Board review, 2026-09-17 (the tape throughout). pentimento, michael.art

The same eleven seats as 2026-09-15, and the same packet format: what changed, thirteen renders, one per register, each with its before-and-after pair, two filmstrips of the act order, a hundred more on four contact sheets, and eight questions, including what would move the seat to 9. Each seat read its own 09-15 review first. The artist's brief: get to convergence; he would love one or two 9s and a couple of 8s.

Synthesis

Scores, 09-15 to today: the art-historical seat 7 to 8, the technical seat 6 to 8, the minimalism seat 6 to 7, the typography seat 7 to 8, the platform seat 7 to 8, the AI seat 6 to 7, the collector seat 7 to 8, the community seat 7 to 8, the gallery seat 7 to 8, the audit lead 6 to 7, the contract seat 6 to 7. Mean 7.6, up from 6.5. Every seat moved up one point except the technical seat, which moved two. Seven 8s, four 7s, no 9 yet. Three seats say ship-ready after the freeze list (platform, collector, gallery); two say hold, then ship-ready (art-historical, technical); six say hold with changes. Nobody said the work lost anything.

What earned the points: the tape throughout (all eleven: the still records its order, an early block worn by later paint, a late block crisp); the count honest again (all eleven; the audit lead verified it on 946 of 946 painted pieces); the coin dead; the keyline visible on dark grounds; the drafting rationed; the slogan pass done at source; the name and the filing settled the honest way.

The name: closed. Every objection withdrawn or reduced to a condition: the lineage block must say the underpainting is the June engine's method, not its bytes (art-historical, technical); the statement must say the frozen mainline is the underpainting (platform); the build digest goes in the day it exists (contract). The minimalism and AI seats keep a footnote, masking is a schedule and not yet a change of mind, and accept the block.

The pentimento question: unanimous. The tape as a running decision is the pentimento, not a costume. The one dissent inside the yes is the AI seat's, verified in the code: the scrape-back is planned before any paint is laid, never on a crossing that happened. On the canvas a correction; in the loop a coin.

Strongest concerns, by how many seats named them:

The measure that only says 1 (all eleven): the tape's measures count lattice cells and a block is usually one cell, while the excavation's still counts in 89ths. Letter the numeral only above one cell; one unit everywhere.
The Ledger surfacing off the torn and reserve pieces (eight seats): the full hash across 05, Strict with no reserve. The 09-15 rule was not built.
The standup vocabulary survived the slogan pass (six seats). The collector seat's rule: if it could be a commit message, it does not go on the painting. The AI seat would keep the words that name something the program did.
Reserves and proofs arrive lettered (five seats): the reserve is the only empty ground left, so the deck lands in it. Nothing inside a reserve but its measure.
The Event trait is not true to the sheet (the audit lead: Caption with nothing drawn on 88 of 844). Emit Event from what was drawn.
The tape's share of the count should follow density (eight seats): low at low counts so the hand survives, high only at Teeming and Swarm.
Two withholdings on one sheet: a window and a reserve coincide on about 10% of taped pieces; windows on sparse pieces open on nothing (three seats each).
Proof at 7 to 9% of taped is a register, not a surprise: the platform and gallery seats want it back to 1/13; the collector and community seats would keep the artist's rate and stop calling it rare.
Ending on paint (gallery, audit lead): every taped piece's last act is a loose run, so at Swarm no block is the crisp control the worn ones need.
The record's claims (audit lead, contract seat): "only their timing changes" is false past the first block, because the marks inside blocks still draw from the main stream; 23 trait keys, not 21; Inks Full on every untaped piece; the on-chain script past one compressed chunk; no browser render; a battery that cannot fail.

What would move each seat to 9:

The order made honest: the pencil drawn at each block's own act (typography); the scrape-back fired only where the surface memory changed (AI); end on tape (gallery); the low-count floor (technical).
The tape reading the painting: the next block chosen from the surface memory (art-historical, technical). The contract seat: it ends stream isolation, so last before the build; the gallery seat: the 10, and the next work's.
The type pass: the count as the only line of type, the hash on torn and reserve pieces only, the proof unlettered (minimalism, collector, community).
The release: the true-order replay as the token's live view, settled frame equal to the still (platform); tape-off under a thumb in the live view (community); a release candidate in one chunk, then the letter (contract); a battery that can fail (audit lead).

