{"claim":{"id":"8ade3a98-a181-4436-9aa4-82921270c8ee","submission_id":"9f641ab2-c310-48c4-8c74-b92588848fcb","submission_revision_id":"b5713e03-79d1-458d-ab22-a1f9cbd2bf34","builder":"builder-01","status":"active","claimed_at":"2026-10-15T15:40:20.682814Z","delivered_at":null,"ended_at":null,"delivery_note":null,"end_reason":null},"task":{"submission_id":"9f641ab2-c310-48c4-8c74-b92588848fcb","submission_revision_id":"b5713e03-79d1-458d-ab22-a1f9cbd2bf34","revision_version":1,"submission_status":"under_review","submitted_at":"2026-10-08T23:22:59.429639Z","submitter_name":"Rongjie Huang","submitter_email":"zjurongjiehuang@gmail.com","task":{"status":"submitted","version":1,"extra_fields":{"difficulty_self_test":{"model":"GPT-6 Astra","estimated_score":"5–10%"}},"submitted_at":"2026-10-08T23:22:59.429639+00:00","industry_domain":"Print production and prepress","task_short_name":"prepress_ultimate","operating_system":"Linux","task_description":"Objective:\nProduce the 12-page Harbour & Hearth catalogue from 24 delivered CGI product scenes: cut out each product at its physical outline, correct each setup from its own chart, place the finally approved live copy around the silhouette, and deliver an editable Scribus project and a print-ready PDF/X-4.\n\nStarting State:\nExtract prepress_ultimate_v13_inputs.zip as workspace/. It contains the public task_prompt.txt, ENVIRONMENT.md, PUBLIC_MANIFEST.json and input/ assets, including product and chart frames, briefs, amendments, the page plan, chart reference, fonts, profiles and the numerical acceptance contract. No hidden renderer truth, reference output, evaluator or model trajectory is solver input. The client job, product catalogue and correspondence are disclosed simulations.\n\nInstructions & Rules:\nPrepress job: Harbour & Hearth autumn/winter catalogue (12 pages, 24 products).\n\nRead input/brief.md first, then input/delivery_note.md, input/page_plan.json, input/catalogue_copy.json,\ninput/copy_amendments.md and input/chart_reference.json. The studio frames are in input/photos (with corrections in input/masters and\ninput/alternates); every frame has a colour-chart frame in input/charts. Put only the requested deliverables in\noutput/. Scribus 1.6.4 (scribus), GIMP 2.10 (gimp), Ghostscript, LittleCMS (Pillow ImageCms), OpenCV and\nscikit-image are installed; there is no network access.\n\n\n# Brief: Harbour & Hearth autumn/winter catalogue (ultimate-v13)\n\nThe catalogue shows every product cut out of its lifestyle frame and set on a full-bleed brand colour, with its\ndescription running around the product's own outline. You are the prepress house: produce the print-ready\ncatalogue and its editable source.\n\n## Pages\n- Twelve A4 portrait pages (210 x 297 mm trim, 3 mm bleed on every edge), in the order of page_plan.json.\n- Each page is a full-bleed fill of its brand colour (the CMYK values in page_plan.json, as process CMYK).\n- Each page carries its two SKUs. Layout, alignment and typography are free within these rules: keep live type\n  at least 8 mm inside the trim, nothing may overlap a product, products may not overlap each other, and every\n  product stays inside the trim.\n\n## Cut-outs\n- Every product is cut out of its delivered frame at its physical outline: none of the room, floor, table,\n  shadow, reflection, chart or other furniture may remain, and no part of the product may be lost. Background\n  seen through handles, between chair spindles and stool legs, inside wire cages, chandelier arms, carved\n  aprons and lattice or fretwork panels, and between fronds and leaves is not product. Lantern and chandelier glass is part of the product and\n  is reproduced opaque (it counts as product coverage). Fur and fibre edges keep their natural soft edge.\n- Soft-edge products (soft_edge_skus in page_plan.json: the faux-fur, sheepskin and long-pile items): the outline\n  of the fibre mass must be followed, neither trimmed inside the fur nor carrying a halo of background; an alpha\n  matte or a feathered path is acceptable at the fibre tips.\n- On press a fringe of background or a bite out of the product wider than about 0.2 mm is visible against the\n  brand colours and is rejected at proof. The quality contract treats a connected region of at least\n  six master pixels beyond its 0.2 mm proof tolerance as a visible local defect; this applies separately to\n  residual background and missing product, after fibre-mass smoothing on the soft-edge SKUs.