Board review, 2026-09-15 (the tape). pentimento, michael.art

The full board, eleven seats: the nine curatorial seats plus the audit lead and the contract seat. Each received the same packet, the fork record and thirteen renders, one per register, including one hash rendered with and without the tape, and seven questions: does it hold as the work, the untaped share, the border dial, the drafting layer, the trait surface, the curation submission on file, and anything nobody asked. Each was prompted for hard critique.

Synthesis

Scores: the art-historical seat 7, the technical seat 6, the minimalism seat 6, the typography seat 7, the platform seat 7, the AI seat 6, the collector seat 7, the community seat 7, the gallery seat 7, the audit lead 6, the contract seat 6. Mean 6.5. All eleven: hold with changes. Nobody said it does not hold; nobody said ship it as is. Ten of eleven open with the same pair, 04 beside 13: the tape turned a painting into a work on a wall.

Does it hold as the work: As a work, yes, unanimously. As Pentimento, 5 to 6. The art-historical seat (if the over model ships), the typography, collector, community and gallery seats: the same work grown up, and a truer pentimento, because the frozen mainline is now literally the underpainting. The technical, minimalism, platform and AI seats: the rule changed, revision became masking, and it needs its own name. The audit lead and the contract seat: a different distribution and a different work until a ledger entry with a lineage block says otherwise. All eleven agree the decision has to be written before anything is packaged, and that the working label on the branch is not a title.

The filing, eleven of eleven: do not swap the work after applying. The submission on file holds no script; the platform reviews the real script at its review build, so an undisclosed fork upload there is the swap. Two honest paths, and the board splits: write to the platform now, before it answers, with the fork beside the filed script (collector, community, gallery, contract); or ship the mainline as filed and submit the tape as its own project, or withdraw and refile (the other seven). The platform seat: collectors buy the distribution, not the superset.

Strongest concerns, three seats or more each:

The untaped coin. Six seats say structural only, kill the 12/13 coin (technical, minimalism, platform, collector, community, gallery): a piece untaped because it lost a roll has no story, and none of the artist's eighteen favorites were untaped. Three say keep about 22% as the floor (art-historical, typography, AI): the untaped pieces are the edition's only pigment-level proof that the mainline is underneath. The audit lead and the contract seat: either value is auditable; publish it with the carve-outs.
The slogan register came back through the side door. Every seat names BREAK THE GRID with its arrow, on three of twelve review pieces, as the excavation-at-39% lesson repeated in words. Chain words are still in the deck and the boosted favorites pool. The dev vocabulary (WONTFIX, OVERFIT, REVERT, SHIPPED BROKEN) is the new crypto register: "the trading floor left and the standup walked in". And the ledger the June board asked to kill surfaces as the full hash on Torn and reserve pieces.
The drafting layer stacks. Four seats: one drafting event per sheet. The dimension numerals are not in the sequence and count nothing a viewer can tally: letter the span in lattice units or drop the number. The strike at 30% will be a tic (art-historical, AI, collector) against typography and platform (the best new mark). Registration crosses and dimension lines split the room: the printer's own marks (art-historical, gallery, platform, collector, contract) against annotations of annotations (technical, minimalism, typography).
The trait surface. Unanimous: cut Taped blocks and Loose blocks, counts that contradict Discipline on 8 of 113 taped pieces. Majority: fold Proof into Reserves or into Tape; fold Hot note into Inks (six seats) against four who would keep it as the grail trait. The contract seat found a ninth undocumented key, Turbulence, residue of a rejected probe: cut. Eight becomes three to five.
Lines that show. Not a dial, a rule: art-historical (condition on discipline), typography (weight Keyline by sparseness), gallery (partway, to a third). Turn it flat: technical, minimalism, collector. Leave it: platform, AI, community. Everyone who looked closely: the keyline is at the threshold of visibility, gone on a phone; widen it before any rate moves.

Verified against the engine (re-derived by the seats): the lineage anchor holds; no randomness outside the hash; traits from a null context equal render traits 40 of 40; the over model is byte-stable; additivity held on 60 of 60 unchanged pieces. BUT: the "byte for byte" guarantee is overstated. With the tape off the fork diverges from the frozen mainline on the Sheet register and the Cartoon, 14 of 153 hashes (9.2%), and the identity tool prints "sheet" for those instead of comparing. The tape switches are live and the split model is dead code. No browser has rendered the fork. And the packet document names an edition size and a price that were never on the application.

