ConvergenceJackson Pollock, 1952 - all-over field, measured as relation system

00 Observe

The field before numbers

The reproduction has no visible frame or mat. The study treats raw edge mode as the honest island read for Pollock, while keeping edge-aware as a comparison check. The newer frame profile qualifies this: primary-color tracks are depleted at the border (outer/inner primary ratio 0.805, pullback 0.195).

01 Standard spine

Texture field, not centered emblem

The kernel calls the image a TEXTURE_FIELD: low cohesion μ=0.166, very low orientation stability θ=0.0032, and low structural mask fraction 0.030. The field centroid is nearly centered, but the linework itself refuses a single dominant direction.

02 Islands

A loose triangular relation, not a single anchor

Raw mode finds 18 islands with entropy 0.881. The top relation is 12→14, with bridge cost 0.501. Edge-aware mode changes count by one but keeps the result stable; the crop stability report is stable.

03 Accent failure

The structural lens says zero

The default structural gap lens returns 0 accents. This is not absence of event; it is a scale/channel mismatch. Pollock's incidents are too entangled with the field to behave like isolated tonal marks.

04 Accent recovery

Chromatic coalescing finds bridges

At brush-scale coalescing, the chromatic channel finds 20 accents: 10 bridges, 7 lures, and 3 snags. The combined channel is nearly the same, with 18 accents. That makes the null useful: Pollock's punctuation is chromatic-relation work, not default structural punctuation.

05 Palette roles

The ground is nearly neutral; the accent is struck

The largest swatch is a dark, low-chroma ground. The accent lift is +0.049, a struck-accent read: smaller cream/ochre/high-key systems carry much more chroma than the field that holds the most territory.

06 Convergence probe

There is a hot node, but not a Degas-style scaffold

The repeated Hough probe locates its hottest intersection node around x=0.70, y=0.48. But the orientation resultant only ranges 0.30-0.48, so the result reads as broad crossing pressure rather than a disciplined radiating armature.

07 Density grid

Evenness with tonal weather

The gradient logic is globally consistent: LG is 0.0082. But tonal local/global reads 0.277, so value planes still differ. The 5×8 grid echoes this: edge density CV 0.21, color density CV 0.35.

08 Frame profile

The edge is active; the color pulls back

The raw edge read is still the honest canvas read for this reproduction, but it should not be mistaken for full all-over color. Primary-color density is lower in the outer 8% than in the inner field: outer/inner primary ratio 0.805, border pullback 0.195. This supports a protected chromatic perimeter even while the line field remains active.

09 Deep claim ledger

What survived, narrowed, or failed

This is the running tracker for the Pollock-specific research path. The important correction is that all-overness now splits into two claims: structural edge activity remains high at the border, while primary color pulls back from it.

10 Primary tracks

The color system is not one field

Targeted hue masks keep the primary-color question separate from the OKLab palette clustering. Red, yellow, and blue do not share one scaling behavior; the combined primary field becomes more plane-filling than any single strand.

11 Chroma territory

Palette roles as canvas commitment

Palette roles are now read territorially, not just as swatches. Strict primary chroma occupies 0.405 of the canvas; broad chroma occupies 0.512; canvas/void occupies 0.180. The central-window canvas lift is 0.900, so the center-as-canvas-depth idea is not globally supported by this threshold, even if local central exposed passages matter visually.

12 Figure / ground depth

Color figure, canvas void, black table

The figure-ground reading is probabilistic. Primary chroma has almost 2.99× the color-field intensity of canvas/void, but only 0.57× its edge intensity. That means color can pull forward without being the hardest edge. Black remains more mark/table than depth in this read; white has lower edge than black but higher color-field value than canvas.

13 Push / pull hypothesis

Diffuse chroma, black scaffold, white lift

This is a working hypothesis, not a proved influence claim. In a Hofmann-like push/pull reading, depth is made by contrasts of color, form, and texture on a flat surface. Pollock's chroma can advance because its color-field intensity is high, but it is prevented from becoming a hard dominant figure because its edge intensity is lower than the canvas/void role. Black then stabilizes the field as an active mark/table: it is large, contact-heavy, and scaffold-like rather than pure recession. White behaves as lift or Z-plane: lower edge than black, higher color-field value than canvas, optically mixing forward while also breaking the black scaffold.

