Rows: 7,005 - 1,056 new rows from eighteen batches of the volume run (540 from sampled specifications, 516 from seeds). From batch 109 each batch leans to real pictures: six series of three seeded from real pictures (long looks with their closer and quick looks cut from them, mostly from DOCCI's descriptions) and single rows from seeds whose pictures are public, so that 60% of a batch shows a real picture. Each batch passed the validator and was read before it was added; at the read a few thoughts were reworded to describe the picture, a few genres and one domain were corrected, and one relation that put a held-out word into a record was removed. Seeds whose description contradicts their own records are skipped. Every row fits the declared features. The card's row counts and status are updated.
Rows: 5,949 - 913 new rows from sixteen batches of the volume run (630 from sampled specifications, 283 from seeds). Batch 096 tries long looks at real photographs: eight series of three seeded from DOCCI descriptions, each long look kept to what its description states, with its closer and quick looks cut from it. Each batch passed the validator and was read before it was added; at the read a few thoughts were reworded to describe the picture (a size order that contradicted its own relation, thoughts that named the specification's categories or read a recorded role back), a few genres and two domains were corrected, and two rows whose seeds held no subject (only light, focus or framing) were removed. Seeds without a subject are no longer drawn, and a relation's ends always name things in the picture. Every row fits the declared features. The card's row counts and status are updated.
Rows: 5,036 - 1,075 new rows from eighteen batches of the volume run (756 from sampled specifications, 319 from seeds drawn from all eleven sources). Each batch passed the validator and was read before it was added; at the read two thoughts were reworded to describe the picture rather than places in general, and two drawn series were given one genre across their three depths (a series takes the kind of picture its deepest specification sets). Seeds whose description never names the specification's main object are now skipped. Every row fits the declared features. The card's row counts are updated.
Rows: 3,961 - 896 new rows from fifteen batches of the volume run (630 from sampled specifications, 266 from seeds). Each batch passed the validator and was read before it was added; at the read one thought was reworded and two genres were corrected (a still life holds no people; a built sculpture is architecture). PartiPrompts joins the seeded rows from batch 065, so seeds now come from all eleven sources. A correction to the previous release's message: batches 036-062 drew their seeds from ten of the eleven sources (PartiPrompts was not drawn), not from all eleven. Every row fits the declared features. The card's row counts are updated, and its description of the batches no longer claims all eleven sources.
Rows: 3,065 - 713 new rows from twelve batches of the volume run (504 from sampled specifications, 209 from seeds drawn across all eleven sources). Each batch passed the validator and was read before it was added; at the read three thoughts were reworded (a definition, or a word about the specification, became an observation of the picture) and one genre was corrected. Seeded subject labels are mapped to kinds of picture (a coin or a cup shown for itself is a product, food on a table a still life, a bus in traffic a street picture). Every row fits the declared features. The card's row counts are updated.
Rows: 2,352 - the pilot's 1,995 repaired after a read by eye, and 357 new rows in six batches with seeds from all eleven sources. Repairs in the pilot: thoughts that defined a word instead of looking at the picture removed (490) or rewritten as observations of the picture (141); every genre now a kind of picture (1,363 rows relabelled: made-up labels and subject categories such as "harbour portrait", "animal" or "vehicle" became portrait, environmental portrait, genre scene, street, still life, documentary and the like); 50 remarks about the writing or the specification made observations of the picture. Every row passes the validator and fits the declared features. The card's status, its row counts (design and sculpture rows now appear) and its description of the genre field are updated.
Rows: fixes from a read of 50 rows by eye, in 89 rows. The genre is now the kind of picture (portrait, environmental portrait, street portrait, street, figure study) in 25 series and one seeded row. In 33 rows, thoughts that defined a word instead of looking at the picture were removed or rewritten as observations of the picture, and remarks about the writing were removed. Row counts unchanged (1,995); every row passes the validator.
