Board synthesis, 2026-06-13. pentimento, michael.art

Provenance. Generated 2026-06-13 from the curation-review-prep scope (each seat's lens brief), not the full agents/<name>.soul.md file. Michael ran this board from his laptop against the v1 static engine and the 2026-06-13 spread. For a canonical record, the team may rerun with full soul files via Eve.

Eve. Post-review synthesis. v1 static aesthetic engine. Read the nine seats yourself; this is the map.

Michael, nine seats reviewed v1 and the convergence is unusually high. I am not going to soften it: the engine cleared the bar, the statement did not, and one seat found a hole nobody else was looking at. Here is where it actually stands.

What each seat said, in a line

the technical seat (code honesty): Advance the engine, the PRNG and render path are provably deterministic and the surface-memory fix is canon-worthy, but two mechanics (Ghost-visibility trait, surfStamp footprint) assert more archaeology than the pixels keep, and do not ship the v2 prose on the v1 bytes.
the typography seat (typography): The LEXICON register is genuinely strong and the curves help, but the M apex, B joins, # colliding with substrate ticks, S softening at small sizes, and flat note-weight hierarchy are real defects, and the static engine cannot deliver the temporal typography the statement promises.
the platform seat (platform fit): Visual system and cohesion are Curated-quality, but Worked at 78.8% is too dominant and the statement promises a living piece the static engine does not deliver; fix the statement (or ship v2) before any submission.
the minimalism seat (minimalist): Piece 1 is the strongest object in the spread, but "5 strokes" that is really ~50 visual elements is a category error in the Sparse label, piece 7 has no center of gravity, and the note system is over-annotating the quiet pieces.
the art-historical seat (institutional): The hand is real and pieces 1 and 4 would hang, but the statement and the engine are arguing about two different artworks, and that gap is disqualifying for a wall label until closed.
the gallery seat (Marfa/long-arc): The dark grounds (2, 3, 4) are gallery-ready and reward sustained looking, but wall text that lies poisons the room, so the statement must match the static engine or v2 must exist before submission.
the community seat (soul/access): The soul is real and lives in the LEXICON and in piece 4, and the v1 archaeology is a complete and true idea, so ship v1 with a statement that tells the truth about v1, the asymptote belongs to v2.
the collector seat (collector): Strong visual engine and real identity, but the statement-to-engine gap will damage the floor within 48 hours of discovery, and the Worked/Saturated/Monumental middle needs a fast-readable secondary differentiator.
the AI seat (meta-honesty): The paintings keep their corrections but the project hides its two biggest ones, that a machine co-wrote the engine and that the temporal soul is not built, so name both in the work or hold the drop.

Where they agreed (this is the signal)

The visual engine advances. Unanimously. Every seat, including the two hardest graders (the art-historical seat, the AI seat), said advance the visual system today. Cohesion is solved, the grain lock and Fibonacci discipline read as one hand, the determinism is clean and verifiable. Nobody asked you to rebuild the engine. That is the headline and you earned it.

The statement cannot ship attached to the v1 mint. Eight of nine, independently. This is the consensus that matters and I want to be precise, because the handoff already told us the statement describes v2 and was written for Marfa. That was correct for an application that is a vision document plus sample renders. It is not correct for a drop page or a wall label, where the buyer receives the object. the technical seat put the sharpest point on it: an immutable on-chain token makes the contradiction permanent and trivially checkable by anyone who opens the script, and collectors in this space read the script. Eight seats reaching the same gate from eight different lenses (code, type, platform, institutional, gallery, soul, collector, meta) is not eight opinions. It is one structural fact.

The fix everyone offered is the same fix, and it is cheap. Rewrite the statement around what v1 actually is: the kept archaeology of revision held in one finished frame. Five separate seats (the art-historical seat, the gallery seat, the community seat, the collector seat, the technical seat) volunteered that this is a strong, sufficient, defensible artwork on its own and does not need the asymptote to stand. the community seat went further and argued the daughters passage lands harder without the temporal promise, because a static finished thing made for someone you love says "this moment is worth preserving," which may be the truer claim. Hold the asymptotic-striving text, intact, for v2.

The LEXICON is the asset, and the personal-word swap is the move. The painter register is named the strongest and most-screenshotted element by nearly every seat. Three seats (the art-historical seat, the typography seat, the AI seat) independently said: replace the generic register words with your own list before anything institutional. This is already the #1 open thread in CLAUDE.md and your memory. The board just confirmed it is the lever that converts a competent generative set into an authored one. If you do one creative thing before submission, do this.

Where they dissented sharply

Piece 8 (Studio Oil, Restless): formal courage or stopped render? The cleanest disagreement in the set. the minimalism seat reads the left-heavy composition with the empty right register as a deliberate formal argument that works. the platform seat and the collector seat read the same pixels as an unresolved image that ran out of canvas and will read as "didn't finish" at thumbnail scale. Note the trap: this dispute only exists because of the statement gap. Under the v2 temporal frame the incompleteness is a live fact; under v1 it reads as a bug. Resolve the statement and this argument partly resolves itself, but the composition risk on the left-clustering Restless outputs is real and worth an audit either way.