Critical path (the board's, in order):

Write the decision: a ledger entry that names this Pentimento with a lineage block, or names a second work; then write to the platform before any review-build upload.
Kill the untaped coin, or publish it with carve-outs at 2000 hashes; the board leans 6 to 3 to kill.
The slogan pass: BREAK THE GRID out of the deck and off the refusal; chain words and the dev vocabulary out of the visible register.
The drafting ration: one event per sheet; numerals in lattice units or none; the strike to 1 in 13; Proof gated to Dense or above.
The keyline made visible, then Border as a rule conditioned on discipline and density.
Traits: cut Taped blocks, Loose blocks, Turbulence; fold Proof into Reserves and Hot note into Inks.
A release battery the fork can fail: identity at 2000 hashes, three browsers diffed at the clip edges, a fresh freeze tag with the switches removed.
Edition: settle 233 against 377, and reconcile the packet document.

Ideas nobody asked for, worth keeping: the tape reading the painting it covers (the AI seat); tape-off in the live view as a way of looking (the community seat); the tape theater as the exhibition and never the token (the gallery and platform seats); the five refusals kept with the work as its record (the art-historical seat).

Bottom line: The tape holds, and the scores sit within a quarter point of June's. The honest reading across eleven seats is one sentence: keep the rule, cut the costume, and write down what this is before anyone else sees it. The rule is tape, paint, lift, and the residue. The costume is everything lettered on top of it, plus a trait table that grew when it was told to shrink. The record has to catch up with the work: a name, a lineage block, a letter to the platform, a battery the fork can fail.

The art-historical seat (score 7/10)

A mask that leaves evidence adds to the field, and under the over model the June painting becomes the literal pentimento; the maker's slogans lettered over it must go.

Put 04 beside 13: 13 is the better painting, 04 the better edition piece. What the fork adds to seventy years of the field is not the grid, which Nake ruled and Molnar disordered, but the mask treated as a substance. Here the tape has width, drift, torn ends and, decisively, residue: paint that ran past the tape stays on the sheet, as Newman left the ragged paint of the taped zip. It holds as Pentimento on one condition, the over model, the frozen mainline painted whole underneath. What is borrowed is the writing; Molnar put her mother's handwriting on the plotter sheet and never once a slogan.

Concerns:

The slogan register returned by the side door: chain words in the deck, ON CHAIN, GM, GRAIL and VIBE in the boosted favorites pool.
The deck is too small for the edition: PHI four times, BREAK THE GRID three, in twelve pieces. A recurring phrase is a logo, not marginalia.
The dimension numbers are not checkable, neither Fibonacci nor a count of anything ruled on the sheet.

Recommendation: Hold with these changes. Keep the untaped share near 22; do not swap the work after applying. Most important move: purge the slogan register, then make every number on the sheet countable or absent.

The tape is the first mask in seventy years of this field that leaves evidence; do not let the maker's slogans be the other thing it leaves.

The technical seat (score 6/10)

The mask is one honest rule and it holds; everything lettered on top of it is the old engine talking over the new one.

I read the compositor before I read the pictures. Every mark in this engine goes through one function, and that function now takes a mask: an inner clip on the block and an evenodd clip outside the tape's outer boundary, so the overrun stays on the sheet. That is the only genuinely new thing in the fork; 06, taped blocks with the underlayer showing as hairlines, is Scully by algorithm. What I do not accept is the layer that arrived after the mask: dimension lines nobody can check, registration crosses, proof lettering, STET. None are consequences of the tape. A mask is a plan, not a revision; a different work, with its own name.

Concerns:

The isolation claim is half true: the fill pulls, the accents and the refusal stroke draw on the mainline stream.
Two bracket systems on one sheet: 07 carries the excavation's brackets and the tape's crop brackets, the same mark drawn twice.
The keyline, about 1.5px at 700, is not there at arm's length.

