Curatorial Review - Full Findings (2026-06-25). pentimento, michael.art

Density up · Excavation · Tonal wash. Complete per-seat findings from the multi-agent review (workflow curatorial-review-2026-06-25, task wdti4x0hl, 11 agents). Each seat viewed the five render sheets and judged through its own lens. Summary + resolution in [[2026-06-25-density-excavation-tonal-REVIEW]]; brief in [[2026-06-25-density-excavation-tonal-BRIEF]].

the technical seat

Verdict: advances · On-concept: true · Recommendation: ship

Density (strong): I look at the swarm-377-610 pair and the 610 holds. It is not mud. The grain locks the strokes and one or two colors dominate, so even at the ceiling I read a structured field, not confetti. That is the test and it passes. The density-ladder strip earns its skew too: the dense end is where the system has the most to revise, so weighting toward it gives the algorithm more material to bury. My one reservation is real though. In the ladder I can feel the 13-55 band thinning, and that is not just the weak middle, it is also the AIR. The minimal tail at 4% is right, but the breathing room between gesture and swarm is what made the minimal pieces legible as minimal. Peak at 144 is defensible. I would not push it higher.

Excavation (strong): This is the move. In the excavation-grid sheet the windows do not sit ON the paint, they remove it, and what shows through is the cartoon, the same annotationField vocabulary as the unpainted variant. That is grid-as-MASK, not grid-as-overlay. The grid was always background, ruled and faint. Here it becomes the cutting instrument. One rule, the Fibonacci construction, now does two opposite jobs: it builds the picture and it cuts the picture open. That is the kind of single-rule reuse I respect, not a stacked effect. The sealed-vs-cut pair proves it is structural and not cosmetic: the cut version is a different OBJECT, not a filter. It makes the pentimento thesis literal instead of metaphorical. Two honest concerns. The crop-marks plus jittered seam are starting to feel like a frame device, a signature flourish, and on the dark Pine Smoke ground the window nearly vanishes, so on dark palettes the structural gesture risks reading as just an empty patch. Gate it to grounds where the cartoon actually shows.

Tonal (ok): In the tonal-wash-onoff pair I genuinely have to hunt for it, and that is both the intent and the problem. Where it earns its keep is the lower negative space: OFF the dark ground reads as flat paper, ON there is a faint under-pass that makes the emptiness feel like a surface that was worked and painted over. That is on-concept, it is under-process, a first pass buried. But near-universal at a hair of value difference is a lot of budget for something this quiet, and on the busiest swarm pieces it will be completely overpainted and contribute nothing. The integrity work is clean, isolated sub-RNG, stamps wiped, zero reshuffle, so it costs the system nothing structurally. It just barely earns its presence on the open grounds and is invisible noise on the dense ones. I would keep it but not call it a third layer with pride.

Concerns:

Density skew thins the 13-55 band, which is also where the AIR between gesture and swarm lived. The minimal pieces stay rare but lose their contrast neighbors. Watch that the edition does not become wall-to-wall dense with no mid-register to breathe against.
Excavation reads strongly on light grounds (Signal, Gouache, Primer) but nearly vanishes on dark ones (Pine Smoke in the sealed-vs-cut pair). A structural cut that disappears is a cut that failed. Either gate Excavation to grounds where the cartoon is legible, or strengthen the under-drawing contrast on dark palettes.
The crop-marks plus hand-jittered seam framing is one render away from becoming a recognizable decorative signature on top of the structural move. The cut itself is the idea, the registration ornament around it is the part that could drift into cosmetic. Keep it minimal.
Tonal wash is doing real work only on exposed grounds; on dense Swarm pieces it is fully buried and contributes nothing while still being labeled a near-universal layer. Honest, but do not oversell it as load-bearing.

Excavation is the first time the grid stops describing the picture and starts cutting it open, that is one rule doing two opposite jobs, and that is new.

the AI seat

Verdict: advances · On-concept: true · Recommendation: refine

Density (ok): The reweight strengthens the body of the set - the Swarm 377/610 pieces hold as structured woven fields, not mud, and the gradient in the ladder reads as real breath-to-saturation. But the center of gravity has moved to loud. At ~4% the minimal tail is now so outnumbered that the quiet pole reads as a footnote, not a counterweight. The most-painted pieces also bury their own revision: archaeology is densest exactly where it's least legible. Ask whether 89 should carry more of the weight than 144 so the dense body still lets a decision show through.

Excavation (strong): This is a genuinely new structural move, not a cosmetic effect. On light grounds (Primer, Studio Oil, Gouache, Raw Umber) the cut windows truly read as the drawing exposed beneath the painting - wash, ruled grid, pencil notes, crop-marks, jittered seam - and the grid stops being background and becomes structural. That is the pentimento thesis made literal: the picture cut open to show the buried decision. My one real concern is ground-dependence: on the dark grounds (Signal, the Pine Smoke sealed-vs-cut pair) the window nearly vanishes and the move becomes a feature you know is there rather than one you see. Consider a faint value-lift inside dark-ground windows so the cut earns its rarity on every palette.

