Orientation, for a cold reader
The painting, the mirror, and the instrument
The facts first, because the picture's structure turns on them. In a tall room of the Alcázar in Madrid, the five-year-old Infanta Margarita stands at the centre of a hushed crowd: two kneeling maids of honour (the meninas who named the picture), two court dwarfs, a dog, two chaperones in half-shadow, a chamberlain paused in a lit doorway at the back, and, at the left, Velázquez himself before a canvas so large its back is turned to us. On the dim rear wall, between framed paintings, a small mirror catches two pale faces: the King and Queen. They are the picture's occasion, the sovereigns for whom a court portrait exists. They are also the only figures in it who are not in the room, only reflected into it.
That inversion has carried three centuries of argument. Foucault opened The Order of Things with it: a painting organised around an absent centre, the royal couple present only as reflection, standing where the viewer stands. Against the metaphysics, art historians offered geometry: Snyder and Cohen showed the perspective's vanishing point falls not on the mirror but at the lit doorway, by the chamberlain's elbow. Whether the mirror reflects the royal couple posing, or the canvas Velázquez is painting, has never been settled. The instrument enters below all of this. It has no iconography and no theory of the gaze; it measures where the picture puts its mass, light, edges and first-pass attention, before any of it means anything. What follows is what those measurements make visible, including two things that turn out to be visible only as failures of vision.
Two conditions scope every number. The object: a 1734 fire, an undocumented trim of the canvas on both sides (more on the right), a repainted cheek on the Infanta, a 1984 cleaning that lifted a century-and-a-half of yellowed varnish. The source: the first file grabbed was a colour-corrected derivative whose lifted shadows the instrument itself flagged as noise; the reading was re-pinned to the raw Prado gigapixel, which restored true tenebrist blacks [Entry 01, stable].
What kind of picture this is
A shadow-key construction with no plan under it
Before any object is named, the whole. Las Meninas reads as a mixed, counterweighted field in the deep shadow register: 89.6% of its tonal mass sits in shadow, and the brightest thing in the frame, the Infanta's dress, accounts for 0.001 of that tonal mass. Colour behaves as a servant of the drawing: almost none of it floats free of form (colour-in-void β 0.049), and its coupling to structure is the highest in the five-work comparison run for this study (+0.301, compare cell) [Entry 06, confirmatory].
And the arrangement is far from where pictures rest. The default-gravity index, which measures distance from the centred, balanced composition a picture falls into on its own, reads 18.8: the most anti-default composition the instrument has measured. Under it sits one technical datum from the object itself: the Prado's reflectography finds no underdrawing. There was no plan on the canvas. Velázquez found this off-balance order in the act of painting it [Entries 06/10, confirmatory · corpus rank].
And the construction has a specific mortality. Perturb the picture and watch what kills it: it survives blur, it survives desaturation, and it dies when the tonal range is compressed (persistence 0.558, fragile to low contrast). That completes a clean three-way split across the corpus: the Bruegel dies by blur (it lives on fine edges), the Klimt by grayscale (it lives on colour), Las Meninas by contrast. This painting is made of tone, and of nothing else it cannot spare [Entry 11, stable · the three-way split exploratory].
The skeleton
Where the weight sits, where the eye goes, and what both ignore
Strip the picture to its structural spine and one figure holds it. The island system's anchor, the region the composition organises around, is the Infanta: stable at her identity across both edge-reading modes (the lock moves between her body and her skirt-base, but never to another figure), inside a single connected bright frieze that runs the width of the room [Entry 06, stable at identity, not location]. Under her, the picture's weight pulls hard down: the bottom third of the canvas holds roughly half the structural mass, the top third under a fifth (δy +0.214), with the horizontal centroid dead-centre to within a hair [Entry 06, confirmatory · horizontal claims provisional on the trim].
