{"session": "8a4bb842-c367-4600-821c-aa2d0e4e0b8c", "env": "main-signor-bernini", "command": "git add -A && git commit -qm \"The sweep run to convergence, five more passes until a whole pass gained less than half a point, the record saved after every piece: colour kind butler 0.794, coach 0.730, colleague 0.793, homie 0.718, squire 0.644\" && git log --oneline -1 && journal helper report \"Alfred, the sweep has converged. Five more passes ran back to back, and the last gained under 0.005 across all five voices together. The record is saved after every piece, so an interrupted sweep loses one piece at most.\n\n**Colour kind, then silhouette; every screen within 0.006 of its drawn size** (committed as $(git rev-parse --short HEAD)):\n\n| Voice | Colour | Silhouette |\n|---|---|---|\n| squire | 0.644 (from 0.527 before any sweep) | 0.881 |\n| homie | 0.718 (from 0.593) | 0.933 |\n| coach | 0.730 (from 0.605) | 0.929 |\n| butler | 0.794 (from 0.666) | 0.913 |\n| colleague | 0.793 (from 0.711) | 0.963 |\n\n**Two faults found and fixed at the source:**\n- **Pieces crowding a screen.** The early sweeps had grown the butler's hat-tipping hand and his hat over the edge of his screen, and moved the colleague's frames off her lenses. Those fits were cleared, and the sweep now refuses any change that takes a screen further from its drawn size.\n- **A misleading height correction.** Correcting screens by the compare page's figure-height estimate had made them worse. That estimate isn't what a screen's fraction is measured against, so the screens stay at the heights they were set for.\n\n**Next, step 2:** widening what a fit may move, one freedom at a time. The first is stretch: a piece's own width, length and depth sized separately, where today it has one size. Then I'll sweep it, and keep it only if it pays.\n\nPreview: http://127.0.0.1:8431/toon-preview/index.html\nOverlays: http://127.0.0.1:8431/toon-preview/rest/overlays.png\" 2>&1 | tail -1", "parts": ["git add -A", "git commit -qm \"The sweep run to convergence, five more passes until a whole pass gained less than half a point, the record saved after every piece: colour kind butler 0.794, coach 0.730, colleague 0.793, homie 0.718, squire 0.644\"", "git log --oneline -1", "journal helper report \"Alfred, the sweep has converged. Five more passes ran back to back, and the last gained under 0.005 across all five voices together. The record is saved after every piece, so an interrupted sweep loses one piece at most.\n\n**Colour kind, then silhouette; every screen within 0.006 of its drawn size** (committed as $(git rev-parse --short HEAD)):\n\n| Voice | Colour | Silhouette |\n|---|---|---|\n| squire | 0.644 (from 0.527 before any sweep) | 0.881 |\n| homie | 0.718 (from 0.593) | 0.933 |\n| coach | 0.730 (from 0.605) | 0.929 |\n| butler | 0.794 (from 0.666) | 0.913 |\n| colleague | 0.793 (from 0.711) | 0.963 |\n\n**Two faults found and fixed at the source:**\n- **Pieces crowding a screen.** The early sweeps had grown the butler's hat-tipping hand and his hat over the edge of his screen, and moved the colleague's frames off her lenses. Those fits were cleared, and the sweep now refuses any change that takes a screen further from its drawn size.\n- **A misleading height correction.** Correcting screens by the compare page's figure-height estimate had made them worse. That estimate isn't what a screen's fraction is measured against, so the screens stay at the heights they were set for.\n\n**Next, step 2:** widening what a fit may move, one freedom at a time. The first is stretch: a piece's own width, length and depth sized separately, where today it has one size. Then I'll sweep it, and keep it only if it pays.\n\nPreview: http://127.0.0.1:8431/toon-preview/index.html\nOverlays: http://127.0.0.1:8431/toon-preview/rest/overlays.png\" 2>&1 | tail -1"]}