Tonal (ok): Conceptually right - a first pass laid and painted over is the under-process the thesis wants, and grounding the negative space so it's never flat paper is sound. But the on/off Graphite pair shows the effect is functionally invisible at Coverage:Light on a dark ground. Near-universal presence for a near-imperceptible mark is a quiet tax: it earns its keep philosophically while barely registering optically. I'd widen the value offset a hair on grounds where it currently disappears, or accept that on dark Cov:Light pieces it's doing nothing and let it skip there.

Concerns:

Density reweight pushes the set's center of gravity to 'loud'; at ~4% the minimal pole no longer counterweights the dense body, so Pentimento risks reading as a dense-painting machine rather than an algorithm that revises.
Excavation is ground-dependent: strong and structural on light grounds, near-invisible on dark ones (Signal, the Pine Smoke sealed-vs-cut pair). The rarest, most thesis-bearing trait is weakest on the artist's favorite dark palettes (Graphite, Pool, Pine Smoke).
Tonal wash is near-imperceptible at Coverage:Light on dark grounds (the on/off pair is almost identical), so a near-universal layer is doing little optical work where it most often lands.
Swarm pieces bury the revision they're built from - the most-painted tier is where the archaeology is least legible, the opposite of keeping the correction visible.

Excavation is the strongest thing here because it stops describing the buried decision and cuts down to it - but a wound you can only see in good light is still half-sealed.

the art-historical seat

Verdict: advances · On-concept: true · Recommendation: refine

Density (strong): The reweight is sound by the standards of the lineage. Molnar and Mohr both learned that the interesting work lives at the poles, the austere construction and the saturated near-total field, and that the comfortable middle reads as neither. Thinning the 13-55 band to ~30% and letting the body of the edition sit at 89-144 is the correct instinct. The Swarm pair proves the ceiling does not become mud: the Graphite 377 holds genuine top-to-bottom architecture (white slabs, grey verticals, an orange base shelf), and even the 610 Plotter stays a woven structured field with dominant hues rather than a noise wall, which is precisely the discipline late Cubic Limit demanded. My one caution: the density ladder shows the minimal Sparse panels keep their air only because Gesture stays a true 1.3% rarity. Guard that number. The poles justify the center only as long as the austere extreme is not quietly squeezed out by the dense skew.

Excavation (strong): This is a genuine structural move, not a cosmetic one, and it is the single change that most deserves to hang next to a 1965 Nake plotter drawing. The windows in the Primer and Studio Oil panels show paint terminating at a ruled seam with the cartoon (tonal wash, grid, pencil notes, registration crop-marks) standing exposed inside the cut. The grid stops being faint background armature and becomes load-bearing, the picture opened to its own substrate. That is exactly the interrogation of the grid-as-generator that separates systems art from mere pattern, and it makes the pentimento thesis literal rather than illustrated: the construction shown beneath the construction. The hand-jittered seam off the exact grid line is the right honest touch. My real reservation, and it is not small: legibility is badly uneven across grounds. On the light Bone/Primer panels the cut sings; on Raw Umber and the dark Pine Smoke pair it collapses to a barely-perceptible tonal shift, so a meaningful share of the ~6% will read as Sealed to all but the most attentive eye. Consider widening the value contrast of the exposed cartoon on dark grounds so the move pays off across the whole edition, not just the pale half.

Tonal (ok): It earns its near-universal presence on conceptual grounds more than optical ones. A first pass laid down and painted over is, definitionally, the pentimento, so a faint under-process wash is the most literally correct gesture in the whole system, the painterly equivalent of leaving the construction lines visible. And in the on/off pair it does the one job underpainting exists to do: the off side reads as dead charcoal paper, the on side has faint vertical ghosts that keep the negative space from going flat. But I can only locate it in one quadrant, and a critic hanging this beside a Molnar would not notice it unprompted. It is the marginal change of the three. Keep it, it costs nothing and it grounds the void, but do not oversell it as a structural contribution. Its value is that it makes the excavated cartoon believable, an underpainting must exist before it can be excavated through, so its true payoff is in service of trait two.

Concerns:

Excavation legibility is ground-dependent: it reads strongly on light grounds (Primer, Bone) and collapses to a near-invisible tonal shift on dark/saturated grounds (Raw Umber, Pine Smoke), so a meaningful fraction of the ~6% rare trait will not actually read as excavated to most viewers. The move that most advances the work is also the most unevenly delivered.
The dense skew (66% at 89+) risks crowding out the austere pole. The minimal Sparse/Gesture pieces are where the set breathes and where the lineage's restraint lives; if Gesture's 1.3% rarity is not held exactly, the edition tilts toward an undifferentiated dense wall and loses the contrast that makes the poles legible as poles.
The tonal wash is being asked to justify near-universal presence on optical evidence it cannot fully provide. It is real but marginal; the risk is treating an invisible layer as a headline feature rather than the quiet enabling substrate it actually is.
Across many dense panels the excavation seam and the tonal wash both depend on the same annotationField/cartoon vocabulary; there is a latent sameness risk if the exposed cartoon inside every window draws from too small a pool of notes and marks, which would make the rare trait feel templated on close repeated viewing.