Then put the two maps together, because that is where the skeleton speaks to the subject. The centre of the picture's weight and the centre of its first-pass attention nearly coincide (divergence 0.080, both low-centre), and this coincidence is composition-dependent: phase-scramble or patch-shuffle the pixels and the coincidence comes apart (p < 0.01) [Entry 09, null-controlled · this is the spectrum/patch null, not the figure-shuffle used later; the value transfers across a scale cell, spine 0.080 to null-cell 0.075, scoped in the lab record]. Of the placement channels, this is the one that survives its null, and it says something the burial numbers do not: eye and weight agree in this picture, and both agree on ignoring the mirror. Wherever the composition's gravity and its pull are pointed, it is at the low-centre of the room, near the child, and not at the King.
The buried subjects
The mirror is not dim. It is small.
Stand in front of the painting and the royal reflection reads as almost nothing, a pale smudge in the dark wall. The instrument's job is to say why, and the answer is a gradient, not a switch. Follow it across the channels. In paint, the reflection is nearly a full body: the Infanta out-does it only 1.54× in local intensity and 1.44× in edge, the same order of substance she has, the same modelled faces and legible red curtain. In first-pass attention (a saliency model: pull before recognition) the deficit suddenly triples, to 5.01×: the paint survives, the pull does not, and by more than the substance can explain. And then extent finishes it, at 15.2× less share of the canvas. So the burial is not one fact but three, compounding: painted at nearly full body, it already fails to grab, and then it is given a fifteenth of the room [Entry 04, confirmatory · floors 13.6× across 200 jitter draws and all three selection conventions].
This is the separation the eye cannot make unaided. Looking fuses substance, salience and extent into a single impression of "presence," and cannot tell you which is failing; the instrument reads them off one at a time. On that first pass the royals are last of the nine named objects, a 0.33% saliency share against the Infanta's 16.6%, and the dog out-masses the royal couple roughly five times over. Once you have been told the mirror is fully painted, you can verify it by looking, and it does not help them: full-strength paint, no pull, no room. The King is present at full body and given the standing of an incidental [Entry 04; ranks box-set-dependent, shares carry].
The "not dim" half of the claim has an older witness, and the measurement sharpens it. Miller argued in 1998 that the mirror betrays itself as a mirror by carrying borrowed light. Measured against its own wall: the mirror's interior reads 2.3–2.7× brighter than the two dark canvases flanking it and 2× the bare wall, while still 0.66× the doorway. So the royal reflection is locally anomalous and globally weak at once: the loudest rectangle on its wall, and a whisper in the room. Velázquez made the mirror bright enough to read as glass, and small enough to read as nothing [Entry 10, measured · Miller's premise confirmed, his inference untested].
The picture has a centre
Geometry and attention agree, and it is not the mirror
Here is the finding the essay is for. The dark rear wall holds two lit rectangles: the royal mirror, and the open doorway where José Nieto pauses on a bright stair. They are structural opposites. The doorway is the densest local pull of first-pass attention in the picture (local saliency 0.556); the mirror is a void (0.055): a 10.2× gap, in every one of 200 perturbed re-reads [Entries 05/12, confirmatory · prediction 8].
And the doorway's pull is built, not bright. In 1953 Ortega y Gasset described the room's light as a three-part structure, lit foreground, dark middle, lit background; measured, it holds exactly (0.232 / 0.074 / 0.274, the dark middle at a third of either lit plane), a hypothesis confirmed 73 years after it was written. The sharpening the numbers add is that the two lit planes are within 18% of each other: the brightest hole in the picture is barely brighter than the figures. What it has that they do not is a wall that gives it nothing to compete with. Its pull is local contrast against surrounding dark. Once you are told this, you can see it; the back wall reorganises around that door [Entry 10, measured · Ortega confirmed; the contrast sharpening exploratory].
The light sorts the figures
Who steps out of the ground, who sinks into it, and where the painter put himself
The instrument reads each figure by finding its connected mass inside
a search box, and the picture splits its cast cleanly in two on whether that
succeeds. Every lit figure separates from its ground, every time, under
every perturbation: the Infanta, the maids, the dwarfs, the dog all resolve
into a solid component (a quarter to a third of their box). The shadowed
figures do not. Box the two background attendants and the self-portrait and each
one fails: the instrument cannot find a component that is the figure rather than
the figure-plus-wall. A dark body on a dark ground does not separate, and the
picture has several [Entry 12; dark_controls.py,
new boxes on the chaperones].