\n- Keep the master pixel grid: crop only. Do not resample, rotate, flip or retouch the product pixels other than\n  the colour correction below. Place every product at 100 % so it prints at the master's 300 ppi.\n\n## Colour\n- Every frame carries a camera colour error: per-channel tone-response errors, crosstalk between the channels\n  (hue and saturation shifts) and an uncorrected white balance; each set-up's chart frame has exactly the same\n  error as its product frame. chart_reference.json gives the chart's published values (patches counted from 1,\n  row-major from the top-left).\n- Correct each product frame from its own chart frame. Profile the set-up so that the chart reproduces its\n  published values: a per-channel tone curve that brings the neutral patches 19 to 24 onto the published neutral\n  steps, followed by a 3 x 3 matrix fitted on all 24 patches. Brightness is anchored on neutral patch 22, whose\n  delivered green level is kept. Apply the same correction to the product.\n- Separate for ISO Coated v2 300% (profiles/ISOcoated_v2_300_bas.icc) with relative colorimetric intent and\n  black point compensation, or place RGB images with an embedded correct profile in a PDF/X-4 whose output intent\n  is that profile (the evaluator then applies the same intent). The printed product colour is what counts.\n\n## Copy\n- Per SKU: the name (Noto Sans Bold, 14 to 18 pt), the price and the SKU code (live text), and the description\n  (Noto Sans Regular, 9 to 11 pt) exactly as approved: catalogue_copy.json with the client's changes in\n  copy_amendments.md applied (a later email overrides an earlier one). Superseded names, prices and sentences\n  must not appear.\n- Each description wraps around its own product: at least four of its lines run beside the cut-out, each of\n  those lines keeping 3 mm from the visible product edge (no closer than 2.5 mm, and no further than 6 mm on the\n  side facing the product except the description's final line), so the text follows the silhouette rather than\n  a rectangle. Copy may not run into gaps inside a product's outline.\n\n## Deliverables (output/)\n- catalogue.sla: Scribus 1.6.4 source of all 12 pages with images linked relatively from output/assets/.\n- catalogue_press.pdf: PDF/X-4, 12 pages with trim and bleed boxes, embedded fonts, output intent ISO Coated v2\n  300%, exported from catalogue.sla.\n- assets/: the cut-out masters you placed (TIFF or PNG; clipping path or alpha channel).\n- handoff.json: {\"version\": \"ultimate-v13\", \"skus\": [{\"sku\": \"HH-1041\", \"source_frame\": \"photos/HH-1041.tif\",\n  \"asset\": \"assets/<file>\", \"crop_origin_px\": [x, y]}, ...]} where crop_origin_px is the top-left pixel of the\n  asset in the source frame.\n\nThe supplied quality_acceptance.md states the scoring weights, measurements and production acceptance tolerances for this delivery.\n","software_licensing":"Free/Open Source","evaluation_criteria":null,"verification_method":"The frozen standalone evaluator tools/grade_catalogue.py grades the delivered catalogue.sla, catalogue_press.pdf, assets/ and handoff.json against host-only validated truth. It returns a weighted score in [0,1], averaged equally over 24 SKUs, with silhouette 75%, colour 10%, wrap 5%, and source/copy 10% within each SKU. The supplied public quality_acceptance.md is the delivery contract. Missing/unreadable candidate deliverables lose dependent items; broken evaluator inputs or software are infrastructure errors, not candidate deductions. Native reopening/export actually runs Scribus 1.6.4 offline with candidate files and public fonts/profiles only; print composition is measured through pinned Ghostscript 10.02.1.\n\nRun in the recorded environment:\nPREPRESS_SCRIBUS_RUNTIME=/path/to/scribus-1.6.4 python3 -B tools/grade_catalogue.py --output /path/to/delivery --package /path/to/package --report /path/to/score.json\n\nSOFTWARE_BINDINGS.json and SOFTWARE_ACQUISITION.json record the provisioned software and resource requirements. Official ALE runnable integration and clean-machine installation remain pending.