4 pieces in 1 set · docade-social@4. Everything has a verdict.
64 attempts replaced or turned down.
| Piece | Set | State | Verdict | Model | Why | When |
|---|---|---|---|---|---|---|
| h2h-week-result | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week-result | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-complete | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-complete | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-bar | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-bar | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week-result | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week-result | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-complete | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-complete | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-bar | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-bar | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week-result | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week-result | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-complete | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-complete | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-bar | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-bar | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week-result | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week-result | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-complete | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-complete | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-bar | — | superseded | — | authored-svg | 2026-08-26 | |
| coop-bar | — | superseded | — | authored-svg | 2026-08-26 | |
| h2h-week-result | — | superseded | — | authored-svg | 2026-08-25 | |
| h2h-week-result | — | superseded | — | authored-svg | 2026-08-25 | |
| coop-complete | — | superseded | — | authored-svg | 2026-08-25 | |
| coop-complete | — | superseded | — | authored-svg | 2026-08-25 | |
| h2h-week | — | superseded | — | authored-svg | 2026-08-25 | |
| h2h-week | — | superseded | — | authored-svg | 2026-08-25 | |
| coop-bar | — | superseded | — | authored-svg | 2026-08-25 | |
| coop-bar | — | superseded | — | authored-svg | 2026-08-25 | |
| h2h-week-result | — | superseded | revise human | flux2-dev | @3 is worse, and the reason ends the diffusion attempt. Banning text in prose made text MORE likely: the model transcribed the ban list itself into the picture — 'BOTH halves finished', 'Writes the real Names', 'Token #7c5...' — set as headlines across the panel. Three passes, three different failures, and the compositions were correct on all three. Wrong tool class. CLAUDE.md already says diagrams with real text are authored, not generated; the inverse binds just as hard. These four are DIAGRAMS — one bar with two contiguous segments, two mirrored halves, a disc measured at 9% — where the requirements are exact symmetry, exact geometry and exactly zero glyphs. That is a drawing problem, not a diffusion problem, and no amount of prompt pressure makes a diffusion model draw a guaranteed-empty band. Lane moved to capability diagram, routed to authored-svg. | 2026-08-25 |
| h2h-week-result | — | superseded | revise human | flux2-dev | @3 is worse, and the reason ends the diffusion attempt. Banning text in prose made text MORE likely: the model transcribed the ban list itself into the picture — 'BOTH halves finished', 'Writes the real Names', 'Token #7c5...' — set as headlines across the panel. Three passes, three different failures, and the compositions were correct on all three. Wrong tool class. CLAUDE.md already says diagrams with real text are authored, not generated; the inverse binds just as hard. These four are DIAGRAMS — one bar with two contiguous segments, two mirrored halves, a disc measured at 9% — where the requirements are exact symmetry, exact geometry and exactly zero glyphs. That is a drawing problem, not a diffusion problem, and no amount of prompt pressure makes a diffusion model draw a guaranteed-empty band. Lane moved to capability diagram, routed to authored-svg. | 2026-08-25 |
| h2h-week | — | superseded | revise human | flux2-dev | @3 is worse, and the reason ends the diffusion attempt. Banning text in prose made text MORE likely: the model transcribed the ban list itself into the picture — 'BOTH halves finished', 'Writes the real Names', 'Token #7c5...' — set as headlines across the panel. Three passes, three different failures, and the compositions were correct on all three. Wrong tool class. CLAUDE.md already says diagrams with real text are authored, not generated; the inverse binds just as hard. These four are DIAGRAMS — one bar with two contiguous segments, two mirrored halves, a disc measured at 9% — where the requirements are exact symmetry, exact geometry and exactly zero glyphs. That is a drawing problem, not a diffusion problem, and no amount of prompt pressure makes a diffusion model draw a guaranteed-empty band. Lane moved to capability diagram, routed to authored-svg. | 2026-08-25 |