Recommendation: Hold with these changes. The coin goes; ship what was filed. The single move: strip the still to what the tape did (mask, residue, keyline, reserve) and let the animation carry the process.

The tape is the first new rule this engine has had; the pencil is the old one refusing to leave.

The minimalism seat (score 6/10)

The tape is the work and the proof sheet is the excuse; keep the rule, cut the costume.

Put 13 next to 04. Same hash. One is a painting that got tired. The other is a painting that got told where to stop. That is the whole experiment and it wins. Then take the color out: in grayscale 04 and 06 still read as blocks with a rule behind them; 08 and 12 turn into grey weather. Not a proof sheet. Masking. Tape, paint, lift. Meanwhile the sheet wears four other trades' uniforms: the draftsman's dimension lines, the printer's crop marks, the proofreader's STET, the copywriter's BREAK THE GRID. Only the painter was in the room when the tape went on. Revision is about what was there; masking is about what never arrived. A different work, one name each.

Concerns:

The reserve is the sharpest idea on the table, a block the paint never reached, the pentimento inverted, and it is filed behind a trait called Proof.
The on-top ledger survived June as the buried Signature, and 07 letters the full hash across the middle of the sheet.
Hot note: "at a tape edge, never centered," and in 03 the tape edges are invisible. A rule the viewer cannot see is a rumor.

Recommendation: Hold with these changes. Untaped share zero: if the tape is the work, every piece is taped. Three traits, not eight; edition 233. Single most important move: take the coin to zero and the annotation devices down to one, the keyline.

Tape. Paint. Lift. Everything else on the sheet is asking for credit.

The typography seat (score 7/10)

The lettering moved on top of the paint, where a mark laid last has to be right; the keyline and the STET are the best typographic decisions since the alphabet, but the drafting layer is spoken by four hands.

I judged it as a mechanical: who drew each mark, and can I trust it. The keyline passes: four pencil strokes, a hair off the tape, a system behind it. The strike passes better; TRY CLEVER STET in 08 is a regret regretted. Where it does not hold is governance: a draftsman, a press, a proofreader and an issue tracker, four hands at one pencil weight. Under the paint a note could be half read and still carry meaning. On top of the paint a dimension line that reads 36 has to be 36 of something, and it is not. It holds as a Pentimento and needs no new name: the whole first painting now shows through.

Concerns:

The numerals are not the signature: none is in the sequence. Letter the span in lattice units, or drop the number.
Events stack: 08 carries BREAK THE GRID with its arrow, the strike and 13 13 13. One sheet, one event.
The strike has two grammars, delete and keep after all, and the edition has both.

Recommendation: Hold with these changes. Keep the untaped 22%, where the under-notes survive; weight Keyline by sparseness, not a dial. The single most important move: one event per sheet; fold Proof, Hot note, Reserves, the strike and the refusal into one Event trait.

A proof sheet is the one page where every mark gets checked; the fork put the lettering on top, and now it has to survive being checked.

The platform seat (score 7/10)

A better edition than the mainline, but not the work on file, and you cannot release B on an acceptance for A.

04 beside 13 is the whole story: the tape turns a weave into a sheet with a shape. My trouble is what it is work of. A Sealed piece like 12 covers the painting that annotates itself completely; the revision the project is named for sits under the tape as hairlines. The favorites board says the artist's eye has moved to the tape. Then the tape is the work, and it needs its own name, statement, freeze and application. If the filing is accepted, the mainline ships, full stop; a fork with 44% more code is a different project in an accepted project's slot.

Concerns:

The fork has no features script, no platform build, no battery record: not submittable today under any name.
Determinism is verified with the tape off, not on: a piece can carry 13 clip edges and no tape-on cross-browser run exists.
Sealed hides the thesis: about one in eight of the edition would carry the name Pentimento with no visible pentimento.

Recommendation: Hold with these changes. Kill the coin; leave the border dial, since borderless at 19% is excavation at 39% again; keep the strike. The single most important move: settle the name and the filing before touching another dial, and never release the tape under the mainline's acceptance. If the rare registers stay, 377 is the floor.

The curators say yes to a painting, not to a promise; if the tape is the work, give it its own name and its own application, and let each be judged as what it is.