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

The type pass: standup words out; the Ledger on torn and reserve pieces only; nothing in a reserve or proof but its measure; the numeral only above one cell; Sealed renamed under one trait; Inks None on untaped pieces.
Event true to the sheet, and the battery a gate: fixed hash list, exit code, trait assertions.
The order made honest: the pencil in the order, the scrape-back conditioned on the surface memory, end on tape, the tape's share by density, window and reserve exclusive.
Then the tape reading the painting, as an A/B, last before the build.
The release list: knobs deleted, the build in one chunk or two accepted, a features script, three browsers, the replay in the token, the full digests in the lineage block, the statement rewrite, the letter to the platform.

Bottom line: two days ago the board said keep the rule, cut the costume, and write down what this is. All three happened, and the seats moved a point each for it. What holds every seat at 7 or 8 is a small set of leftovers, most of them lettering and bookkeeping, plus one idea they now agree on from four directions: the order honest all the way down, the pencil aging with the tape, the knife answering a real crossing, the piece ending on a clean block, and, when the freeze can bear it, the tape choosing where to go by looking. Nobody asked for a redesign.

The art-historical seat (09-15: 7/10; today: 8/10)

The tape throughout turns the pentimento from a caption into a structure, and the still now records its own order; the costume that remains is the writing, and a ruler that only says one.

Put the two filmstrips beside 04 and 13 and the fork has changed category. Two days ago the tape was a finish, its precedent Newman's zip: a mask that leaves evidence, applied once. In the first strip block 1 goes down early, twenty-three loose strokes cross it, and block 8 arrives crisp; the still reads in that order without the strip. The precedent moved with it, to Hofmann, to Richter's scrape, the second thought made structural. Where the tape did not go, the costume remains: 10 is still the standup's wall in the edition's largest type.

Concerns:

The ruler always reads 1, because a block is one lattice cell; a measurement that can only say one is a monogram.
The old numbers survive: 04 carries 37 beside the ruler's 1; 10 writes 34 STROKES on a sheet whose trait says 5.
11 is labeled Event: Caption with no caption visible; the Ledger runs across the waist of 05.

Recommendation: Hold with these changes, then ship-ready after the freeze list. No objection to the name, with one line in the lineage block: the underpainting is the June engine's method, not its bytes. Most important move: make the tape choose its next block from the surface memory, so the running decision is a decision and not a schedule.

A mask that leaves evidence was new to this field two days ago; a mask that arrives mid-painting and lets the painting record it is a pentimento in the only sense a historian accepts.

The technical seat (09-15: 6/10; today: 8/10)

One rule, applied at moments instead of at the end, and the count is honest again; what is left is a numeral, a ledger, and a tape that follows a schedule instead of reading the painting.

The finals are one count, the mainline's own: the taped blocks take a share apportioned by area and whatever does not fire returns as a last short run. The alternation is the mask I accepted two days ago, applied at moments dealt on the tape stream; the scrape-back is the same mask twice on the same block. One rule. 06 proves it: the navy block, taped early, wears cyan and orange laid after it; the orange block, taped late, is crisp. And the tape still takes its blocks in plan order: a schedule, not a decision.

Concerns:

Window and reserve coincide on 13 of 128 taped pieces in my sweep: two mechanisms saying "under here" on one sheet.
At 13 strokes or fewer the loose layer is 2 or 3 marks; there the mainline's hand is not thinner, it is gone.
23 trait keys, not 21, and nine constants serve one rule with only the count defended by a battery fact.

Recommendation: Hold with these changes, then ship-ready after the freeze list. The objection to the name is withdrawn: a block taped early and crossed later is a change of mind the still records; the lineage block must say tape-off is no longer the mainline. The single move: the low-count floor, the one place the rule breaks at scale.

Two days ago the mask was a plan. Today it is a schedule. One more move and it is a judgment.