The current numbers fit the hypothesis enough to keep it alive: strict primary chroma area 0.405, broad chroma area 0.512, chroma/canvas color ratio 2.99×, chroma/canvas edge ratio 0.57×, black/chroma 3px contact 0.830, white/chroma 3px contact 0.765. The hard part to prove is intention; the measurable part is the push/pull behavior.

The new Canny-gap test makes the evidence stricter: if chroma mostly lives in the Sobel-active / not-Canny gap, it is acting as soft optical field rather than hard contour. If it lives on Canny, the hypothesis weakens. The current role-map verdict is supported_as_behavioral_push_pull.

Interpretation: the data does not say that Pollock's color avoids contour. It says the color has a double life. Some chroma becomes hard event, but a larger share behaves as transition field: strict chroma in structural void is 0.593, while the strict soft/hard ratio is only a modest 1.137. That makes the best claim less absolute and more interesting: the layers need transition zones. Chroma advances and recedes through soft/void passages; black gives those passages a contact-heavy scaffold; white interrupts the scaffold into lift. Push/pull is therefore not a single depth trick but a relay between hard events, soft chroma, black contact, and white breakage.

14 Chroma diffusion vs contour

Can color advance without becoming outline?

The diffusion probe is the stricter proof attempt for push/pull. It compares top-quartile saliency and saturation-gradient activity against luminance Canny capture. The verdict is chroma_diffusion_evidence_mixed. Strict chroma saliency lift is 1.41×, Canny capture share is 0.814, and soft/hard salient ratio is 0.02×.

Interpretation: this asks whether chroma's perceptual force is contour-dependent. The key comparison is against black: strict chroma boundary-Canny alignment is 0.888, while the strict/black boundary-alignment ratio is 0.94×. If chroma is salient but less boundary-Canny aligned than black, then it can read forward through saturation diffusion rather than through conventional outline.

15 Transition zones

The handoffs are the spatial engine

The transition cartography tests the user's strongest refinement: all layers need transition zones. The result is transition_zones_supported. The union of transition bands is large at 0.739, but this should not be read as a literal territory claim because interface bands overlap heavily in an all-over painting. The more reliable interpretation is qualitative and comparative: the transition union is only 0.258 Canny-committed, while void plus soft-Sobel behavior dominates. The softest and largest interface is broad_chroma_to_canvas, which confirms that broad chroma-to-canvas is not a small border but a pervasive atmospheric handoff.

Strict chroma-to-canvas is the key transition for optical depth: it has softness 0.753 and Canny share 0.277. Strict chroma-to-black is nearly as soft at 0.744, while black-to-white is more edge-committed, with Canny share 0.426. That makes white's interruption role sharper than the chroma/canvas atmosphere.

16 Role-by-border all-over audit

All-over splits by visual role

The second border pass tests the all-over claim by role rather than as one field. The verdict is structural_allover_chromatic_pullback. Edge density is border-active with outer/inner ratio 1.230, black is strongly border-active at 1.445, and canvas/void is also border-active at 1.447. Strict primary chroma retreats at 0.805, while broad chroma is closer to neutral at 0.937.

The primary tracks split sharply. Red/orange retreats from the edge, blue is the most protected role by far, and yellow is the exception: its outer/inner ratio is 1.272. That suggests yellow may act as a perimeter tether or atmospheric guide while red/orange and blue concentrate farther inward. So the all-over claim survives only if it means structural all-overness; it fails if it means every role carries equal edge-to-edge weight.