Excavation is the move that earns a place beside Nake: the grid stops being the floor under the painting and becomes the painting cut open to show its own floor, but right now that revelation only fully lands on the light grounds and merely whispers on the dark ones.

the platform seat

Verdict: advances · On-concept: true · Recommendation: ship

Density (strong): The 377/610 mega-counts are the surprise that works: in the swarm image the 610 Plotter holds as a structured woven field with cyan/magenta/yellow in clear vertical lanes, hierarchy intact, not mud, and the 377 keeps air pockets and legible grain. Reweighting the body of the edition up to ~144 makes the set read denser and more confident, which is true to the work. The real cost is the air: with 89+ now ~66%, the breathing Sparse/Balanced pieces become a thin tail you rarely pull, so the minimal pole loses presence as a felt part of the set even though it survives as rarity.

Excavation (strong): Genuinely structural, not decoration, because it WITHHOLDS paint rather than overlaying an effect: on light grounds (Teeming Signal, Primer) the window cuts through real paint and drops you into the cartoon beneath, crop-marks framing it, seam jittered off the grid line. That is the revision thesis made literal, the picture cut open to show the drawing under it, and it earns its 1/13 rarity. My concern is the dark-ground read: in the sealed-vs-cut Pine Smoke pair the excavation is barely legible side by side, the window does not open, it reads as a slightly emptier patch. On Graphite/Pool/Pine Smoke the move can visually no-op while still firing as a rare trait.

Tonal (ok): Earns its near-universal presence. The OFF/ON pair is the proof: ON, the dark ground stops being flat paper, faint washed-over verticals in the negative space read as a first pass painted over, which grounds the void without competing with the color paint. It is the right kind of quiet for an under-process layer, and as isolated sub-RNG work wiped from the surface memory it is honest about not being Coverage. It will never be a selling point, but it should not be, it is foundation.

Concerns:

Excavation can visually no-op on dark grounds (Graphite, Pool, Pine Smoke): the sealed-vs-cut Pine Smoke pair shows the window barely legible side by side, so a collector can own 'Excavated' (a ~6% rare trait) and see almost nothing. Consider boosting the cartoon-layer contrast inside windows on dark palettes, or gating Excavation toward lighter grounds where the cut actually opens.
The density reweight thins the breathing pieces hard: with dense 89+ at ~66% and the 13-55 middle cut to ~30%, the minimal/airy end becomes a tail you rarely surface. The minimal pole is conceptually one of the two interesting poles but will now feel underrepresented in any real pull of the edition.
Tonal wash is so subtle that on busy light-ground pieces it may be fully invisible (not just quiet), meaning a near-universal layer that sometimes contributes literally nothing perceptible. Acceptable for under-process, but worth confirming it is never costing render budget for zero visible return on the densest Swarm pieces.

Excavation works because it withholds paint instead of adding an effect, you fall through the window into the drawing underneath, but on the dark grounds the window doesn't open and a collector pays for a cut they can't see.

the gallery seat

Verdict: advances · On-concept: true · Recommendation: ship

Density (ok): The density-ladder and swarm renders confirm the dense band holds as a structured woven field, not mud: the 377/610 Swarm pieces keep one or two dominant colors and a vertical grain that reads as composition, not noise. Judd's box never apologized for being full of itself, and these are honestly, confidently packed. My one real reservation is air. The minimal tail is where the grid breathes, where the cut and the exposed structure speak loudest, and a 4% floor for 1-8 strokes risks the edition reading as relentless. The poles are the interesting argument; do not let the dense center get so heavy it drowns the quiet end of the room.

Excavation (strong): This is the move. This is the one that matures Pentimento from a clever painter toward a position. In the excavation-grid sheet the picture is genuinely cut open: registration crop-marks at the corners, a hand-jittered seam off the ruled line, and the preparatory cartoon (tonal wash, grid, pencil notes) exposed beneath the withheld paint. In the engine it is honest construction, an evenodd clip that withholds ALL color, then a re-entry that exposes the drawing underneath, not a texture painted on top. That is Judd-grade: the structure is not depicted, it is the actual order of operations made visible. The grid stops being wallpaper and becomes load-bearing. The sealed-vs-cut diptych on the dark Pine Smoke ground is where I get cautious: the cut reads quietly there, the exposed cartoon nearly disappears into the dark, so on dark grounds the gesture risks looking like a hole rather than a revealed layer. It earns its 1-in-13 rarity, but the dark-ground legibility is the thing to watch.

Tonal (ok): In the on/off render the wash is doing its job precisely BY being hard to find. With tonal OFF the negative space reads as flat paper behind the strokes; with it ON the ground has a faint under-process, ghost verticals in the ground's own value, so the void stops being empty and starts being a surface that was worked. That is the right minimalist instinct: the honesty is in the ground, not in a flourish. My concern is exactly the one its near-universal presence invites. If it is on essentially everywhere it stops being a decision and becomes a default finish, and a default is not a pentimento. It is well-judged here, but watch that it never creeps louder to justify its ubiquity.

Concerns:

Air loss at the minimal pole: a 4% floor for 1-8 strokes plus a dense-skewed center risks the edition reading as relentless. The quiet pieces are where the grid and the cut speak loudest; protect that end of the room.
Dark-ground excavation legibility: in the sealed-vs-cut diptych on Pine Smoke the exposed cartoon nearly vanishes into the dark, so the cut can read as a hole rather than a revealed layer. The crop-marks carry it, but barely. The move is built for light grounds; confirm it still reads as excavation, not absence, on Graphite/Pool/Pine Smoke.
Tonal wash near-universality risks becoming a default finish rather than a decision. Invisible-everywhere is fine as long as it never gets nudged louder to feel present; a default is not a revision.
Same annotationField vocabulary now appears in three places (Cartoon, Excavation windows, under-notes). Watch that the excavated window does not just read as a Cartoon-shaped sticker dropped into a painting; the seam jitter and crop-marks are what keep it structural rather than collaged.