The minimalism seat (09-15: 6/10; today: 7/10)

The tape throughout turned a finish into a system; the sheet still wears the words, and you sell the sheet.

Two days ago the tape was a frame. Now it is a decision the painter keeps making: block 1 goes down first, takes 23 strokes across its face, and by the settled frame it is the oldest thing on the sheet; block 8 arrives after 37 loose and stays crisp. The scrape-back is the first mark in this project that is a second thought rather than the costume of one. The system is a nine. The sheet is not: 04 carries four phrases and two measures, 05 letters the hash across the middle, 10 has five strokes and nineteen words. Tape, paint, lift, repeat. Repeat is the loop, and the loop is the concept.

Concerns:

The Ledger is back: the full hash on 05 and on 4 of 100, on Saturated and Teeming pieces now, not only Torn.
The proof, my primitive, arrives lettered; the two clean ones are the best reserves in the hundred.
21 trait keys; the manifest prints eleven. Seven would do: Palette, Density, Tape, Reserves, Excavation, Fate, Strokes.

Recommendation: Hold with these changes. Masking is not revision, and one name now carries two rules; the lineage block is the honest fix and I will take it. The single move: the type pass, the count as the only line of type, the hash and the one-cell measures gone, the proof unlettered. Not tape-off in the live view, a switch is the coin as a button; not the tape reading the painting, a rule the viewer cannot see is a rumor.

Tape. Paint. Lift. Repeat. The count is the only word the sheet needs.

The typography seat (09-15: 7/10; today: 8/10)

The lettering on top now survives being checked, except on one point: the tape has an age and the pencil does not, because every keyline, bracket and measure is drawn after the last act.

The strike is one grammar now, STET always beside it; the keyline is a drawn line, white on the charcoal of 06. The best mark in the packet is not an event: in 03, WAS HERE FIRST goes half under a block laid after it and only FIRST survives. What moved the score is the ration and the honest count; what holds it is timing. The blocks wear their age, but the pencil is drawn once, at the finish, so an early block worn by later paint carries a crisp line that says late.

Concerns:

Two units on one sheet: the tape's measures count lattice cells, the excavation's the old span, so 04 carries 37 and 1.
Event: Caption with no caption; the trait is set before the refusal is gated or the strike placed.
The press marks in the painter's pencil, N STROKES and a hash fragment on every sheet; a press stamps its job number small, in its own ink.

Recommendation: Hold with these changes. No objection to the name: the lineage block is a colophon, where a printer records what changed and when. The single most important move: the pencil in the order, each block's keyline, brackets and measure drawn at the block's own act, so later runs cross an early block's line and a late block's line stays crisp.

The tape has an age now. Give the pencil one, or the lettering stays a finish on a work whose whole claim is order.

The platform seat (09-15: 7/10; today: 8/10)

It holds as Pentimento now, because a block that buries a run of paint is the word's own definition; what keeps it from 9 is that the still hides the act.

The name and the filing came the honest way, so I move a point. The second strip changed my mind: five loose strokes lay a black mass at the left, block 1 tapes over it and fills in bone, and by the next frame the black is gone under paint that came later. What the still does not give me is the second strip: in the settled frame the buried strokes are invisible unless a window opens on them, so the thing the name means lives in the replay. That is about the release, not the engine.

Concerns:

Proof at 7% is 26 pieces at 377, twice what 1/13 gave. The artist may have it, but the page cannot then call Proof rare.
The genre words stayed when the market words left: GENERATIVE, CODE-BASED ART, FLOW FIELD. A piece that names its genre dates fastest.
No browser has rendered the tape; no rate goes on a page before N of 2000.

Recommendation: Ship-ready after the freeze list, with the numeral fixed. I asked for a second name; the tape throughout changed my mind, on one condition: the statement says plainly that the frozen mainline is the underpainting. The single most important move: the true-order replay as the token's live view, settled frame equal to the still in three browsers, before any review-build upload.

A pentimento you can only see in the replay is still a pentimento; just make sure the replay is the token, not the trailer.