\n\nFull public numerical contract:\n# Grading (ultimate-v13)\n\nEach of the 24 SKUs is worth 1/24 of the score. Within a SKU (weights fixed before any trial of this version;\nthe cut-out is the paid deliverable of a silhouetting job, the other items are routine production checks):\n\n- 75% silhouette: the product's effective coverage as actually printed in catalogue_press.pdf (the combined\n  effect of any clipping path, soft mask or alpha, recovered by substituting two flat colours for the image\n  data and comparing Ghostscript 10.02.1 RGB renders with printer options;\n  printable annotations and declared Usage/Print layer states participate, and nonprinting annotations do not;\n  declared ON/OFF print states are normalized consistently before all print-composition checks) is mapped to the master pixel grid through the image placement and the\n  declared, verified crop origin, and compared with the true product coverage of the frame. Boundary F-measure at\n  2 px (0.17 mm at 300 ppi): 0.95 or better earns full credit, 0.70 or worse none, linear between. For the\n  soft-edge SKUs both coverages are first smoothed with a 3 px Gaussian, so the outline of the fibre mass is\n  graded and individual fibres are not (same F-measure and thresholds). Silhouette credit additionally\n  requires no visible local residual background or missing product beyond the physical 0.2 mm proof band (2.3622 px at 300 ppi). After the same\n  soft-mass treatment, extra and missing coverage are checked separately: an 8-connected wrong-coverage\n  region outside that proof band with at least six pixels (the area of a 0.2 mm square rounded up at 300 ppi)\n  rejects that SKU's silhouette item. Continuous coverage is checked as well: at each pixel it must stay\n  within the minimum and maximum true coverage in the same circular 0.2 mm neighbourhood, allowing\n  3/255 coverage for eight-bit raster quantization. Connected excess or deficit beyond that envelope\n  follows the same six-pixel rule. For soft-edge SKUs the candidate coverage and the permitted minimum/maximum envelope are smoothed by the same 3 px Gaussian, preserving the geometric freedom of individual fibres; partially transparent\n  background residue and faded opaque product interiors cannot evade the check by staying on one side\n  of the half-coverage threshold. Tiny isolated quantization pixels do not trigger this condition.\n- 10% colour: the product region (true coverage, eroded 3 px) as printed is converted through ISO Coated v2 300%\n  to CIELAB D50 and compared with the correctly profiled frame (the frame rendered with a correct camera, at the\n  delivered brightness of neutral patch 22), reproduced through the same press condition. Two equal parts: CIEDE2000 between the region means (2 or\n  less earns full credit, 6 or more none) and the 90th percentile of CIEDE2000 over 16 x 16 px block means (3 or\n  less full, 8 or more none; local blocks carry the frames' sensor noise and shadow quantisation).\n- 5% wrap: description lines beside the product (sharing its vertical band) are measured from their glyph boxes to\n  the true silhouette at its printed position. A line qualifies when it keeps at least 2.5 mm from the silhouette and\n  its facing edge is within 6 mm of it (the description's final line only needs the 2.5 mm clearance). Credit = qualifying lines / max(4, lines beside the product).\n- 10% source and copy, five equal parts: correct source frame (the declared frame shows this SKU's product and\n  follows the delivery-note corrections, and the printed product keeps the master's fine detail: at least 85% of its fine-detail pattern is retained: after a 0.6 px Gaussian, the demeaned Laplacian of the candidate's press lightness is projected onto the corrected delivered master's demeaned Laplacian inside the product; both the retained contrast and normalized fine-pattern correlation must be at least 0.85. The corrected master is produced from the public source and measured chart and includes the source sensor detail. Unrelated noise, denoising and sharpened upsampled proxies cannot replace the original spatial pattern); effective resolution\n  at least 280 ppi; the finally approved name and price (copy deck plus client amendments) and the SKU code as\n  visible live text on the SKU's page, with no superseded name or price on that page; the complete finally approved\n  description (at least 98% of its words in order) with no superseded sentence; the placed asset linked from catalogue.sla (all pages) with the handoff version: a readable relative PNG/TIFF link, matching master pixels and placement, approved live copy that remains visibly complete on the native page after an actual reopening/export, and a Scribus reopening/export that reproduces the delivered product region within the same full-credit colour tolerances (mean ΔE00 ≤ 2 and block p90 ≤ 3), with its effective coverage agreeing\n  with the delivered PDF at boundary F ≥ 0.95 and with no visible local extra/missing region under the same\n  0.2 mm / six-connected-pixel condition, including the continuous-coverage envelope and its 3/255 quantization allowance. Empty spaces in the image crop are included in this native comparison.