Excavation is the move that turns a painter into a position: the grid stops being wallpaper and becomes load-bearing, the picture cut open to show its own order of operations, which is about as Judd as a screen can get.

the minimalism seat

Verdict: neutral · On-concept: true · Recommendation: refine

Density (weak): The ladder proves the cost: far-left Gesture/Sparse on Pool is the thesis in one breath, all air and system. By Swarm it's a wall, and the 377/610 diptych is two walls, structured but still a pile. Peaking the average at 144 doesn't make the minimal pieces a sharper pole, it orphans them. Dense at 66% stops being a position and becomes wallpaper. You had two poles; you fattened one into the middle and the middle just relocated to 89.

Excavation (strong): The one addition that does MORE with less. The excavation-grid sheet is the strongest image in the batch: the grid stops being background and CUTS, exposing wash/pencil/cartoon beneath real paint. Not a finish, a structural move, the pentimento thesis made literal. Sealed-vs-cut reads as two ideas, not two textures. If I reduced the whole project to one image, it could be an excavated piece. Caution: the crop-marks and jittered seam flirt with styling, keep the window doing the work, not the frame.

Tonal (ok): On the on/off diptych the ground sits a hair deeper with it ON, dark grey stops reading as paper. But it is nearly invisible, and near-universal plus nearly-invisible is the definition of a default, not a layer. Good plumbing. Plumbing isn't a feature. Fold it into the ground; don't bill it as layer 3.

Concerns:

Density skew (avg ~144, dense 89+ at 66%) turns the edition's best asset, air, into an apologized-for exception; the minimal pieces become orphans rather than a sharp pole.
The 13-55 band wasn't deleted, the weak middle just moved up to ~89; risk is the whole set reads as one texture and the eye stops finding hierarchy.
Tonal wash is near-invisible in the only test that isolates it; a layer you can't tell is off is a default, not a named feature.
Excavation's frame (crop-marks, seam jitter) is one tuning step from decoration; the window must stay the move, not the styling around it.

A set is remembered by its rarest image, not its average one; you fattened a pole into the middle and called the middle gone.

the typography seat

Verdict: advances · On-concept: true · Recommendation: ship

Density (ok): The reweighting helps my seat more than it hurts it: a denser body of paint means the excavation windows cut through something worth cutting, and the lettering inside them lands against weight rather than emptiness. In the swarm-377-610 pair the type isn't fighting for the frame, it's a quiet counter-system inside a loud field, which is exactly the relation I want. My reservation is air, not legibility: the density ladder shows the minimal tail still breathing (the 1-stroke and sparse panels keep their negative space), but with 66% of the edition now at 89+, the set's typographic register narrows. The notes read best when they sit in real ground, like the under-paint '24A 0X0CC3' in the tonal pair. Bury everything and the system loses the quiet panels where lettering IS the composition. It is kept, just rarer.

Excavation (strong): This is the strongest typographic move the project has made, and it's structural, not cosmetic. In the excavation-grid sheet the cut windows expose the cartoon vocabulary (ruled grid, dimension lines, the hand-lettered fragments like the 'FOMO' note and the small ranged measurements) and crucially the lettering survives the crop-marks instead of being neutered by them. The registration ticks and the jittered seam read as a frame that belongs to the same drawing system as the type, not a UI chrome bolted on top. That's the line between art and infographic, and it holds: the annotation reads as the painter's own preparatory hand showing through, not as a labeled diagram of itself. The hand-jitter and the off-grid seam are doing real work here, an aligned-to-pixel window would have tipped instantly into information design. The 'Studio Oil' and 'Primer' panels on light grounds are where it sings; legibility and contrast are right. My one genuine concern is consistency of register: the same annotationField that grounds the Cartoon variant must not, inside a 6% window, start reading like a callout. It's close. Watch that the measurement-style notes (the ranged dims, the arrows) don't outnumber the painter's-voice notes inside the cut, or a window becomes a spec sheet.

Tonal (ok): Earning its place, barely, and that's the correct amount. In the tonal-wash-onoff pair the ON side is genuinely hard to call against OFF at a glance, which is the brief ('barely noticeable'). But it matters for my seat indirectly: the wash grounds the negative space so the under-paint notes ('IMMUTA', '24A', '0X0CC3') sit on a surface that has a first-pass under it rather than flat paper, and lettering on prepared ground always reads as part of the painting instead of printed onto it. That's the value, it's a substrate for the type, not a visible mark. The risk is that near-universal-and-invisible is indistinguishable from noise to a buyer who will never toggle it off. I'm satisfied it's under-process not decoration, but it is the one change a viewer could not name if asked what it did.