| h2h-week | — | superseded | revise human | flux2-dev | @3 is worse, and the reason ends the diffusion attempt. Banning text in prose made text MORE likely: the model transcribed the ban list itself into the picture — 'BOTH halves finished', 'Writes the real Names', 'Token #7c5...' — set as headlines across the panel. Three passes, three different failures, and the compositions were correct on all three. Wrong tool class. CLAUDE.md already says diagrams with real text are authored, not generated; the inverse binds just as hard. These four are DIAGRAMS — one bar with two contiguous segments, two mirrored halves, a disc measured at 9% — where the requirements are exact symmetry, exact geometry and exactly zero glyphs. That is a drawing problem, not a diffusion problem, and no amount of prompt pressure makes a diffusion model draw a guaranteed-empty band. Lane moved to capability diagram, routed to authored-svg. | 2026-08-25 |
| coop-complete | — | superseded | revise human | flux2-dev | @3 is worse, and the reason ends the diffusion attempt. Banning text in prose made text MORE likely: the model transcribed the ban list itself into the picture — 'BOTH halves finished', 'Writes the real Names', 'Token #7c5...' — set as headlines across the panel. Three passes, three different failures, and the compositions were correct on all three. Wrong tool class. CLAUDE.md already says diagrams with real text are authored, not generated; the inverse binds just as hard. These four are DIAGRAMS — one bar with two contiguous segments, two mirrored halves, a disc measured at 9% — where the requirements are exact symmetry, exact geometry and exactly zero glyphs. That is a drawing problem, not a diffusion problem, and no amount of prompt pressure makes a diffusion model draw a guaranteed-empty band. Lane moved to capability diagram, routed to authored-svg. | 2026-08-25 |
| coop-complete | — | superseded | revise human | flux2-dev | @3 is worse, and the reason ends the diffusion attempt. Banning text in prose made text MORE likely: the model transcribed the ban list itself into the picture — 'BOTH halves finished', 'Writes the real Names', 'Token #7c5...' — set as headlines across the panel. Three passes, three different failures, and the compositions were correct on all three. Wrong tool class. CLAUDE.md already says diagrams with real text are authored, not generated; the inverse binds just as hard. These four are DIAGRAMS — one bar with two contiguous segments, two mirrored halves, a disc measured at 9% — where the requirements are exact symmetry, exact geometry and exactly zero glyphs. That is a drawing problem, not a diffusion problem, and no amount of prompt pressure makes a diffusion model draw a guaranteed-empty band. Lane moved to capability diagram, routed to authored-svg. | 2026-08-25 |
| coop-bar | — | superseded | revise human | flux2-dev | @3 is worse, and the reason ends the diffusion attempt. Banning text in prose made text MORE likely: the model transcribed the ban list itself into the picture — 'BOTH halves finished', 'Writes the real Names', 'Token #7c5...' — set as headlines across the panel. Three passes, three different failures, and the compositions were correct on all three. Wrong tool class. CLAUDE.md already says diagrams with real text are authored, not generated; the inverse binds just as hard. These four are DIAGRAMS — one bar with two contiguous segments, two mirrored halves, a disc measured at 9% — where the requirements are exact symmetry, exact geometry and exactly zero glyphs. That is a drawing problem, not a diffusion problem, and no amount of prompt pressure makes a diffusion model draw a guaranteed-empty band. Lane moved to capability diagram, routed to authored-svg. | 2026-08-25 |
| coop-bar | — | superseded | revise human | flux2-dev | @3 is worse, and the reason ends the diffusion attempt. Banning text in prose made text MORE likely: the model transcribed the ban list itself into the picture — 'BOTH halves finished', 'Writes the real Names', 'Token #7c5...' — set as headlines across the panel. Three passes, three different failures, and the compositions were correct on all three. Wrong tool class. CLAUDE.md already says diagrams with real text are authored, not generated; the inverse binds just as hard. These four are DIAGRAMS — one bar with two contiguous segments, two mirrored halves, a disc measured at 9% — where the requirements are exact symmetry, exact geometry and exactly zero glyphs. That is a drawing problem, not a diffusion problem, and no amount of prompt pressure makes a diffusion model draw a guaranteed-empty band. Lane moved to capability diagram, routed to authored-svg. | 2026-08-25 |
| coop-bar | — | superseded | revise human | flux2-dev | @2 fixed both faults it was written for: the markers are plain flat discs with nothing inside them, no map pin anywhere, and no text at all — the name bands came back correctly EMPTY. h2h-week-result holds its whole social contract: mirrored halves, identical confetti, identical completed marker, a checkered flag spanning both, no winner. Traded for a new fault of the same family as the room walls: the composition rendered as a dark CARD floating on white, so the cutout had nothing flat to key (h2h-week came back 1.7% opaque, h2h-week-result kept a rectangle of ground). Cause: nothing said the shapes sit directly on the ground, so 'a panel in docade's shared household surface' was read as an object called a panel. @3 bans the card by name — this art goes INSIDE docade's card, so a card here is a card inside a card. | 2026-08-25 |