Is the static engine even the right object, or just the cheaper one? the art-historical seat dissents hardest here: a static set, however handsome, competes in a crowded field of generative abstraction where it is "good but not singular," and the asymptote is the only idea with the conceptual weight to clear a museum bar. the community seat, the gallery seat, and the collector seat push back: the v1 archaeology concept is complete and true on its own. This is not resolvable by editing, it is a strategic call about ambition, and you should make it consciously rather than letting the v1/v2 ordering decide it for you.

The CRYPTO register: kill it, thin it, or own it. the art-historical seat wants it pulled back hard or fully committed to as declared irony, arguing GM/COPE/DEGEN on the same surface as a daughters statement reads as the artist hedging. the typography seat wants it thinned to the 8 to 10 strongest and the dated entries (LFG, SER, IYKYK, WEN, NGMI) retired. the community seat flags the specific collision of "COPE" and "1 OF 1" sitting together as cutting against the emotional project. The disagreement is degree, not direction: everyone agrees it is currently the weakest-aging register, nobody agrees on how much to cut.

the AI seat stands alone, and that is exactly why to weigh him. No other seat raised AI co-authorship. He argues a project whose entire ethic is "keep the correction visible" has an aesthetic obligation (not a legal one) to admit the engine was substantially written by a model, and that the "agent team" framing must be declared a fiction or stripped from anything a juror sees. This is the one genuinely new, uncomfortable finding. You do not have to agree it is a blocker. You do have to decide on the record, because if you do not and a reviewer notices, it reads as the un-kept pentimento in a project about kept ones. He also gives you the reframe that turns it into an asset: the daughters and the agents are the same tension, the work you reach for is literally made of the agents.

What I would change before you apply

I am sorting these by whether they gate a submission. Do not let the long tail of craft nits distract you from the two things that actually gate.

Gating (do before any application):

Decide the statement, in one direction, and commit. Either ship a v1-honest statement (archaeology in a finished frame) now, or hold the drop until v2 exists. Do not present the v2 statement over the v1 engine to anyone who receives the object. This is the single decision the whole board is waiting on.
Decide the authorship question on the record. Either add one honest note to the GENERATIVE register (the AI seat's "you already write SFC32, you are one word from honest"), or consciously decide not to and strip the agent-team fiction from the Marfa application and drop page so it never reads as a literal claim.

High-value (cheap, high-leverage, do alongside): 3. Swap in your personal LEXICON. Already your top open thread; three seats reinforced it. Add specificity to the generic verbs (COVER, START AGAIN, NO) that will repeat hundreds of times across 720 pieces. 4. Fix the two mechanics that lie. Derive Ghost visibility from surviving ghost area (you already have SURF to measure it) so a piece cannot report archaeology it buried, and either tighten surfStamp to the ribbon footprint or write one comment owning the approximation. For a project named "the truth showing through the paint," a trait and a surface model that misreport are the one failure mode you cannot ship. 5. Audit the Fate distribution. Worked at 78.8% is roughly 568 of 720 pieces telling the same story. At minimum lift Abandoned (conceptually central, currently 7.9%) and decide Sparse (1.3%, under 10 pieces) deliberately, either own it as explicit scarcity or bring it to 3 to 5% so it anchors a real tier.

Craft (real, not gating, batch them): 6. Typography: lower the M apex (~0.35), fix the B doubled cap, make the # distinct from substrate ticks, raise the SMOOTH threshold (~70°) or make it scale-aware so the S holds at small sizes, give over-paint notes a heavier weight range (THIRTEEN should be the heaviest thing on the canvas), and replace 0X0D so it does not collide with the 0X#### hash notes. 7. Thin the CRYPTO register to the strongest 8 to 10 and retire the dated entries. 8. Decide whether Sparse pieces should cap note count (the minimalism seat: six notes on five strokes is the artist explaining himself in the margin), and add a fast-readable secondary differentiator to the Worked/Saturated/Monumental middle (the collector seat: rarity must be visible at a glance or the floor flattens). 9. Re-verify Fibonacci grid legibility on all four dark grounds. If the grid is the cohesion signature, it has to survive on Pool, Graphite, Deep Teal, and Slate.

The honest readiness call

The engine is close. The project is not.

The visual system would pass a Curated conversation today, and that is a real achievement that I do not want buried under the criticism. But "confident ship" is gated on two decisions you have not made, not two bugs you have not fixed: what the statement claims, and how the project accounts for how it was made. Both are decisions, not engineering. You can make them this week.