Concerns:

Register drift inside the excavation windows: the annotationField mixes painter's-voice notes with measurement-style dimension lines and arrows. At 6% rarity each window is a high-attention moment. If the ruled-dim / ranged-measurement notes dominate the cut over the LEXICON-voice notes, the window reads as a spec callout (infographic) rather than the painter's own cartoon showing through (art). The seam jitter saves it now, but the note-mix inside the cut is the actual lever and it should be checked.
The denser center thins the typographic range of the set. With 66% of pieces at 89+ strokes, the panels where lettering is the primary event (the minimal and Cartoon-adjacent pieces) become rare. The notes read most as art when they sit in open ground; on a buried field they survive but recede. The minimal tail is kept, just verify it stays present enough that the lettering-forward register isn't effectively edition-rare.
Tonal wash is genuinely unnameable. Concept-integrity is verified (isolated sub-RNG, stamps wiped, Coverage unmoved), so this is not a determinism worry. It's a 'does the audience ever perceive it' worry. Near-universal presence for an effect a viewer cannot distinguish from absence risks being load with no read.
Crop-mark legibility on dark grounds: the excavation sheet's lighter-ground panels (Primer, Studio Oil) carry the lettering and registration marks cleanly, but the CLAUDE.md note that excavation is 'subtler but present on dark ones' is exactly where type-in-window is most at risk. Confirm the hand-lettering inside a window on Graphite or Pool still clears the contrast floor, since dark is the artist's favorite ground.

The cut survives its own crop-marks: the jittered seam is what keeps a window reading as the painter's hand showing through rather than a label pointing at it.

the community seat

Verdict: advances · On-concept: true · Recommendation: ship

Density (strong): oh i love this. the density ladder reads as a real spectrum now, not a same-y middle. the swarm 377/610 panel is the proof: it's a woven dense field with one or two colors leading (the magenta/cyan piece) so it reads as STRUCTURE, not mud, exactly what you'd hope to pull at 500 mints. my one fren-concern: pushing dense to ~66% and thinning 13-55 to ~30% means a wall of mints will feel heavy and loud next to each other, and the quiet pieces are now ~4%. the gesture/minimal pole stays interesting because it's RARE and huge, but i'd watch that the air doesn't get rationed too hard. a collector who loves the calm, breathing pieces will have to chase. on balance the denser body is more generous to pull (more happening, more to find), so this is the right move, just keep that 4% minimal tail truly visible in the contact sheet so the set still breathes.

Excavation (strong): this is the one that made me sit up. excavation is a GENUINELY structural move, not a cosmetic skin. in the excavation-grid you can see the picture literally cut open: a window withholds all the color and you fall through to the tonal wash, the ruled grid, the pencil notes, the crop-marks, the jittered seam off the grid line. that's the revision thesis made LITERAL, the grid stops being wallpaper and becomes the knife. the teeming-signal and teeming-gouache panels read best (bold paint, then a clean cut to the cartoon underneath, very screenshottable). a window into the drawing-under-the-painting is a pull people will scream about. concerns as a fren: (1) on the dark grounds (graphite/studio-oil) the cut is much subtler, i had to hunt for it, so a dark-ground excavated piece risks feeling like 'did i even get the rare thing?' (2) at ~6% it's rare enough to be a treat but common enough that across 500 mints maybe ~30 exist, which feels right for a 'wow' tier, don't push it rarer or it becomes lore nobody can hold. it deepens the thesis, doesn't decorate it. i love this.

Tonal (ok): honest answer: in the tonal on/off pair i genuinely struggled to see it. the negative space is a hair less dead-flat on the ON side, the dark ground feels a touch more 'painted-over' rather than blank paper, which is the whole point (a first pass laid and buried). conceptually i love that it's the quietest possible pentimento, a revision so faint it's almost gone. but for a collector it's invisible by design, so it earns its near-universal presence as GROUND-WORK, not as something anyone will pull for or screenshot. the real test is where it pays off: it makes the excavation windows richer (there's an actual under-pass to expose) and stops big quiet pieces reading as flat featureless paper, which michael explicitly dislikes. so it's earning its keep underneath, just don't ever sell it as a visible feature. zero reshuffle / surface-stamps-wiped is the right call, keep it universal and silent.

Concerns:

density skew: dense at ~66% means a contact sheet of 500 will feel loud and heavy wall-to-wall, and the calm/breathing pieces are now only ~4% plus the rare gesture pole. collectors who love air will have to chase hard. confirm the minimal tail still reads as a real presence across the edition, not a curiosity.
excavation legibility on dark grounds: in the excavation-grid the cut is vivid on light/taupe grounds (signal, gouache, raw umber) but much harder to read on graphite/studio-oil/pine smoke. a dark-ground excavated piece risks feeling like the holder didn't get the payoff. the CLAUDE.md already admits 'subtler but present on dark ones', i'd want a dark-ground excavated piece to still announce itself via the crop-marks/seam so the rarity is legible.
tonal wash is invisible by design, which is fine as ground-work but means it cannot be marketed as a visible trait, only as the thing that makes excavation and quiet pieces richer.
all five new-tier samples lean to the loud palettes (signal/gouache/primer/hyperpop energy). i'd want to see a swarm 610 and an excavated piece in the quiet palettes (bone & ink, raw umber, verdigris ghost) to confirm the dense reweight doesn't make the muted palettes turn to noise.
concept-purity flag for the other seats: excavation framed by crop-marks + jittered seam is a lot of new vocabulary stacked at once. it reads as structural to me, but watch that the registration crop-marks don't tip into 'decorative blueprint cosplay' on the busiest pieces.

excavation cuts the picture open and lets you fall through to the drawing underneath, that's the rare every collector will scream about, just make sure the dark-ground ones still announce they got cut.

