assets/boards/morphology_references.json
{
"_what": "MORPHOLOGY STAGING REFERENCES -- the images the rung-4 identity rows condition on, and the provenance of each. These are GENERATION INPUTS, not delivered artifacts: like a prompt or a seed, a reference shapes a plate and is never itself a plate. That is why they carry NO per-artifact licence record. harness/check_generation_lifecycle.py tooth L9 is correct to call a licence record with no consuming row an orphan -- 'a stale declaration rots into a standing pre-authorization' -- and the right answer was to fix the ontology rather than to enrol a non-artifact as an artifact. Their identity is still enforced mechanically: each reference's sha256 rides inside the hashed sampler_config of every row that conditions on it (docs/generation_lifecycle.json prompt_hash), so replacing a reference changes artifact identity exactly as replacing a seed does, and every consuming plate's own licence record names the reference by id, source and sha.",
"_why_staged": "The task's preferred reference is a LANDED plate of the same subject class. For these two species that route is unsafe and the evidence is a LOOK: all nine landed shell plates were opened and read and every one carries the hard-scuted carapace the REC line names, and all twelve landed macaque plates carry the bright red-pink face mask its REC line names. Conditioning on them would have PROPAGATED the defect. So the reference was staged from the creature registry's own species name, through the same Apache-2.0 Z-Image-Turbo weights every plate in this lane uses -- never an un-licenced photo, and never a plate a LOOK has just convicted.",
"generating_model": "Tongyi-MAI/Z-Image-Turbo",
"model_revision": "z_image_turbo_bf16.safetensors",
"model_hf_revision": "f332072aa78be7aecdf3ee76d5c247082da564a6",
"model_licence": "Apache-2.0",
"licence_class": "shippable",
"previz_only": false,
"rights_value": "generated_local",
"sampler_config": {
"sampler": "res_multistep",
"scheduler": "simple",
"steps": 8,
"cfg": 1.0,
"width": 1344,
"height": 1344
},
"references": [
{
"key": "softshell",
"reference_id": "PLT_S001_1",
"path": "E:/art/boards/_staging/morphology/softshell__PLT_S001_1.png",
"sha256": "d938707ab25fb62bf81dbe2dbd2311a462f9ecc9481701a28ce4bbf0514f4f74",
"bytes": 2654278,
"seed": 7401,
"canon": "CR_0153 Flores Giant Softshell Turtle (Chitra chitra javanensis), FAM_01 Shell",
"prompt": "An Asian narrow-headed softshell turtle, Chitra chitra, lying flat on wet sand, photographed from DIRECTLY ABOVE as an anatomical morphology study. Its back is a wide flat oval of BARE SMOOTH LIVING SKIN, like a slab of wet grey-olive leather stretched taut over the body, one unbroken rubbery surface finely vermiculated with pale wandering lines. THE OUTER THIRD OF THAT OVAL IS A WIDE FLOPPY SKIN FLAP that lies loose and softly rippled where it settles on the sand, the way the disc of a stingray settles. The whole animal is as flat as a pancake in profile. At the front a small narrow head sits on a long extensible neck and ends in a SHORT FLESHY TUBULAR SNOUT like a little snorkel, the eyes small and set high and far forward near the snout base. The four limbs are broad flat webbed paddles of loose folded skin, each carrying three short claws, and the tail is a short soft stub. Even flat overcast daylight, painterly AAA concept-art realism, the whole animal in frame.",
"consumed_by_plates": [
"PLT_F041_1",
"PLT_F042_1",
"PLT_F043_1"
]
},
{
"key": "macaque",
"reference_id": "PLT_S002_2",
"path": "E:/art/boards/_staging/morphology/macaque__PLT_S002_2.png",
"sha256": "258c0730242c8a573c26d3fc9bc2d98b6f14badaf5fdd92a19fc90c93ecebf4a",
"bytes": 1793384,
"seed": 7402,
"canon": "long-tailed macaque, Macaca fascicularis, Flores highland troop",
"prompt": "A single long-tailed macaque seated on a plain pale studio ground, full body in three-quarter view with the head turned to camera, an anatomical morphology study. Its bare facial skin is MUTED GREY-BROWN FLESH, one even quiet skin tone across brow, eyelids, cheeks and muzzle, the muzzle a shade paler grey and the whole face matte. A crest of brown-grey crown fur sweeps back from the brow, the cheek whiskers are pale, the eyes are amber-brown and close-set. The coat is olive-brown above and pale grey-cream on the chest and belly, the hands and feet dark grey with five separated fingers, and the tail is long and slender, longer than head and body together. Even flat studio light, plain neutral ground, painterly AAA concept-art realism, the whole animal in frame.",
"consumed_by_plates": [
"PLT_F075_1"
]
},
{
"key": "buffalo",
"reference_id": "PLT_S003_2",
"path": "E:/art/boards/_staging/morphology/buffalo__PLT_S003_2.png",
"sha256": "0eef9eca3c5e93291d1b3ee689c2abac02d6b6680b8db742a8afcdab1568fd4a",
"bytes": 1987795,
"seed": 7404,
"canon": "CR_0012 Water buffalo (Flores domesticated), Bubalus bubalis, the Ch-3 megafauna row",
"prompt_note": "Staged 2026-08-04 for the Ch-3 leg because CR_0012 carries ZERO landed plates, so there was no same-class reference to open. Affirmative-only on the same law as every dispatched prompt: the bison prior is displaced by stating what IS there (bare near-hairless slate hide, a level back, a smooth shoulder line), never by naming what is not.",