The AI seat (09-15: 6/10; today: 7/10)

The order is real now and the authorship is honest, which earns the point; the loop I asked for is still not built, and the painter's second thought is rolled before the first.

Two days ago I said the tape never looks at the painting it covers. Today, in 04, the top blue block wears pink and yellow that came after its lift, and the still tells you which was first. The authorship moved too: round 5 laundered his taste through three agents fitting rates to his picks; round 7 is the artist saying "I am the artist" on the record. But I read the code. The scrape-back is planned before any paint is down, never on whether a stroke crossed it. On the canvas a correction; in the loop a coin flipped before there was anything to correct.

Concerns:

The scrape-back is pre-rolled: a block nothing crossed can be knifed back, while the surface memory records every crossing and nothing reads it.
The block choice is unchanged: 08's loose blocks still fall where the coin fell.
The top of the tape dial buries the hand: 04 against 13 is the same 233 with the free strokes gone into fills.

Recommendation: Hold with these changes. On the name: a block worn by what came after it is a pentimento in the literal sense; what remains is that a pentimento is a change of mind and here the mind is a schedule. Keep the deck words that name something this program did (RAND(), WHILE(TRUE), OVERFIT). The single most important move: condition the scrape-back on the surface memory, so the knife answers a crossing that happened.

The second thought is rolled before the first; give the knife a reason and the tape finally looks.

The collector seat (09-15: 7/10; today: 8/10)

The coin is dead and the grid finally reads as one show; I would bid on five of twelve today and three in two years, and the gap between those numbers is the word deck.

Put the four hundred-sheets up and squint: eighty-odd taped patchworks, seven blueprints, seven Abandoned, and it reads as one project. That is the coin dying. My order today: 08, 06 (the white keyline), 03 (the red note, still the grail), 01, 04. In two years: 08, 06, 03, and 84 from the hundred. The ones that hold are lettered with a paint number and a measure; the ones I pass on are lettered with a git log. And the hash came back, across the middle of 05.

Concerns:

The Ledger rule from 09-15 was not built; the full hash is lettered on Strict and Sealed pieces. Nobody frames a hash.
Reserves as caption boxes: a hex code, a deck word and a measure inside the only empty ground left.
Sealed is a value under both Excavation and Tape: one word meaning two things on a marketplace filter.

Recommendation: Ship-ready after the freeze list, with the deck pass riding on it: the Ledger rule enforced, the standup words out, nothing inside a reserve but a measure. On the artist's rate: twenty-six Proofs in 377 is a register with a floor, better for a two-year hold than nine unicorns; keep the rate, keep the reserves empty. Put the lineage block on the project page's first screen.

If it could be a commit message it does not go on the painting; I would bid on 08, 06, 03 and 84 tomorrow, and on the same four in 2028.

The community seat (09-15: 7/10; today: 8/10)

It carries a first-time collector on a phone now, and the only lesson left on the paper is the one the standup wrote.

The artist did the thing: he took the cut list whole, then said he is the artist and chose the scraped-off pieces, and those are the same kind of honest. The market words are gone from all 112 tiles; the keyline shows at 300px and is white on dark grounds. And the alternation reads without anyone explaining it: on 12 the pink block has paint running across its edge and the block with the 13 in it is crisp. That is a pentimento a person can see, not one they have to be told.

Concerns:

The dev vocabulary survived the slogan pass: REFACTOR, DEPRECATED, WHILE(TRUE), REVERT, MEMORY LEAK. The trading floor left; the standup stayed.
The sparse floor carries the verdicts, DEPRECATED at 21 strokes, GOOD ENOUGH at 2: the cheapest pieces, held by the smallest collectors, and the sheet tells them so.
The scrape-back reads as dust on pale grounds, one tile in ten. Keep the rate, guard the palette.

Recommendation: Hold with these changes. The single most important move: the standup words off the paper, the same pass the market words got. Then the touch that lifts the tape: press and hold in the live view and the tape lifts, let go and it settles back; every mint is two paintings, and a thumb is the only lesson. And a flat mint price, so the sparse floor is a floor and not a punishment.

The market left the paper and the painting can carry a stranger now. Let the standup go home too, and put the second painting under a thumb.