This corrects a claim worth correcting, because it is the difference
between a fact about Velázquez and a fact about tenebrism. It is tempting to say
the painter dissolved himself into the dark, painted without the edge
every other figure got. The measurement refuses it. His self-portrait is not
even the softest-edged dark figure in the room: read side by side in one
producer's cell, his silhouette contrast runs 0.006
and the shadowed guardadamas behind the Infanta reads
0.003, twice softer; both are far below any lit
figure, which is simply what dark-on-dark costs. (The comparison is
within-cell, which is the only way it means anything: a second producer, with a
looser silhouette threshold, reads him at 0.011 on the same channel. Two
producers, one metric, two cells, both stated.) And the merge is not
the 1984 fill along his seam either: his boundary is uniformly weak, on the
left edge far from any seam as much as on it. So the honest finding is not a
special softness. It is a placement: of everyone in this picture,
Velázquez painted the author into the shadow-merge condition, out at the far
margin with the room's attendants, rather than into the lit frieze with the
child. Where he stood himself is the choice the numbers can see
[dark_controls.py + painter_boundary.py,
measured · the earlier 'one figure of nine' was a box-set artefact, the seam
confound ruled out].
Nieto, in the doorway, is the sharp exception that completes the pattern. His is the only figure who fails to separate while standing in light. Read as structure his aperture is an annulus: a ring of brightness with a dark body at its centre, so that reaching for "the lit doorway" finds the man-shaped hole instead. The painter is an absence inside shadow; the chamberlain is an absence inside brightness. The two figures who frame the act of picture-making, the one who paints it and the one who holds its threshold, are both among the figures the paint declines to hand over whole [Entry 12, I-18 · rendered].
Two footnotes, one historical and one measured. The historical one: the two framing figures held the same office. José Nieto was the Queen's aposentador, her palace chamberlain; Velázquez had been the King's aposentador mayor since 1652. The painter at the far margin and the chamberlain at the threshold, the two the paint declines to resolve, are counterparts by rank as well as by structure, one at each end of the room [historical context, Palomino-tier · not a measurement]. The measured one: the single strongest edge in the entire painting is the right edge of the canvas-back Velázquez stands behind, the picture's own turned back, stronger than any contour of any figure. The loudest line in Las Meninas belongs to the painting we are not allowed to see [Entry 02, measured · handled as an artefact for the edge lenses].
The punctuation
Twelve small accents, spread wide, none on the mirror
Where does the picture place its sharp chromatic incidents, the small, high-contrast notes that punctuate a composition? Twelve, once the dark surface's physical artefacts (a canvas-back edge, two seams, a doorjamb) are set aside: on the painter's eye and palette, the búcaro-menina, the Infanta, the dwarfs, the shadowed chaperones, the dog. Their leverage declines smoothly, 0.83 to 0.49, with no single exclamation mark anywhere. And none on the mirror: given four times the detection headroom, the lens still finds nothing there. The royals carry no punctuation [Entry 03, pixel-verified · cap-checked].
One accent needed checking against the object's history. The red cross of Santiago on Velázquez's breast is later paint, added around 1659, after his knighthood, three years after the picture. It is not one of the twelve: the marks inside the self-portrait sit on his eye and on the red of his palette, both original; the cross carries none. The composition's punctuation is Velázquez's own [Entry 03, checked].
The centre, given away
The strongest central figure, in an arrangement that does not use her
A last, more delicate observation, offered at its honest weight. The picture holds moderate bilateral symmetry about its locked central axis (address 0.425), and the Infanta owns that column. Is her centrality authored emphasis? The experiment: keep the seven figures' exact bodies and permute only their left-right positions, then see where the real arrangement falls. Run twice at independent seeds, 500 permutations in all, it lands at the 70th percentile of reshufflings of its own cast (95% CI 66th–74th): real central content, mildly above chance [Entry 08, measured · two seeds, n=200 + 300 pooled; robust across background fills].