"rolls_read": "FOUR rolls were generated (seeds 7403-7406) and ALL FOUR were opened and read before one was used, per the standing rule. Roll 1 was REJECTED on its horns; roll 2 (seed 7404) was accepted.",
"accepted_because": "The four bison-prior cells the defect line names are answered: bare near-hairless slate-grey hide caked with dried pale mud, a back running flat and level from withers to rump, a smooth un-humped forequarter, and a near-side horn that is a long heavy transversely-ridged crescent sweeping BACK along the plane of the neck to a blunt tip behind the ear.",
"named_residue": "THE FAR HORN. On all four seeds the near horn sweeps back correctly and the far horn rises as a short pale up-spike, so the pair reads asymmetric. That is a reproducible model prior across four seeds, not seed noise. It is recorded rather than averaged away, and the consuming row ran two LoRA-only candidates beside the two conditioned ones so a judge could decide the residue on the image.",
"residue_outcome": "THE RESIDUE TRANSFERRED, exactly as this note predicted it might. The four-axis pick pass read all four candidates on the consuming row and returned REGENERATE_ALL: no candidate carried the wide backward horn sweep. This entry is therefore evidence FOR the standing rule rather than against it - naming the residue before use is what made the transfer legible instead of mysterious.",
"consumed_by_plates": [
"PLT_M0024_1",
"PLT_M0024_2"
]
},
{
"key": "python_cr0002",
"reference_id": "PLT_A0062_3",
"path": "E:/art/boards/D1-15/CR_0002/study__080fa93d1421.png",
"sha256": "20cd5652086d38531118cc1387b37de19fafecbf64ee34665bcd62ac53574b84",
"bytes": 2706729,
"provenance": "A LANDED PLATE, not a staged one - the first entry here of that kind, and it is the case the ruling section 2a says is PREFERRED when it is safe. It is recorded here anyway, because the rule that matters is not where a reference came from but whether it was read: 'condition on a plate whose morphology has been READ FOR THIS PURPOSE, and where none exists, stage one.'",
"canon": "CR_0002, the reticulated python that carries the roused naga tana rendering",
"read_before_use": "All complete same-class candidates were opened 2026-08-04 against this row's own defect line (a heavy-bodied constrictor with a flat wedge head, olive-and-black net patterning and a smooth unhooded neck; anything hooded, crested, horned, whiskered or dragon-like propagates the defect). PLT_A0062_1 was REJECTED - it carries a duplicate second head, and a reference with an extra head installs an extra head. PLT_A0062_2 was usable. PLT_A0062_3 was CHOSEN: one animal, one head, flat wedge skull, smooth unhooded neck, correct net, and a composition that is ALREADY a guardian entering a plunge pool off wet black rock. PLT_C074_1 was kept as a fallback and not used - correct head, but a studio plate on plain ground where the staging rows want the staged composition.",
"fallback_not_taken": "The synthesize-instead branch did NOT fire for this species. Unlike the softshell and the macaque, this corpus is not convicted on python morphology - the landed plates are right, and the note is recorded so the next reader does not re-stage what already exists.",
"outcome": "MIXED across its three consuming rows. On the register plate the conditioned candidate WON the pick. On both staging rows the conditioned candidates grew a SECOND HEAD out of the reference's far coil while the unconditioned candidate came back clean and took the pick. Read together with the softshell row, which lost its child to the reference's single-animal composition, the finding is that reference conditioning is a REGENERATION remedy for a row with a named defect, and is not safe as a first-generation mechanism: on a first pass there is no defect to repair and the imported composition is the only thing it reliably brings.",
"consumed_by_plates": [
"PLT_M0021_1",
"PLT_M0021_2",
"PLT_M0030_1",
"PLT_M0030_2",
"PLT_M0040_1",
"PLT_M0040_2"
]
}
],
"attempted_and_failed": [
{
"key": "tracks",
"canon": "Timor deer (Rusa timorensis) hoof sign, Flores highland volcanic mud",
"attempts": "two prompt rounds x two seeds = 4 stagings, all generated and LOOKED AT",
"finding": "every attempt returned a HUMAN bare-foot or boot-sole impression. 'Prints in mud' reaches this model's footprint prior before it reaches any animal -- which is precisely the defect row 77 names. No landed track plate exists anywhere in the corpus either, so no reference could be produced licence-cleanly.",
"consequence": "Row 77 was dispatched with the style LoRA plus the affirmative morphology clause and NO reference conditioning, and that was recorded in its payload as reference_source 'none_producible' rather than hidden. IT THEN PASSED ITS LOOK: at 2x each mark resolves as two teardrop gouges with a mud wall between them, no boot sole anywhere. The words that could not stand alone as a staging worked once the LoRA carried the render."
}
]
}