The AI seat (score 6/10)

It holds as a work, but it is not Pentimento: the tape never looks at the painting it covers, and the fork's only real feedback loop ran on the artist's taste.

04 is the June painting with eleven of thirteen blocks taped and filled, and the mainline surviving in two windows and in the residue past the tape: a pentimento in the literal sense. What I mind is how the windows were chosen. No branch of the discipline code consults the surface memory the engine already keeps; a mask that does not look is a stencil. Then the authorship: the Sheet accident is the only moment a machine gave him something he had not specified. The engine lettered OVERFIT on 06 and WHOSE VOICE? on 12. It is asking my question in pencil.

Concerns:

The loose-block choice is blind to the painting: in 08 the loose blocks fall where the coin fell.
The hot note is a loved accident made a rule: five favorites carried one red mark by chance; now every two-ink paper piece gets one.
The untaped coin went to 12/13 on a sample that could not contain an untaped piece; the remaining 22% is the only pigment-level evidence the mainline is underneath. Keep it as the floor.

Recommendation: Hold with these changes, under its own name. Keep brackets and dimension lines; cut the refusal's caption; ship what was judged. The single most important move: let the tape read the painting it covers, driving the loose-block choice from the surface memory, no new model, so the fork earns the word reflection instead of borrowing it.

The tape never looks at the painting it covers; the painting already wrote OVERFIT in its own margin, and nobody in the loop answered it.

The collector seat (score 7/10)

The tape turned a painting I would have scrolled past into four pieces I would bid on, but one in five is still the old show and the words will date it.

Put 13 next to 04 and the argument is over: a decent blob beside a piece I would hang. Which would I own? 01 first, the Proof on the light sheet with the taped-shut reserve; then 05, 06 and 03, the red note on Bone and Ink, the grail trait. 02 is tagged Proof and does not read as one; 08 and 12 are borderless dense pieces where the tape is a rumor. 10 has ON CHAIN and SHIPPED BROKEN written on it, a 2021 sentence on a 2026 wall. A better Pentimento, not a new name; two names would split one collector base.

Concerns:

A trait priced by its name has to look like its name on every piece; Proof does not.
The untaped ~22% is the floor family, 6 to 9 points of it only because a coin came up wrong; the market will price them the way the artist did.
At two words a piece is a painting; at five it is a poster. And nobody frames a hash.

Recommendation: Hold with these changes. Turn the border to 19% borderless; cap deck words at two per piece; drop the strike; ship the tape and tell the platform before it answers. Edition 233. The single most important move: kill the 12/13 coin so the only untaped pieces are the ones with a reason to be.

I would bid on 01, 05, 06 and 03 tomorrow, and I would not bid on anything with a hash or ON CHAIN written on it in two years.

The community seat (score 7/10)

A better wall and a better phone than the mainline, but it is drifting from a painting anyone can hold toward a proof sheet you have to be taught to read.

The artist found this by accident and kept it, the most honest thing in the packet. 03 is the piece I would take first: bone and ink, one red note that lets the whole sheet breathe. The long tail meets a piece on a phone before a wall, and 11 survives one. My worry is the words. STET and a registration cross are the artist's own trade language, which is why they feel true where NUMBER GO UP felt borrowed, but a collector who never sat at a drafting table has to be taught them. It is the same work grown up. Keep the name.

Concerns:

The deck words repeat: BREAK THE GRID on three of twelve sheets. At 377 that is a logo.
The untaped mainline is the orphan register: the cheapest pieces, held by the people with the least money, and their story is "mine lost the toss".
The keyline does not survive the phone, and the market is still writing on the sheet: ON CHAIN on 10, the full hash across 07.

Recommendation: Hold with these changes. Structural only; leave the border dial. Write to the platform now, before the answer: the community forgives "I kept working and it grew" and never forgives a quiet swap. The single most important move: one drafting event per sheet and the market words off the paper.

The proof sheet is for the printer. The painting under the tape is for everyone else, and every piece has both, so let people see both.

The gallery seat (score 7/10)

The tape is the pentimento the mainline only claimed, and it hangs; but the drafting layer speaks in the standup's vocabulary, and the fork reads as a sequel until the name is settled.