17 Eye-path hypothesis

Attention sits on the relays

The eye-path probe asks whether spectral saliency supports a plausible sequence from canvas/void into chroma, then into black scaffold and white release. The verdict is attention_relay_mixed, which is the right caution. It does not support a simple one-way path beginning mainly in canvas/depth: hot saliency on canvas/void is 0.230, while hot broad-chroma share is higher at 0.409. The dominant hot role is broad_chroma.

But the relay hypothesis is strong. Hot saliency on the full role-contact relay is 0.762, with chroma-to-black at 0.393 and black-to-white at 0.386. So the better claim is not that the eye follows a fixed route. It is that likely attention is repeatedly captured at contact zones where depth, color, scaffold, and lift exchange force.

18 Directional color roles

The color roles are mostly isotropic

This probe tests whether red/orange, yellow, blue, black, and white carry distinct directional jobs. The verdict is role_directionality_mixed. The result limits the hunch rather than confirming it strongly: all orientation coherence values are low. Red/orange is the most directional role at 0.036, blue is the least directional at 0.012, and yellow sits between them at 0.026.

Interpretation: yellow's specialness in the border audit does not come from a strong directional vector. Red/orange does have the clearest orientation bias, but even there the coherence is weak. So the current evidence says color roles differ more by territory, border behavior, contact, and saliency relay than by stable directional flow.

19 Role rhythm

The colors behave like pulses, not arrows

The rhythm probe asks whether the weak direction result is better understood as recurrence. The verdict is role_rhythm_supported. The strongest result is role differentiation: the most pulsed role is black, the tightest-spacing role is broad_chroma, and the most clumped role is broad_chroma. That supports the language of rhythm only if rhythm means distributed recurrence, not a single beat or visual march.

Interpretation: red/orange acts more like an accent pulse than a path; yellow's value is its recurrence and contact behavior, especially where it tethers other roles; blue is sparse and less tempo-driving, but often reads as a cooling interruption because it lives in contact with white and chroma. Black supplies the ground beat: not passive depth, but a recurring scaffold whose contacts regulate the field. White is the release rhythm, resetting local density and pushing marks optically forward.

20 Rhythm stress test

Recurrence survives; contact is controlled

The pushed rhythm test asks whether role recurrence is more than mask area. It compares each role's grid arrangement against shuffled grid cells and compares role contacts against shifted-role nulls. Yellow's mid-scale Moran z is 3.19, black's is 3.87, and white's is 2.99. Positive z means dense cells neighbor other dense cells more than random placement would predict. The verdict here is recurrence_survives_shuffle_null.

The contact null is the surprise: direct role contacts are not enriched against shifted-role controls. Red/orange-to-black lift is 0.73×, yellow-to-black lift is 0.62×, and blue-to-white lift is 0.71×. Since these are below 1.0, the better read is not simple attraction. Pollock's roles recur in organized clusters while also keeping enough separation to prevent one role from swallowing another. Rhythm is therefore pulse plus spacing, not just contact.

21 Near-miss spacing

Composition lives in the interval

The spacing probe follows the rhythm stress test directly. If direct contact is depleted, the next question is whether roles are held in near-miss bands. The verdict is near_miss_spacing_supported. The strongest near-miss pair is black_to_white with lift 1.61×. The strongest canvas interval is black_to_white with canvas share 0.338.

Interpretation: this tests whether Pollock composes by controlled intervals rather than collision. A near-miss lift above 1.0 means the role pair appears in the 4-16px optical neighborhood more often than shifted-role placement predicts. High canvas interval share means exposed ground/void is functioning as an active spacing medium between roles, not merely as leftover background.

22 Interval corridors

The network connects; single pairs fragment

The corridor probe asks whether enriched near-miss intervals remain local halos or connect into longer routes. It skeletonizes the strongest interval masks and measures continuity, reach, border contact, and canvas share. The verdict is interval_network_supported_pair_corridors_fragmented. The strongest single-pair corridor is blue_to_white, but its largest skeleton-component share is only 0.067, even though its reach is 0.976.