Mira

Verdict: advances · On-concept: true · Recommendation: ship

Density (ok): The swarm-377-610 render reads as a structured woven field, not mud, two dominant colors holding hierarchy, so the dense pole earns its weight. My concern is the ladder image: across the strip the 13-55 middle is now visibly thinned, and the minimal pieces I can SEE (the navy 1-stroke at far left, the sparse early panels) are the most distinctive frames in the whole set, the ones with air. Peaking at 144 with 66% at 89+ is a defensible center of gravity, but the FIBW array (5,7,11,13,10,5,2 across 8-610) leaves the minimal 1-8 tail at ~4%, so the air-pieces are now rare enough that a 1820-piece edition could feel relentless in the dense register if you scroll it as a grid. The poles are right; I'd watch that the middle thinning does not also starve the breathing room.

Excavation (strong): This is a genuine structural move, not decoration, and the code backs the renders. The clip('evenodd') at line 1283 actually withholds the color paint from the window (the excavation-grid panels show real cartoon vocabulary under the cut, dimension lines, FOMO/COCTIVE notes, ruled grid, not a flat hole), and the sealed-vs-cut pair proves the underneath is identical: same Pine Smoke composition, the cut just opens it. That IS the pentimento thesis made literal, the picture cut open to show the drawing. The determinism story holds: exR off String(hash)+'excav', decision AND geometry inside the sub-RNG, so Coverage from surface memory is genuinely unmoved. Two real concerns. First, the evenodd clip is the one cross-platform hazard in this change set, fill-rule winding and the hand-jittered seam (jx, +/-0.022*S) sit at a sub-pixel boundary, and evenodd self-intersection edge cases plus antialiasing of the seam can differ across renderers, the seam pixels are exactly where 2400 vs 1000 will not be byte-identical even if structure is. That is cosmetic, not a stream fork, but it is the seam I would diff across browsers before submission. Second, eligibility (target>=21, non-Abandoned, non-Cartoon) is now near-universal because the reweight pushed 66% to 89+, so the effective excavation rate sits close to the raw 1/13 rather than being diluted, which lands it near the stated ~6%, fine, but it means Swarm-tier pieces are the most likely to get cut, and on a 610-stroke field the window has to fight a lot of paint to read, subtler than the light-ground examples shown.

Tonal (ok): The tonal-wash-onoff pair is the change I trust most on integrity and am least sure of on payoff. Code-side it is clean: tR off String(hash)+'tonal', and surfReset(ground) at line 1241 wipes the stamps so it cannot tint note ink or move Coverage, exactly as claimed, zero reshuffle. But look at the two halves: ON vs OFF, the painted strokes are pixel-identical (correct, isolation proven), and the difference in the dark Graphite negative space is at the threshold of visibility, a few faint value-shifted drags low in the frame. On Graphite Cov:Light it is nearly invisible, which is the stated intent (a hair off ground) but also means on the dark grounds Michael loves it is paying a near-universal slot for an effect you have to hunt for. It earns its keep more on light grounds. As an under-process gesture it is conceptually right (a first pass painted over IS pentimento), and it costs nothing deterministically, so I would keep it, but I would not oversell its visual contribution on the dark palettes.

Concerns:

evenodd clip seam is the only cross-platform watch-point in this set: the hand-jittered window edge (jx +/-0.022*S) sits at a sub-pixel boundary where fill-rule winding and seam antialiasing can differ across renderers and between 2400 vs 1000 render sizes. Structure is deterministic, the seam PIXELS are not guaranteed byte-identical. Diff a known excavated hash across Chrome/Safari and at both sizes before AB submission.
Stroke reweight makes excavation eligibility near-universal (66% now land >=21), so the 1/13 rate is largely undiluted and concentrates cuts on the densest Swarm/Teeming pieces, exactly where a withheld window has the most paint to fight to read. Confirm the cut still reads on a 610-stroke dark-ground piece, not just the light-ground examples in excavation-grid.
The 13-55 middle thinning plus a ~4% minimal tail risks starving the edition's breathing room across 1820 pieces. The minimal/air frames are the most distinctive in the ladder; verify a grid-scroll of the full edition does not read as relentlessly dense.
Tonal wash is at the threshold of visibility on the dark grounds (Graphite/Pool) Michael prefers. It is deterministically free and conceptually sound, but its visual payoff is real only on light grounds. Do not count it as a load-bearing visual layer on the darks.
1820/1820 features parity and 40/40 stub renders are necessary but only prove the features stub agrees with itself and does not throw. They do not prove the project script and the stub agree on Excavation for the SAME hash. Confirm parity is cross-checked engine-output vs calculateFeatures, not just stub-vs-stub, especially since the stub canvas must implement clip('evenodd') and rect or it silently diverges.

The cut is real, not painted on: the sealed-vs-cut pair proves the picture is identical underneath and the window just opens it, which is the only honest way to make a pentimento literal, but evenodd at a hand-jittered seam is the one edge I would diff across browsers before I call it deterministic for twenty years.