| coop-bar | — | superseded | revise human | flux2-dev | @2 fixed both faults it was written for: the markers are plain flat discs with nothing inside them, no map pin anywhere, and no text at all — the name bands came back correctly EMPTY. h2h-week-result holds its whole social contract: mirrored halves, identical confetti, identical completed marker, a checkered flag spanning both, no winner. Traded for a new fault of the same family as the room walls: the composition rendered as a dark CARD floating on white, so the cutout had nothing flat to key (h2h-week came back 1.7% opaque, h2h-week-result kept a rectangle of ground). Cause: nothing said the shapes sit directly on the ground, so 'a panel in docade's shared household surface' was read as an object called a panel. @3 bans the card by name — this art goes INSIDE docade's card, so a card here is a card inside a card. | 2026-08-25 |
| h2h-week | — | superseded | — | flux2-dev | 2026-08-25 | |
| h2h-week | — | superseded | — | flux2-dev | 2026-08-25 | |
| h2h-week-result | — | superseded | — | flux2-dev | 2026-08-25 | |
| h2h-week-result | — | superseded | — | flux2-dev | 2026-08-25 | |
| coop-complete | — | superseded | revise human | flux2-dev | @2 fixed both faults it was written for: the markers are plain flat discs with nothing inside them, no map pin anywhere, and no text at all — the name bands came back correctly EMPTY. h2h-week-result holds its whole social contract: mirrored halves, identical confetti, identical completed marker, a checkered flag spanning both, no winner. Traded for a new fault of the same family as the room walls: the composition rendered as a dark CARD floating on white, so the cutout had nothing flat to key (h2h-week came back 1.7% opaque, h2h-week-result kept a rectangle of ground). Cause: nothing said the shapes sit directly on the ground, so 'a panel in docade's shared household surface' was read as an object called a panel. @3 bans the card by name — this art goes INSIDE docade's card, so a card here is a card inside a card. | 2026-08-25 |
| coop-complete | — | superseded | revise human | flux2-dev | @2 fixed both faults it was written for: the markers are plain flat discs with nothing inside them, no map pin anywhere, and no text at all — the name bands came back correctly EMPTY. h2h-week-result holds its whole social contract: mirrored halves, identical confetti, identical completed marker, a checkered flag spanning both, no winner. Traded for a new fault of the same family as the room walls: the composition rendered as a dark CARD floating on white, so the cutout had nothing flat to key (h2h-week came back 1.7% opaque, h2h-week-result kept a rectangle of ground). Cause: nothing said the shapes sit directly on the ground, so 'a panel in docade's shared household surface' was read as an object called a panel. @3 bans the card by name — this art goes INSIDE docade's card, so a card here is a card inside a card. | 2026-08-25 |
| h2h-week | — | superseded | revise human | flux2-dev | @2 fixed both faults it was written for: the markers are plain flat discs with nothing inside them, no map pin anywhere, and no text at all — the name bands came back correctly EMPTY. h2h-week-result holds its whole social contract: mirrored halves, identical confetti, identical completed marker, a checkered flag spanning both, no winner. Traded for a new fault of the same family as the room walls: the composition rendered as a dark CARD floating on white, so the cutout had nothing flat to key (h2h-week came back 1.7% opaque, h2h-week-result kept a rectangle of ground). Cause: nothing said the shapes sit directly on the ground, so 'a panel in docade's shared household surface' was read as an object called a panel. @3 bans the card by name — this art goes INSIDE docade's card, so a card here is a card inside a card. | 2026-08-25 |
| h2h-week | — | superseded | revise human | flux2-dev | @2 fixed both faults it was written for: the markers are plain flat discs with nothing inside them, no map pin anywhere, and no text at all — the name bands came back correctly EMPTY. h2h-week-result holds its whole social contract: mirrored halves, identical confetti, identical completed marker, a checkered flag spanning both, no winner. Traded for a new fault of the same family as the room walls: the composition rendered as a dark CARD floating on white, so the cutout had nothing flat to key (h2h-week came back 1.7% opaque, h2h-week-result kept a rectangle of ground). Cause: nothing said the shapes sit directly on the ground, so 'a panel in docade's shared household surface' was read as an object called a panel. @3 bans the card by name — this art goes INSIDE docade's card, so a card here is a card inside a card. | 2026-08-25 |
| h2h-week-result | — | superseded | revise human | flux2-dev | @2 fixed both faults it was written for: the markers are plain flat discs with nothing inside them, no map pin anywhere, and no text at all — the name bands came back correctly EMPTY. h2h-week-result holds its whole social contract: mirrored halves, identical confetti, identical completed marker, a checkered flag spanning both, no winner. Traded for a new fault of the same family as the room walls: the composition rendered as a dark CARD floating on white, so the cutout had nothing flat to key (h2h-week came back 1.7% opaque, h2h-week-result kept a rectangle of ground). Cause: nothing said the shapes sit directly on the ground, so 'a panel in docade's shared household surface' was read as an object called a panel. @3 bans the card by name — this art goes INSIDE docade's card, so a card here is a card inside a card. | 2026-08-25 |