Decomposing that modest number sharpens it. When the Infanta lands central she is the strongest central figure available (her group averages 0.494, every other figure 0.27–0.37), yet the real picture, which does place her centrally, scores 0.419, below its own group's mean: a deficit of −1.32σ pooled across both runs (−1.38σ and −1.27σ), with only a handful of random placements of the other six figures scoring lower. The centrality she could have anchored is given back by where the rest of the cast stands [Entry 08, exploratory · post-hoc, conditional groups n=40 and n=41].
Read alongside the punctuation (twelve accents, spread wide, no climax), this suggests a composition that holds its best central figure and then declines to concentrate on her: attention distributed rather than gathered. A third, candidate-grade reading points the same way: the picture's global torque is subadditive (0.15–0.26), the whole quieter than its loudest part, a stillness made by cancellation rather than by rest [Entry 07, candidate C-3 · cut-stable]. That the three agree is an interpretation of converging channels, not a measurement, and the experiment that grounds the central one can only compare arrangements, never the choice to have a centre at all [interpretation · the −1.38σ is measured, the convergence is read].
Tone: a relative of the Caravaggio
Two Spanish tenebrists at the same corner of the space
The study's own first draft called this the deepest shadow-key in the corpus, and the claim did not survive its own comparison: run through one identical pipeline, Caravaggio's Calling of Saint Matthew reads shadow-mass 0.903 against Las Meninas's 0.896: a statistical tie at the shadow pole, inside a measured reproduction-chain wobble of ~0.13 on this very painting. The superlative is withdrawn, and what replaces it is better: an image fact. The two Spanish tenebrists, painted a half-century apart, occupy the same corner of tonal space: shadow-heavy, colour bound to form (compare-cell β 0.060 and 0.061; Meninas's coupling 0.301 the highest of the five, against Caravaggio's 0.195 and near-zero for the moderns). The catalogues have long said they are family; the numbers agree [Entries 06/11, compare · n=5, placement not ranking].
Same burial, another century
Where Las Meninas lands among the works the instrument has read
This is the corpus's second displaced-subject picture. Bruegel's Fall of Icarus buries a mythological hero in a landscape; this buries a living king in a mirror. The two sit at comparable depth, with the split stated evenly: on absolute share of frame Icarus is buried deeper (0.18% mass and 0.22% saliency against the royals' 0.61% and 0.33%); on the ratio to each picture's biggest actor, mass favours Icarus (25.5× vs 15.2×) and saliency the royals (50.2× vs 21.6×). The same grammar in two idioms (a subject painted at full strength and given the room of an incidental), and unique in kind here, because these buried subjects are the picture's ostensible occasion, standing where the viewer stands [Entry 04, cross-study].
What the instrument cannot touch
The trim, the mirror's referent, and the gaze
Three limits bound the reading. The trim: the canvas was cut down after the fire, more on the right, by an undocumented amount, so every horizontal-centre claim is provisional. Two independent estimates agree only on direction (right-heavy): the seam-derived panel widths, which rest on an untested assumption of uniform loom widths, and a quarters-and-sevenths grid fit whose significance did not survive a threshold sweep [Entries 10/11, exploratory · direction only]. The mirror's referent: whether it reflects the posing couple or the hidden canvas is geometry the instrument does not adjudicate; it measures the reflection's weight, not what casts it. (The two flanking canvases used as brightness controls are themselves contested in attribution (Kahr against the National Trust catalogue), but the control needs only two dark rectangles, which both readings grant.) The gaze: who looks out of the picture, and what that address means, sits above the instrument's ceiling; its symmetry channel is a proxy, and the study marks every gaze-reading as prose [stated scope].
