Built against constructed cases it behaves exactly as it should: a pure color boundary at equal lightness reads at its maximum, a pure lightness boundary at its minimum. But it carries no defense against image noise, and canvas and brushwork are noise: a trace of grain is enough to collapse the clean case from its maximum to near its floor [anchor, Entry 04]. So the number could not be quoted as it stood, and three separate controls were run before this essay leaned on it.

The decisive one turned out to be compression. The two plates are not matched on it: the auction catalogue reproduction of No. 31 is some fourteen times richer per pixel than the museum's web reproduction of Green on Maroon, and heavy compression coarsens color into eight-pixel blocks, exactly the fabric this measure reads [Entry 11]. That asymmetry could have manufactured the whole result, so it was tested rather than argued. It cuts the other way. A genuine color boundary at equal lightness is untouched by compression, holding its maximum even at punishing quality settings, while a reading that rests on noise is destroyed by it: push No. 31 down to the museum plate's regime and its color-autonomy number falls toward zero [Entry 12]. Green on Maroon's number survives that treatment; No. 31's does not. The compression the two plates do not share therefore separates them the same way the clean controls do, and if anything it works against Green on Maroon, suppressing the very value it reads high on.

The denoising control agrees, across twelve settings, with the gap widening rather than closing as it is pushed harder. It is reported as an ordering and not as a corrected value, because the denoiser fails its own theorem case: run it on the clean color boundary and it destroys the signal the measure exists to find [Entry 12]. It tells you reliably which of these two paintings is which; it does not tell you what either number would be on a perfect plate. Three controls, then, from three directions, and the direction never moves: the split is the paintings', not the instrument's.

It is worth saying that this measure is doing something the standard tools for this subject do not. A published computational study of Rothko's color clusters the pixels and measures the perceptual distance between the clusters: what colors, how different [cited]. That is a real and useful question, and it is a different one. A symmetric color distance cannot tell a boundary drawn by light from a boundary drawn by color, which is the distinction this whole reading turns on. The instrument's contribution here is not that it finds Rothko's colors; it is that it can say which kind of edge each color makes.

One temperature each

He commits to warm or cool and works inside it

A smaller finding, clean and non-obvious. Sort each painting's color into warm and cool. No. 31 is almost entirely warm: salmon, crimson, yellow. Green on Maroon is almost entirely cool: green, grey-violet. Neither works the tension between warm and cool at all [narrowed]. Rothko, in each of these, commits to a single temperature and modulates within it, warm against warm or cool against cool, rather than setting one against the other. The pair, chosen for its warm-cool contrast, turns out to hold that contrast between the works and refuse it inside them. Whatever vibration each painting has, it is not the vibration of complementary color; it is the close interval of a family.

What the instrument cannot touch

The sublime, the intention, and a color that may have moved

Almost everything ever written about these paintings lies outside what was measured, and the border should be plain. The instrument has nothing to say about the sense of the sublime that Rothko's fields are built to induce (the reading Robert Rosenblum placed at the centre of this painting in "The Abstract Sublime"), about the enveloping scale, about the near-religious quiet of a room hung with them. It cannot read intent: that Green on Maroon's boundary is color rather than light is a measurement, but that Rothko meant it that way, though the conservation record makes it likely, is a historical claim resting on how he built his surfaces, not on anything here. A color field has no prior for the instrument to lean on, which makes the wall between structure and intention higher here than anywhere.

Two cautions specific to these objects. First, the color measured may not be the color painted. Rothko's unvarnished, light-sensitive surfaces are known to shift; analysis of another of his paintings could not rule out a color having drifted from brown toward black over decades of light [cited]. The warm-cool and color-boundary readings of Green on Maroon rest on a color that may itself have moved since 1961, a risk a resolution sweep cannot touch and only a condition report could resolve. Second, Rothko used "maroon" loosely, as a word rather than a coordinate; no single color number here should be read as evidence about his intent. Both are reasons to hold the color findings as structure, not as the artist's mind.

What comes next

The three measurements this pair makes possible

The strongest of these is confirmation from the object itself. If Green on Maroon's boundary is color-only because Rothko built it that way, a cross-section or condition report, of the kind the Tate produced for its own Seagram canvas, would show the paint layering that suppressed a lightness step, and would confirm or refute the pixel finding against the physical fact. That is the move that would turn a perceptual reading into a technical one. Second, the chronology: this pair is two points on a line from 1958 to 1970, and the natural question is whether the count and the color measure move steadily as the palette darkens or whether 1961 sits at a bend. Third, the pair should be run through the standard cluster-and-distance method beside this one, to show a reader exactly where a color-distance reading and a boundary-type reading agree and diverge.