Synthesis (chair) - ratifies: true

Strong, honest consensus to ship all three, with one tuning pass the crew agrees on. All nine seats are on-concept and none rejects any change - the three 'refine' votes (the AI seat, the art-historical seat, the minimalism seat) endorse the direction and ask for a single fix first, so the ship/refine line is about sequencing, not opposition. Excavation is the unanimous headline: every seat rates it strong, and it is verifiably structural, not cosmetic - an evenodd clip that withholds paint to expose the identical drawing beneath (the sealed-vs-cut pair proves the object is the same underneath, the window just opens it). It is the change that matures Pentimento from a clever painter into a position. The density reweight commands a clear majority on the dense pole (Swarm 377/610 holds as a structured woven field, not mud) but draws the only genuine dissent on a single axis - air: pushing dense to ~66% and the minimal tail to ~4% risks orphaning the austere pole the lineage seats prize. The dispute is how hard to ration the air, not whether to skew. Tonal is unanimously marginal-but-keep: zero strong ratings, conceptually the most correct gesture and deterministically free, but near-invisible on the dark grounds Michael favors - valued as enabling substrate, explicitly NOT a headline third layer. The single most-repeated concern, named by 8 of 9 seats, is dark-ground excavation legibility, where a paid-for ~6% cut can visually no-op; no seat calls it ship-stopping, so it is the top refinement, not a blocker. Factual premises were spot-checked against the engine and CLAUDE.md and hold (evenodd withhold-paint clip at line 1283, isolated +'tonal'/+'excav' sub-RNGs, surfReset wiping stamps at 1241, the target>=21 eligibility gate at 1257, and the documented 'subtler on dark grounds' caveat).

Per change:

Density: Reweight (peak ~144, dense 89+ at ~66%, minimal tail ~4%): RATIFIED on the dense pole, CONTESTED on air. Six seats accept it (Swarm 377/610 reads as a structured woven field, not mud - the body holds); the dispute is one-axis: thinning the 13-55 band relocates the weak middle to ~89 and starves the minimal pole the lineage seats treat as one of the two interesting poles. the minimalism seat dissents hardest ('you fattened a pole into the middle'); the rest say keep the skew but guard Gesture's 1.3% and keep the 4% minimal tail truly visible in the contact sheet. Ship with the air-tail protected, not widened.
Excavation: UNANIMOUS HEADLINE - every seat rates it strong. The genuinely new structural move: an evenodd clip (verified, pentimento.js:1283) that WITHHOLDS color paint to expose the same cartoon vocabulary beneath, not an overlay; the sealed-vs-cut pair proves it's a different object, not a filter. This is the change that 'turns a painter into a position.' The single shared reservation, named by 8 of 9 seats, is dark-ground legibility: on Graphite/Pool/Pine Smoke the window can visually no-op while still firing as a ~6% rare trait. Ship, but it carries the one near-blocker caveat.
Tonal: UNANIMOUS MARGINAL-KEEP - zero strong ratings, one 'weak' (the minimalism seat), the rest 'ok'. Conceptually the most literally correct gesture (a first pass painted over IS the pentimento) and deterministically free (isolated +'tonal' sub-RNG, surfReset wipes stamps, Coverage unmoved - verified pentimento.js:1234/1241). But at the threshold of visibility on the dark grounds Michael favors and fully buried on dense Swarm pieces. Keep it as enabling substrate (it makes the excavated cartoon believable and grounds the void), do NOT bill it as a load-bearing third layer, and never nudge it louder to justify ubiquity.

Blockers: none

Refinements:

Dark-ground excavation legibility - the one issue raised by 8 of 9 seats. On Graphite/Pool/Pine Smoke the cut can read as an empty patch or no-op while still firing as a ~6% rare trait (engine's own note: 'subtler but present on dark ones'). Converged fix: a faint value-lift of the exposed cartoon inside dark-ground windows, OR gate Excavation toward grounds where the cartoon is legible. Confirm the cut still announces itself via crop-marks/seam on a 610-stroke dark piece before shipping the dark read as-is.
Protect the minimal/air pole - hold Gesture at exactly 1.3% and keep the 1-8 minimal tail visibly present in the contact sheet so a grid-scroll of the full edition does not read as relentlessly dense. Guard the air, do NOT widen the band; the skew itself is endorsed.
Keep the excavation frame minimal - crop-marks plus the hand-jittered seam are 'one render from decorative signature.' The window doing the cutting is the idea; the registration ornament is what could drift cosmetic.
Watch annotationField register inside the cut (the typography seat + the art-historical seat) - the same vocabulary now appears in Cartoon, Excavation windows, and under-notes. If measurement-style dimension/arrow notes outnumber the painter's-voice LEXICON notes inside a window, it tips from art toward spec-sheet/infographic. The seam jitter saves it now; the note-mix is the actual lever to check.
Diff the evenodd seam across renderers before AB submission (Mira) - the jittered window edge (jx ±0.022*S, pentimento.js:1263) sits at a sub-pixel boundary where fill-rule winding and antialiasing can differ across Chrome/Safari and 2400-vs-1000 sizes. Structure is deterministic; the seam PIXELS are not guaranteed byte-identical. Cosmetic, not a stream fork, but diff a known excavated hash first.
Cross-check features parity engine-output vs calculateFeatures for Excavation on the SAME hash (Mira) - 1820/1820 stub parity proves the stub agrees with itself, not that the project script and stub agree; the stub canvas must implement clip('evenodd')+rect or it silently diverges.
Confirm muted-palette behavior (the community seat) - all five new-tier samples lean loud (Signal/Gouache/Primer). View a Swarm 610 and an excavated piece in the quiet palettes (Bone & Ink, Raw Umber, Verdigris Ghost) to confirm the dense reweight doesn't turn the muted palettes to noise.