\n\nA product whose placement cannot be identified earns no silhouette, colour or wrap credit. Missing or unreadable\ndeliverables lose the dependent items only. Broken evaluator inputs or software are infrastructure errors (no\nscore). The PDF/X-4 output intent, trim/bleed boxes and embedded fonts are checked as preconditions of the\ncolour item (PDF/X-4 declaration in XMP, the complete supplied ISO Coated v2 300% profile, all four bleed edges and actual embedded font programs; a missing precondition earns no colour credit). The ICC creation date and profile ID may differ without penalty. Colour is measured from the composed page through Ghostscript 10.02.1 with relative colorimetric intent and black-point compensation; overlays and transparency therefore count. A native validation export runs offline with only candidate files, public fonts/profiles and the specified Scribus software. Unreadable image objects are omitted from that temporary validation copy and cannot earn their own native credit; unrelated linked products remain assessable.\n\nThe unchanged page and pixel-grid requirements qualify the existing components. Both image axes must be\nupright at 300 ± 3 ppi for silhouette, colour and wrap measurements; the resolution source sub-item requires\nboth axes to be at least 280 ppi. Silhouette credit also requires the product coverage to stay inside trim and\nnot overlap another product. The colour item requires the full-bleed brand fill outside product and glyph\nareas to meet the same full-credit mean ΔE00 ≤ 2 and block p90 ≤ 3 tolerances against the prescribed process\nCMYK colour. The flat fill also requires every measured 72-ppi background sample to have ΔE00 ≤ 3;\nsmall foreign patches cannot be diluted in the page average. Boundary antialiasing is excluded at the\n72-ppi background check. Product exclusion includes all true matte coverage above 1/255, so natural\nlow-opacity fur tips are excluded from flat-background samples.\n\nApproved name/price/code credit requires visible glyphs, 8-mm trim clearance and no product overlap; names\nuse Noto Sans Bold 14–18 pt. Approved description credit and wrap credit require visible Noto Sans Regular\n9–11 pt glyphs with the same page and product clearance. Font sizes allow 0.05 pt serialization rounding.\nVisible copy is checked against the actual embedded glyph program as well as Unicode and font labels.\nFor Noto Sans, painted outlines must match the supplied font's corresponding Unicode glyph or legitimate\nOpenType alternate; subset numbering and font names do not establish character identity. Other embedded\nprice/code fonts must retain their program's Unicode-to-glyph correspondence. Copy is also checked with\ntext rendering modes and opacity, plus a 300-ppi comparison of the actual printed composition against the same\npage with only live text removed. Both rasters use the pinned Ghostscript press conversion with print\noptions and annotations enabled; printable and print-only annotations participate in occlusion. Nonprinting\nannotations do not hide copy in the print visibility check. A retained non-space glyph must contribute at least one pixel with\n8/255 RGB contrast attributable to that glyph. If live glyph cells overlap, changes assigned to a later\npainted glyph cannot be borrowed by an earlier hidden glyph; antialiasing is included with a one-raster-pixel\ncell margin. Invisible, transparent, occluded, clipped and white-on-white text cannot earn credit.\nThese checks qualify the original items; no weights or scoring budget are changed.\n\nCopy and description wrap must stay outside the product’s outer outline: enclosed transparent gaps are excluded from copy placement and wrap-distance credit. This affects copy/wrap checks only; actual cut-out transparency, product colour, background and silhouette metrics retain the true coverage.