Interpretation: this is not one clean highway per pair. The individual role-pair corridors fragment into many short segments. But when the main interval pairs are unioned, they become a connected field infrastructure: union largest-component share 0.844, left-right reach 1.000, top-bottom reach 1.000, and canvas share 0.189. The syntax is therefore distributed: no single pair carries the route, but the interval system as a whole does.

23 Interval junction grammar

The switches are not neutral

The junction probe asks what connects the fragmented corridor pieces. It skeletonizes the union interval network, marks junctions, then compares role presence at junction pixels against ordinary corridor pixels. The verdict is junction_grammar_supported. There are 2322 junction clusters, with 229.9 junction pixels per 1000 skeleton pixels and 42.9 endpoint pixels per 1000 skeleton pixels.

Interpretation: this tests whether the all-over field has local grammar, not just connected texture. The most junction-enriched role is blue at 1.32× ordinary corridor presence. Yellow's junction lift is 1.04×; black is 1.01×; white is 1.07×; canvas is 1.00×. The dominant local role signature is black+white+yellow. So if yellow is guiding, this pass can now say whether it guides as a junction/switching role rather than as a directional arrow.

The pair-level reading is the sharper one: the most junction-enriched corridor pair is blue_to_white at 1.24×. Black-white lift is 1.07× and yellow-black lift is 1.14×. Junction border share is 0.073, while safety-ellipse share is 0.514; that keeps the protected-border question tied to the syntax itself.

24 Phrase grammar

The field starts to read as clauses

The phrase probe asks whether junctions merely switch locally or whether they form recurring clauses. Because the painting does not provide a time arrow, phrases are treated as undirected local syntax: corridor arm A, junction signature, corridor arm B. The verdict is phrase_grammar_supported. The probe found 1449 eligible junctions, 2669 arm-to-arm clauses, and a mean of 2.39 arms per eligible junction.

The dominant phrase is black_to_white__red_orange_to_yellow|black+white+yellow, count 179, share 0.067. The most lifted arm transition is red_orange_to_black__red_orange_to_yellow at 1.26× independent arm-label expectation. Phrase entropy is 0.742, where lower means stronger concentration into repeatable clauses.

Interpretation: this is the closest pass yet to a measurable grammar. It does not prove a fixed viewing path. It tests whether the painting repeatedly stages similar local exchanges. If the dominant phrases keep returning to black/white/yellow while blue/white remains overrepresented in lifted transitions, then rhythm has become syntax: chroma does not simply pulse, it enters the field through repeatable switching clauses.

25 Phrase geography

The grammar is tested against the protected field

The geography pass asks the simple follow-up: if phrase grammar exists, does it behave all-over, or does it gather inside the central safety field? The current verdict is phrase_geography_mixed. Eligible junctions already lean internal, with safety-ellipse share 0.515 and border share 0.208. Against that baseline, black/white/yellow phrase safety lift is 1.01× and border lift is 1.00×.

Interpretation: the phrase geography does not reduce to one protected center. Black/white/yellow phrases sit close to the eligible-junction baseline, so the dominant stage is broadly distributed rather than strongly central. Blue/white is the protected exception: safety lift 1.18×, border lift 0.58×. Red/orange clause-making is different: safety lift 0.97×, border lift 1.07×. So the cleaner claim is split: blue/white helps protect and release the interior, while red/orange and the dominant black/white/yellow stage help carry syntax across the field.

26 Border syntax audit

Which phrases tolerate the protected perimeter?

The third border pass tests phrase families against the safety perimeter and the outer 8% border. It does not claim a literal painted line crosses the oval; it asks whether phrase events sit inside, outside, near the perimeter, or close to the canvas edge. The verdict is border_syntax_split_supported. Eligible perimeter-band share is 0.359, and eligible border share is 0.196.

Interpretation: the split we care about is behavioral. Black/white/yellow behavior is baseline_like; blue/white behavior is protected_or_refusing; red/orange behavior is border_tolerant. The cleaner question is no longer whether Pollock is all-over in general. It is which syntax families are allowed to approach or cross the protected field, and which ones operate as interior release.

27 Phrase saliency coupling

Which clauses become attention events?