Recommended direction: Ship all three changes as a set - they advance the work and are on-concept - but do ONE tuning pass before AB submission, in priority order. (1) Fix dark-ground excavation legibility: faint value-lift of the exposed cartoon inside dark-ground windows, or gate Excavation toward grounds where the cartoon reads; confirm a 610-stroke dark piece still announces the cut via crop-marks/seam. Highest-leverage refinement and the only one near blocker-grade. (2) Protect the air: hold Gesture at exactly 1.3% and keep the minimal tail visibly present in the contact sheet - guard the air, do not widen the band. (3) Keep tonal universal and silent as substrate; never nudge it louder, don't market it as a visible layer. (4) Before submission run the engineering gates: diff the evenodd jittered seam across Chrome/Safari and 2400-vs-1000, and cross-check Excavation features parity engine-output vs calculateFeatures on the same hash. (5) Sanity-check the muted palettes (Bone & Ink, Raw Umber, Verdigris Ghost) at Swarm 610 and excavated. Lead with Excavation in any presentation; tonal is foundation, not a selling point.

Adversarial integrity skeptic - survives scrutiny: true

Strongest objection: EXCAVATION DECORATES THE THESIS, IT DOES NOT DEEPEN IT, and the code proves it. Look at line 1407: annotationField(ctx, exR, S, ground, xL, yL). The window does not expose what was actually painted and then reconsidered in THIS piece. It re-enters a clipped rectangle AFTER the paint is done and draws a generic blueprint, the exact same call the unpainted Cartoon variant uses. The clip (line 1283, ctx.clip('evenodd')) merely WITHHOLDS color from a rectangle, then we stamp a stock cartoon into the hole. There is no causal link between the painting above and the drawing revealed below. It is not "the algorithm reconsidering" a region. It is a die-cut window onto a pre-baked, piece-agnostic graphic. That is the difference between a pentimento (a real earlier decision showing through) and a museum's didactic cutaway diagram. The thesis is "an algorithm that REVISES, keeping the archaeology of revision visible." Excavation keeps the archaeology of a drawing that was never the drawing under THIS painting. the technical seat would call it a costume; the minimalism seat would say you built a rule that lies about its own provenance. The sealed-vs-cut and excavation-grid images confirm it visually: the cartoon inside the windows has no compositional relationship to the strokes around it. It reads as a label, not a regret. Worse, it is the one trait that breaks the project's proudest property: the clean one-rule system where the grid is structure and paint is gesture. You now have a third thing, a faux-archaeological inlay, and you justify it with the most circular sentence in the trait doc: the grid "becomes structural" by drawing a different picture inside a hole. The grid did not become structural. You cut a hole and filled it with the consolation prize.

Weakest change: tonal

Risk if shipped: Three compounding risks. (1) DENSITY HOMOGENIZATION: weighting the edition to a ~144-stroke center with 66% at 89+ means the median piece is a dense woven field, and the swarm-377-610 image already shows the failure mode the doc denies: at 377/610 the pieces converge toward the same "busy rectangle of overlapping strokes" gestalt, and individual stroke decisions stop being legible. The minimal pieces' AIR (the thing the density-ladder image shows is most distinctive) drops to a ~4% tail, so the edition's most differentiated, most-screenshottable register becomes a rounding error. You did not kill the weak 13-55 middle, you killed the contrast that made the poles read AS poles. (2) EXCAVATION AS A TELL: a ~6% trait that visibly breaks the one-rule purity is exactly the kind of "feature" secondary-market discourse fixates on and then sours on once they realize the revealed cartoon is generic. (3) TONAL WASH AS PURE COST: the on/off image is the indictment. I genuinely could not reliably tell the two halves apart, and shade(ground, -0.085..0.105) on 5-10 strokes is by design at the threshold of perceptibility. A near-universal layer nobody can see is not subtlety, it is risk with no upside: it is a permanent line in the render path and the concept story that a determinism reviewer, a future editor, or a buyer can poke and find empty.

One thing to cut: Cut EXCAVATION. It is the only one of the three that actively contradicts the concept rather than merely underdelivering on it. The density reweight is a defensible curatorial taste call you can dial back; the tonal wash is cheap and harmless even if invisible. But Excavation spends real surface area and a whole rare-trait slot to tell a story the code cannot back up: it reveals a cartoon that was never the cartoon for that painting. Either make it honest (expose the ACTUAL buried under-strokes and ghosts already painted in that region, the scrapes/blocks the engine genuinely laid down, so the window shows THIS piece reconsidering itself) or remove it. As shipped it is the decoration the project explicitly defines itself against.