Two paintings, and a third

Where the counter-pole stops being a framing and becomes a result

Set these two beside the painting the instrument read just before them, Malevich's Black Square, and the three resolve into one statement. The Black Square is all edge and no interior: a hard black boundary around a field the instrument reads as empty, structure committed entirely to the edge. Green on Maroon is its exact inverse: all interior and no countable edge, a field the instrument reads as full of color and empty of any boundary light can find, structure refused to the edge and given to hue. No. 31 sits between them, an interior held together by edges of light. Three paintings, three answers to the same question, where does a picture keep its structure. Malevich committed it to the boundary and made the boundary hard; the earlier study found the square's interior empty to the gradient system, its structure all at that boundary. Rothko, in Green on Maroon, refused the boundary to light and dissolved it into color, and the interior count found the inverse: no boundary a light-reading eye can hold, and a field that is nonetheless full. The instrument did not impose that contrast; each study measured its own painting on its own terms, and the two results are the negative of each other. That is what a pre-semantic instrument is for: not to say what these paintings mean, but to measure, in numbers a reader can check, the different places three artists chose to put the thing that holds a picture up.

What this does not prove

The scope, stated plainly

• Not meaning, and not intention. Structure was measured: a countable interior, a boundary of light or of color, an atmosphere of value or of hue. The sublime, the void, and Rothko's aims are his and his readers'; the numbers neither confirm nor touch them.
• The color measured may not be the color painted. Rothko's light-sensitive, unvarnished surfaces can drift; the color-boundary and warm-cool readings rest on a color that may have moved since 1961. A condition report, not a better scan, would settle it.
• The color-autonomy number is controlled, not raw, and it is an ordering. The measure has no noise control of its own, so it is reported here in its noise-controlled form, and its bare threshold is not relied on. The denoiser separates these two paintings consistently but fails on a clean test case, so it ranks them without establishing what either would read on a perfect plate. The two threshold-free findings, the interior count and the light-versus-color blindness, carry the argument on their own.
• The two plates are not matched on compression. An auction catalogue master against a museum web reproduction, some fourteen times apart per pixel, on the channel a color measure reads. The direction was tested against that gap rather than assumed, and survives it, but the color numbers should be read as a within-pair ordering on unequal sources and never as values in their own right.
• Bounded by two reproductions. An auction plate and a museum plate, not the objects, and the plates flatten Rothko's built surface.
• The mark in Green on Maroon is measured, not attributed. Its origin, the artist's hand, an underlayer, or handling, is not decidable from the plate and is not claimed.
• Two works are not the turn. The chronology reading is a hypothesis about a line this pair only begins; it is not established from two points.

Sources & record

Citations and artifacts

• Mark Rothko, No. 31 (Yellow Stripe), 1958, oil on canvas, 198.8 by 175.9 cm, private collection (Christie's, New York, 2025). Untitled (Green on Maroon), 1961, mixed media on canvas, 258 by 229 cm, Museo Nacional Thyssen-Bornemisza, Madrid, inv. 729. © 1998 Kate Rothko Prizel & Christopher Rothko / ARS, New York and VEGAP, Madrid; reproduced for private structural research, not redistribution.
• On the Seagram commission and the darkening register: Artlyst, "Mark Rothko's Seagram Murals." On the engineered surface, resins and glues and layer-separation and the hard/soft edge tests: the "Rothko Methods" conservation summary and the National Gallery of Art, "How Mark Rothko Made Paintings on Paper." On pigment instability and a possible brown-to-black drift, and "maroon" as a loose term: Kunsthaus Zurich Rothko restoration notes. On the paint stratigraphy a cross-section can recover: Tate, on the conservation of Black on Maroon.
• Art-critical literature for the readings the instrument does not make: Robert Rosenblum, "The Abstract Sublime" (ARTnews, 1961), on Rothko and the sublime; Rothko's own framing, which was emotion rather than color (to Selden Rodman, 1957: "I'm not interested in relationships of color or form... I'm interested only in expressing basic human emotions"), so that the instrument's "color carries it" is a claim about the picture's mechanism, not the artist's stated aim; Christopher Rothko, Mark Rothko: From the Inside Out; Anna Chave, Mark Rothko: Subjects in Abstraction; David Anfam, Mark Rothko: The Works on Canvas (catalogue raisonné), for the works' identity and dating; James E. B. Breslin, Mark Rothko: A Biography, for the Seagram commission and its withdrawal.
• The cluster-and-distance precedent this reading is set against: a published K-means and Delta E analysis of color transitions in Rothko (ScienceDirect, 2025).
• Full measurement record: labbook.md (Entries 01–10, the color/tone subset), PREREGISTRATION.md, and PROVENANCE.md. Producers: read1_gate.py, read2_color.py, read4_A.py, color_subset.py, cbi_anchor.py, diag_pm_sweep.py, edge_void.py, lcr_atmosphere.py, figures.py. The color/tone measures used here (CBI, void topology, tonal planes) are the instrument's own; the study-local additions and the essay's reading list are in CANDIDATES_local.md and CRITIQUE_NOTES.md.
• A note on this study's corrections: the instrument's edge and gravity measures, and its color-autonomy number, each failed on these paintings in an instructive way and were set aside or controlled before this essay leaned on them; the record of how is in the lab book, not here.

Instrument lineage: Matisse, Degas, Pollock, Caravaggio, Soejima, Klimt, Bruegel, Cartier-Bresson, Sora Pair, MidJourney Ensemble, Velazquez, Malevich, and Rothko