The saliency pass asks whether phrase grammar is only structural or also perceptual. Each eligible junction receives a local spectral-saliency score, then phrase groups are compared against that eligible-junction baseline. The verdict is phrase_saliency_coupling_supported. The strongest attention group is dominant_phrase, with hot-patch lift 1.46×.

Interpretation: this is the bridge between our grammar claim and a viewer-facing claim. If a phrase group lifts above the eligible-junction baseline, it is not merely a repeated local structure; it is more likely to become a perceptual event. Blue/white hot lift is 0.79×, red/orange clause hot lift is 0.94×, and black/white/yellow hot lift is 1.19×. This gives us a way to separate field syntax from actual attention pressure.

28 Phrase stability

Does the phrase-saliency claim survive other settings?

The stability pass reruns phrase grammar under tighter, broader, and wider-context settings. The verdict is phrase_saliency_stability_supported. The dominant phrase match share is 0.25, and attention-supported share is 1.00. The minimum dominant-phrase hot lift is 1.46×; the minimum black/white/yellow hot lift is 1.10×.

Interpretation: this is the goldilock test. A stable result does not require every setting to choose the exact same top clause; it requires the same family of behavior to persist. If black/white/yellow saliency lift remains above baseline and blue/white remains border-protected, the theory survives as a family-level claim even when the exact phrase label changes.

29 Phrase route simulation

Attention pressure forms local route circuits

The route pass keeps saliency-active phrase occurrences and links nearby nodes. This is not eye tracking and it does not prove temporal order. It asks a narrower question: do the hot phrase events remain isolated, form local circuits, or resolve into a broad connected route field? The verdict is attention_route_network_mixed. Active nodes: 420; graph edges: 1258; largest component share: 0.08.

Interpretation: the active phrases create route pressure without route closure. The small largest-component share argues against one continuous eye-path highway, while the edge and group counts show recurring local handoffs. Current active black/white/yellow-family share is 0.57; blue/white release share is 0.31. Active safety share is 0.57, and active border share is 0.17.

30 Attention stations

What replaces a single eye path?

The station pass decomposes the active route graph into local components. This tests whether the failed single-route claim leaves behind repeatable catch/switch/release stations. The verdict is local_attention_stations_supported. Components: 44; station components: 23; station node share: 0.72; top-10 component share: 0.43.

Interpretation: if route simulation says the eye does not get one continuous highway, station analysis asks whether Pollock substitutes repeated local holdings. Mixed stations matter because they are not just same-role clusters; they are places where several phrase families touch. Current mixed-station share is 0.87. The dominant station family is black_white_yellow_stage, with station safety share 0.56 and border share 0.19.

31 Frame audit ledger

Pollock full frame; Degas cropped for comparison

Degas stays on Degas Cropped.jpg for compositional comparison. The full Degas reproduction is useful only as an artifact audit. Pollock, by contrast, is tested on the full canvas because the reproduction has no visible frame or mat.

32 Current synthesis

All-over does not mean undifferentiated

The strongest Pollock result is the combination: globally consistent gradient field, significant tonal divergence, chromatic bridge accents, protected primary-color borders, role-specific border behavior, large chroma territory, chroma diffusion vs contour, broad transition zones, relay-heavy saliency, weak directional specialization, role-specific rhythm with non-random recurrence/contact, near-miss spacing, interval corridors, junction grammar, phrase grammar, phrase geography, border syntax audit, phrase saliency coupling, phrase stability, phrase route simulation, attention stations, and stable self-similar structure D=1.490 with slope stability 0.940. The painting behaves less like a centerless fog than like a distributed network whose hierarchy keeps reappearing at smaller scales.

n = 1, and one reproduction at that. Every number is from a single image. JPEG compression, the photographer's lighting, and color profile all directly affect edge detection, chroma thresholds, saliency, and box-counting. The raw vs edge-aware checks and reproduction-artifact flags reduce the risk, but they cannot fully separate a fact about the painting from a fact about this photograph of it.