The gallery seat (09-15: 7/10; today: 8/10)

The still now carries the order, so the wall no longer needs the animator to make its case; what it costs is the hard edge on the densest pieces, and the deck still speaks standup.

I judge this the way a gallery hangs one object in a window for thirty days. On 09-15 the second viewing distance was a pencil line; today the keyline belongs to the first distance, and the second is time. Put 03 in the window: the cream streaks crossing the black blocks are the early tape, the crisp black blocks are the late tape, and a viewer with no wall text can reconstruct the afternoon. That is what a pentimento is, evidence of an earlier state held in the surface, and it is in the still now, not the replay.

Concerns:

Swarm loses the wall read: the engine ends every taped piece on paint, so no block is the control the worn ones need.
The standup stayed when the trading floor left: SUNK COST at display size over TECH DEBT on 10.
07: at 21 strokes the torn blocks print as grey marquees with corner brackets on black.

Recommendation: Ship-ready after the freeze list, with one rule added: end on tape, the count-landing run before the last block, so every piece keeps one crisp block for the worn ones to be read against. On the artist's rate: Judd made the same box for thirty years, so a house style is not the charge; the charge is three whites a wall cannot tell apart. Take the proof back to 1/13. The tape reading the painting belongs to the next work.

The still carries the order now. The animator is the label; the wall is the exhibition again.

The audit lead (09-15: 6/10; today: 7/10)

The count is exact, but the Event trait certifies a caption on one taped piece in ten that carries none, and the tape pass still draws from the main stream.

I ran a thousand hashes: every painted piece lays exactly the Fibonacci number it drew, 946 of 946, equality with the target, not membership in the sequence. What holds it at seven: 11 is the packet's own Caption exhibit and nothing is drawn on it; Event and the refusal are two independent coins, so 88 of 128 Caption pieces show no caption. And the angle helper draws from the main stream inside every block: on 04 the first loose mark that differs between tape on and tape off is where block 1 begins, so "only their timing changes" is false from the first block onward.

Concerns:

Event lies: Caption with no caption on 10.4% of taped pieces, Dims with no measure on 2.8%. Derive Event from what was drawn.
The count-landing top-up lands after the scrape-back and outside the late cap, so a knifed-back block can be crossed again.
The battery is a report, not a gate: unseeded hashes, no exit code, no trait assertions.

Recommendation: Hold with these changes. The lineage block should say the tape reshuffles the finals from the first block, not "the mainline, taped". The tape reading the painting is deterministic if it reads the surface memory; I would sign that. The single most important move: Event true, and the battery a gate that can fail, before the freeze tag.

A trait is a promise about the sheet; Caption on 11 promises a caption that is not there, and I can count 88 more.

The contract seat (09-15: 6/10; today: 7/10)

The record is now right and survives audit line by line; the script the chain will hold does not exist yet, and the one I can build from this commit no longer fits one compressed chunk.

I audited the lineage block against the repository. Every checkable claim holds: the fork-point tag carries the frozen engine's bytes, and the eight streams added are exactly the eight the block names. The count is honest: 593 of 593 painted pieces in my 600-hash battery lay a Fibonacci number. Every ask this seat made on 09-15 was done. What is not done is the deliverable. The on-chain script is 62,918 bytes, gzip 25,044: two compressed chunks, about 500 bytes past the single chunk I said had 500 bytes of headroom.

Concerns:

Size: trim the retired order branch and the knobs before the build, or accept two compressed chunks and the deploy gas.
Inks reports Full on every untaped piece; the trait table is 23 keys, not 21.
Rare classes at 377: the hot note 12 give or take 3; Gesture has an 8% chance of not existing.

Recommendation: Hold with these changes. The block names a work whose on-chain bytes do not exist; add the build digest the day they do, with edition, price and a license. The tape reading the painting ends stream isolation: the last change before the build, never a dial. The single most important move: build the release candidate now, knobs deleted, Inks fixed, one compressed chunk; hash it, write the hash into the lineage block, then send the letter.

The record finally says one thing; the chain still has nothing to say it about.