\n","software_and_version":"Scribus 1.6.4, Ghostscript 10.02.1","specific_requirements":null,"expected_completion_time":null},"files":[{"file_id":"47b9085d-8e30-4334-9488-ec6c85d1f31e","filename":"prepress_ultimate_v13_standalone_scorer.zip","category":"reference_output","content_type":"application/zip","size_bytes":79099},{"file_id":"a9a57efa-3ccc-4f17-ba8e-061c5610826b","filename":"prepress_ultimate_v13_reference_outputs.zip","category":"reference_output","content_type":"application/zip","size_bytes":78585542},{"file_id":"02370d4b-9a26-468e-a9ec-fbe63dfd1558","filename":"prepress_ultimate_v13_evaluator_truth.zip","category":"reference_output","content_type":"application/zip","size_bytes":421276654},{"file_id":"3fe8cc35-62c4-4a59-abc5-b0e2eb4712a1","filename":"prepress_ultimate_v13_astra_evidence.zip","category":"difficulty_self_test_evidence","content_type":"application/zip","size_bytes":86930113},{"file_id":"948269a3-06a8-468d-abc9-a298514d6647","filename":"prepress_ultimate_v13_inputs.zip","category":"input_materials","content_type":"application/zip","size_bytes":1071173202},{"file_id":"fe475248-b59b-4c38-9aed-2bb89a208bfb","filename":"prepress_ultimate_v13_submission.zip","category":"reference_output","content_type":"application/zip","size_bytes":1693171006}],"deep_review":{"overall_verdict":"Accept","review_text":"**Overall Verdict:** Accept — complete, coherent, objectively evaluable, and reachable; disclosed simulation provenance and pending clean-machine integration are non-blocking.\n\n**Completeness Check:**\n- Instruction: Complete — detailed 12-page catalogue objective, cut-out, color, copy, layout, and delivery requirements.\n- Inputs: Complete — 24 product frames, 24 chart frames, corrections, copy/amendments, page plan, profiles, fonts, and acceptance contract are present and schema-consistent.\n- Toolchain: Complete — exact Scribus/Ghostscript versions, offline environment, software bindings, acquisition notes, and licensing are documented.\n- Reference output: Complete — reference archive contains 24 assets, editable Scribus source, PDF, handoff data, and layout state; evaluator truth contains per-SKU alpha/target/meta data.\n- Rubric: Complete — numerical acceptance contract, standalone grader, native reopen/export checks, and a passing reference score are supplied.\n\n**File and Task Reachability:**\nAttachment classification: reachable. Task classification: reachable. I inspected `FILES.json` and all six ZIP attachments: the standalone scorer, reference outputs, evaluator truth, Astra evidence, input materials, and submission packet. ZIP headers, central directories, entry counts, manifest sizes, encryption flags, path safety, representative JSON/text schemas, TIFF/PNG/PDF/Scribus/NPY/font/ICC signatures, reference scoring, and candidate scoring evidence were checked. The solver-visible input archive contains no reference truth or evaluator payload; those are separated into protected reference/evidence archives. Bulk raster payloads and most command-log members were not fully decompressed because bounded sampling was sufficient; this is a reviewer-tooling limitation, not an attachment defect. No checksum verification is claimed.\n\nLicense keys, accounts, tokens, and other credentials are not required from the contributor.  \nConfidential, proprietary, or internal material may be evaluated as submitted evidence without requiring additional disclosure.\n\nThe reference delivery reports a successful native 12-page export and score of 1.0; the single Astra screening run also completed its evaluator path, though its candidate scored 0.06.\n\n**Special Screening:**\nNone. The provenance openly identifies simulated CGI material and AI-assisted drafting, but does not establish material AI fabrication of the core evidence under the policy threshold.\n\n**Recommended Changes:**\nThe submission is ready as provided; no contributor revision is required. Optional hardening would be a clean-machine installation/integration run and human expert verification of the simulated prepress reference.","reviewed_at":"2026-10-08T23:37:35.525592Z"},"claim":{"id":"8ade3a98-a181-4436-9aa4-82921270c8ee","submission_id":"9f641ab2-c310-48c4-8c74-b92588848fcb","submission_revision_id":"b5713e03-79d1-458d-ab22-a1f9cbd2bf34","builder":"builder-01","status":"active","claimed_at":"2026-10-15T15:40:20.682814Z","delivered_at":null,"ended_at":null,"delivery_note":null,"end_reason":null}},"revision_is_current":true}