Optional interpretation
Clearly marked, and downstream of every number above
One reading, offered as a reading. A court portrait exists to present its sovereigns, and this one gives their presence away in every channel it has: full-substance paint at one-hundredth of the canvas; the strong aperture to a chamberlain, the void to the King; no punctuation anywhere on the mirror; and a perspective and a first-pass eye that agree the picture's true centre is a door at the back of the room. And the light that is withheld from the sovereigns sorts the rest of the cast too: it hands the picture over to the figures it falls on, the child and her maids, and lets the shadowed ones sink toward their ground, the attendants and, by his own placement, the painter, who put the author of the picture out at the dark margin rather than in the lit company of the subject. The one wholly gathered, wholly lit, wholly found figure is a five-year-old girl who is nobody's subject here. Velázquez composed all of this without a plan on the canvas, hard against the balanced picture a canvas falls into on its own. Whatever the mirror means, the paint has already ruled: the kingdom's centre and the picture's centre are not the same place [reading, downstream of Entries 03–12].
What comes next
The measurements this one makes possible
Three are queued in the study record. The figure-position experiment built here is a new instrument, and the corpus's other central figures (the Klimt, the AI works) now owe a pass through it. The trim is the most decidable open question: a documented conservation measurement would turn every horizontal-centre claim from provisional to firm. And the gaze waits on the one experiment the instrument names but cannot run: eye-tracking on this painting, against which the first-pass model here (the doorway over the mirror, the dress as the point of maximum pull) becomes a genuine prediction rather than a model standing alone.
What this does not prove
The scope, stated plainly
- Not intention. The allocation of frame, light and attention is measured; that Velázquez meant it is a human reading, marked as one.
- Not the mirror's meaning. 0.61% of frame neither confirms nor refutes Foucault; it measures the reflection's structural weight, and "the centre is a door" is a claim about geometry and attention, not about what the picture is about.
- The saliency ranking is over the nine named objects, not the whole canvas; unnamed dark passages are emptier than the mirror.
- The darkness is chain-scoped. The shadow-pole tie is rank-robust but partly a claim about conservation history (fire, trim, repainted cheek, faded azurite, 1984 cleaning); "deepest in corpus" is withdrawn outright.
- Every horizontal-centre claim is provisional on the undocumented trim; both trim estimators speak to direction only.
- The dispersal reading is exploratory and post-hoc (−1.38σ, n=40 of the 300-permutation run; the null run is a separate n=200), and the two-channel convergence is interpretation.
- The absences are selector facts, rendered: "the painter does not separate" is a measured behaviour of a specific reading procedure, shown in the images above, not a claim that no procedure could find him.
- One reproduction chain, one scan. Crisp decimals carry an unmeasured variance bar.
Sources & record
Citations and artifacts
- Museo del Prado, P01174 (Las Meninas, Diego Velázquez, 1656; 320.3 × 279.1 cm); source image the Prado/Google gigapixel via Wikimedia Commons (raw scan pinned; the colour-corrected derivative measured and rejected).
- Snyder & Cohen 1980 and Steinberg 1981 (the vanishing-point geometry
the attention reading independently corroborates); Foucault, Les Mots et
les choses, ch. 1; Clark 1960; Ortega y Gasset 1953; Miller 1998;
López-Rey 1999; Garrido 1992; Kahr; Laing / National Trust, per the
study's graded source list (
meninas_sources.md). - Full measurement record:
labbook.md(Entries 01–12) and its producers:subject_probes.py,jitter_probes.py,convention_jitter.py,lock_margins.py,null_figure_shuffle.py/_robust.py/shuffle_variance.py,clark_grid.py+_sweep.py,accent_clean.py,ortega_probe.py,miller_probe.py,painter_boundary.py,dark_controls.py,shuffle_replicate.py,compare_out/. Instrument findings I-1…I-19, candidates C-3/C-4 (toolkit/CANDIDATES.md,OBSERVATIONS.md). - Instrument lineage: the ISR library: Matisse, Degas, Pollock, Caravaggio, Soejima, Klimt, Bruegel, Cartier-Bresson, Sora Pair, MidJourney Ensemble, Velazquez, and Malevich