You are deliberately not rushing, and that instinct is correct. The board did not find a project that needs more building. It found a project that needs you to decide what it is honest about before it becomes immutable. Make those two calls, do the LEXICON swap and the two trait-honesty fixes, and you are genuinely close. Ship the v2 statement on the v1 bytes and you hand every reviewer who reads the script the one flaw that retroactively discounts everything else you got right.

Consensus

The visual engine advances unanimously: all 9 seats, including the two hardest graders, said advance the visual system as-is. Cohesion is solved, grain lock and Fibonacci discipline read as one hand, determinism is clean and verifiable. Nobody asked for an engine rebuild.
The artist statement cannot ship attached to the v1 mint: 8 of 9 seats independently flagged the temporal-statement vs static-engine gap as the gating issue. An immutable on-chain token makes the contradiction permanent and checkable by anyone who reads the script.
The fix the board converges on is the same and is cheap: rewrite the statement around what v1 actually is (kept archaeology of revision in one finished frame). Five seats independently called that a strong, sufficient, defensible artwork that does not need the asymptote to stand.
The LEXICON painter register is the project's strongest and most-screenshotted asset, and swapping in Michael's personal word list (already the #1 open thread) is the move three seats independently reinforced as the lever from competent-generative to authored.
Pieces 1 (Bone & Ink, Sparse) and 4 (Slate, Abandoned/Bias) are the spread's standouts; piece 4 was named strongest by five separate seats as the one where concept and surface land simultaneously.

Sharp dissents

Piece 8 (Studio Oil, Restless): the minimalism seat reads the left-heavy composition as deliberate formal courage that works; the platform seat and the collector seat read identical pixels as an unresolved render that ran out of canvas. The dispute exists largely because of the unresolved statement gap.
Strategic ambition: the art-historical seat argues a static set is 'good but not singular' in a crowded generative-abstraction field and only the asymptote clears a museum bar; the community seat, the gallery seat, and the collector seat hold that the v1 archaeology concept is complete and true on its own. This is a conscious strategy call, not an edit.
The CRYPTO register: the art-historical seat wants it pulled back hard or fully committed to as declared irony; the typography seat wants it thinned to the strongest 8 to 10 with dated entries retired; the community seat flags the COPE / 1 OF 1 collision specifically. Agreement on direction (weakest-aging register), sharp disagreement on how much to cut.
AI co-authorship: the AI seat stands entirely alone, arguing a project about keeping corrections visible is obligated to admit a machine co-wrote the engine and to declare the agent-team framing a fiction. No other seat raised it, which is precisely why it deserves a deliberate on-the-record decision.

Recommended changes before applying

GATING: Decide the statement in one direction and commit. Ship a v1-honest statement (archaeology in a finished frame) now, or hold the drop until v2 exists. Never present the v2 temporal statement over the v1 engine to anyone who receives the object.
GATING: Decide the AI-authorship question on the record. Either add one honest note to the GENERATIVE register acknowledging the machine co-authorship, or consciously decide against it and strip the agent-team fiction from the Marfa application and drop page so it never reads as a literal claim.
HIGH-VALUE: Swap in Michael's personal LEXICON word list (top existing open thread); add specificity to the generic verbs (COVER, START AGAIN, NO) that will repeat hundreds of times across 720 pieces.
HIGH-VALUE: Fix the two mechanics that misreport. Derive Ghost visibility from surviving ghost area (SURF already measures it) so a piece cannot report archaeology it buried; tighten surfStamp to the ribbon footprint or document the approximation as deliberate.
HIGH-VALUE: Audit Fate distribution. Worked at 78.8% is ~568 of 720 pieces. Lift Abandoned (conceptually central, 7.9%) and make a deliberate call on Sparse (1.3%, under 10 pieces): own the scarcity explicitly or raise to 3-5% to anchor a real tier.
CRAFT (batch): Typography fixes: lower M apex to ~0.35, fix B doubled cap, distinguish # from substrate ticks, raise SMOOTH threshold to ~70 degrees or make it scale-aware so S holds at small sizes, give over-paint notes a heavier weight range (THIRTEEN heaviest), replace 0X0D so it does not collide with 0X#### hash notes.
CRAFT (batch): Thin the CRYPTO register to the strongest 8-10 entries and retire the dated ones (LFG, SER, IYKYK, WEN, NGMI, DEGEN, TO THE MOON).
CRAFT (batch): Consider capping note count on Sparse pieces (six notes on five strokes reads as self-annotation); add a fast-readable secondary differentiator to the Worked/Saturated/Monumental middle so the floor does not flatten.
CRAFT (batch): Re-verify Fibonacci grid legibility on all four dark grounds (Pool, Graphite, Deep Teal, Slate). If the grid is the cohesion signature it must survive on dark grounds.

Ship readiness

The visual engine would pass a Curated conversation today, but confident-ship is gated on two decisions Michael has not made (what the statement claims, and how the project accounts for being AI-co-authored), not on bugs, so v1 is close but not yet shippable until those two calls are made.