| h2h-week-result | — | superseded | revise human | flux2-dev | @2 fixed both faults it was written for: the markers are plain flat discs with nothing inside them, no map pin anywhere, and no text at all — the name bands came back correctly EMPTY. h2h-week-result holds its whole social contract: mirrored halves, identical confetti, identical completed marker, a checkered flag spanning both, no winner. Traded for a new fault of the same family as the room walls: the composition rendered as a dark CARD floating on white, so the cutout had nothing flat to key (h2h-week came back 1.7% opaque, h2h-week-result kept a rectangle of ground). Cause: nothing said the shapes sit directly on the ground, so 'a panel in docade's shared household surface' was read as an object called a panel. @3 bans the card by name — this art goes INSIDE docade's card, so a card here is a card inside a card. | 2026-08-25 |
| coop-complete | — | superseded | revise human | flux2-dev | The compositions are right — coop-bar built ONE bar with two contiguous segments and no divider, coop-complete spread its confetti over the whole bar, and h2h-week-result nailed the hard one: mirrored halves, identical celebration on both, a checkered flag across the full width, nothing that reads as a winner or a loser. The declared per-piece compositions did their job. Two failures, both from the same cause — 'an ABSTRACT MARKER' was left open, so the model invented one. (1) It chose a MAP PIN, which means place, not person; then filled the pin with a head-and-shoulders PERSON GLYPH on coop-complete, which is a silhouette of a person and the one thing this lane bans outright. (2) h2h-week drew TEXT into the name bands — 'Head beantre', 'Hreak Bheatnes' — garbled, and in the exact bands the app writes real names into. Also the whole set renders as soft-bevelled UI widgets rather than flat shapes. Fix in @2: the marker is a plain flat filled disc, stated as a measurement, with nothing inside it and no tail. | 2026-08-25 |
| coop-complete | — | superseded | revise human | flux2-dev | The compositions are right — coop-bar built ONE bar with two contiguous segments and no divider, coop-complete spread its confetti over the whole bar, and h2h-week-result nailed the hard one: mirrored halves, identical celebration on both, a checkered flag across the full width, nothing that reads as a winner or a loser. The declared per-piece compositions did their job. Two failures, both from the same cause — 'an ABSTRACT MARKER' was left open, so the model invented one. (1) It chose a MAP PIN, which means place, not person; then filled the pin with a head-and-shoulders PERSON GLYPH on coop-complete, which is a silhouette of a person and the one thing this lane bans outright. (2) h2h-week drew TEXT into the name bands — 'Head beantre', 'Hreak Bheatnes' — garbled, and in the exact bands the app writes real names into. Also the whole set renders as soft-bevelled UI widgets rather than flat shapes. Fix in @2: the marker is a plain flat filled disc, stated as a measurement, with nothing inside it and no tail. | 2026-08-25 |
| coop-bar | — | superseded | revise human | flux2-dev | The compositions are right — coop-bar built ONE bar with two contiguous segments and no divider, coop-complete spread its confetti over the whole bar, and h2h-week-result nailed the hard one: mirrored halves, identical celebration on both, a checkered flag across the full width, nothing that reads as a winner or a loser. The declared per-piece compositions did their job. Two failures, both from the same cause — 'an ABSTRACT MARKER' was left open, so the model invented one. (1) It chose a MAP PIN, which means place, not person; then filled the pin with a head-and-shoulders PERSON GLYPH on coop-complete, which is a silhouette of a person and the one thing this lane bans outright. (2) h2h-week drew TEXT into the name bands — 'Head beantre', 'Hreak Bheatnes' — garbled, and in the exact bands the app writes real names into. Also the whole set renders as soft-bevelled UI widgets rather than flat shapes. Fix in @2: the marker is a plain flat filled disc, stated as a measurement, with nothing inside it and no tail. | 2026-08-25 |
| coop-bar | — | superseded | revise human | flux2-dev | The compositions are right — coop-bar built ONE bar with two contiguous segments and no divider, coop-complete spread its confetti over the whole bar, and h2h-week-result nailed the hard one: mirrored halves, identical celebration on both, a checkered flag across the full width, nothing that reads as a winner or a loser. The declared per-piece compositions did their job. Two failures, both from the same cause — 'an ABSTRACT MARKER' was left open, so the model invented one. (1) It chose a MAP PIN, which means place, not person; then filled the pin with a head-and-shoulders PERSON GLYPH on coop-complete, which is a silhouette of a person and the one thing this lane bans outright. (2) h2h-week drew TEXT into the name bands — 'Head beantre', 'Hreak Bheatnes' — garbled, and in the exact bands the app writes real names into. Also the whole set renders as soft-bevelled UI widgets rather than flat shapes. Fix in @2: the marker is a plain flat filled disc, stated as a measurement, with nothing inside it and no tail. | 2026-08-25 |