I judge this the way a gallery hangs one object in a window for thirty days. 13 is a good painting; 04 is a work on a wall. The taped edge carries the room read; the keyline, two millimeters of pencil at 48 inches, is the second visit: the first time this project has had two viewing distances. 01 is the piece I would put in the window; 12 is the Judd piece, sealed, no drawn line. WONTFIX, REVERT, SHIPPED BROKEN, FOR LOOP: the June board cut the trading floor and the standup walked in. UNDER THIS, START AGAIN, NEGATIVE SPACE will still be true in 2030. The same work, not a sequel: painted complete, then reconsidered under tape, which is what a pentimento is.

Concerns:

BREAK THE GRID, with its arrow, on three of twelve: a slogan telling the viewer what the unclipped stroke already does.
Lapsed at 36% of taped pieces means a third of the taped edition hangs like 08; the register named for lapsing should be the minority.
The buried Signature is not buried on Torn and reserve pieces: 07 letters the full hash across bare sheet.

Recommendation: Hold with these changes. Structural only; turn the border partway, to about a third borderless; keep the printer's marks; edition 233. The single most important move: one pass that cuts the coin and the slogans together, then write to the platform before it writes to you. The animation is the exhibition.

The mainline was the underpainting. The tape is the pentimento. Now hang it where the words have to stay true for thirty days.

The audit lead (score 6/10)

The tape is deterministic, but the guarantee is stated wider than it is tested, and the metadata contradicts the geometry on one taped piece in fourteen.

I ran it. Tape off against the frozen mainline, 153 hashes with no carve-outs: 139 identical, 14 diverged, thirteen the Sheet register and one the kept Cartoon accident. The fork's own tool prints "sheet" for those instead of DIVERGED: a check that could not fail on the axis that matters. Over model byte-stable, 24 of 24 at two sizes; traits from a no-op context, 40 of 40 identical. What I cannot sign: the trait line under 04 reads Sealed, Reserves 2, Loose blocks 2; 8 of 113 taped pieces in my sweep disagree with themselves. "The mainline, taped" is not an honest description.

Concerns:

State the guarantee as "identical except the Sheet coin and the Cartoon lane floor, about 9%" and make the tool count them.
Stream isolation is partial: the fill, the accents and the refusal pull from the main stream, so a future reorder reshuffles every taped piece.
No browser has rendered this fork; parity at the clip edges is unmeasured.

Recommendation: Hold with these changes. Whatever the untaped number, publish it with the carve-outs at 2000 hashes. The dimension label is a span in 89ths, not a Fibonacci number; call it what it is. The single most important move: a release battery the fork can fail. None of this is a determinism fault; all of it is what a collector with a diff tool finds in 2031.

Same hash, same pixels, I can sign; a trait line that says Sealed and Loose 2 on the same piece, I cannot.

The contract seat (score 6/10)

The engine is deployable; the record is not, because the chain will hold one script and the paper on file describes a different one.

I re-derived the numbers rather than take them. The lineage anchor is real: the fork point, the engine on disk and the ledger entry agree on one checksum. The engine grew 44%; the on-chain script goes from 42.7 KB to 60.6 KB. The chain will see a second hash-to-output map on 95% of hashes and a trait table that grew from 18 keys to 27 against a board instruction to shrink it. The fork's own rule says a fork output is not Pentimento without its own name and record, and only a ledger entry can supersede it.

Concerns:

Turbulence, a ninth undocumented key, is residue of a rejected probe that changes nothing on 94% of the edition. Cut it.
The tape switches are mutable globals and the split model is dead code on-chain; any page that embeds the script can flip the canonical output.
Rare classes at 233: about 10 Proofs, 6 hot notes and 11 Sheets; the hot note lands at 3 or fewer one time in eight. At 377: 17, 10 and 18.

Recommendation: Hold with these changes. Keep what a collector can verify (the block measures, the token's own hash) and ration what is derived from nothing (STET, BREAK THE GRID, the registration cross). Tell the platform before any review-build upload, or file the fork as its own project. The single most important move: a decision-ledger entry that names this Pentimento with a lineage block, or names the second work. No packaging until it is written.

The chain will store one script and one story; right now there are two of each, and the one on file is the one nobody wants to ship.