Paper Card Delivery

Core Contract

Treat paper-card work as source verification, not formatting. A finished card is allowed only after inspecting the official full-paper source end to end enough to verify the card fields: abstract, method, experiment/evaluation, conclusion/discussion, limitations when present, affiliations, datasets, code/project links, and selected figure captions.

Two failures disqualify a paper card from delivery. First, do not write plausible-sounding claims by inference when the official original paper has not been read; that is fabrication, even if the claim sounds likely from the title, abstract, project page, or method family. Second, if the official original paper is accessible by PDF, HTML, publisher/conference page, arXiv/OpenReview/CVF page, project page, official code, local rendering, MinerU extraction, or OCR, do not leave 待核验; continue source extraction and resolve the field.

If a field is 待核验, the required next action is to inspect the official full-paper source or other official source until the field is resolved. Do not leave 待核验 in reader-facing paper cards. A finished paper-card delivery must contain no 待核验, candidate, TODO/TBD/pending verification text, or other unresolved verification markers. For missing information, use 未报告, Not reported, N/A, or 不适用 only after checking the official paper/project/code sources and confirming the item is not reported or not applicable. 无法获取原文 is allowed only after every reasonable official acquisition route has failed, including official HTML, official PDF, arXiv/OpenReview/CVF/publisher/conference pages, DOI landing page, project page, author/lab page, official repository, local PDF download/rendering, and available extraction/OCR routes such as MinerU. When that happens, report a blocker with the attempted routes instead of shipping a finished card.

For formal delivery, every paper card must include a verified method/process figure or a user-approved figure, with a complete Chinese caption. Do not deliver cards with 配图待补, 图注待补, 图片待补, 图像待补, 待补配图, 待补图注, empty image slots, or text-only figure/caption TODO/TBD/pending notes. Any reader-facing sentence like 配图待补：优先从官方论文 HTML / PDF 中提取... is a hard delivery failure, not an acceptable caveat. If the figure cannot be obtained within the current turn, continue extraction, explicitly downgrade the page to a draft, reduce scope with the user's approval, or report that the deliverable is blocked. Never mark a paper-card page complete while any card is missing its figure or figure caption.

Paper Card and Paper Abstract Card are distinct deliverables. A Paper Card is the verified seven-field reader index defined below. A Paper Abstract Card is a smaller research-tracker unit containing only resource links, a Chinese abstract translation, and representative figures. Do not silently expand a Paper Abstract Card into the full Paper Card schema or collapse a requested Paper Card into an abstract-only entry.

Neither card is a full deep dive, PDF-to-Markdown manuscript, complete paper translation, or substitute for paper-deep-dive. If the user asks to 深读 / detailed-read / dive into one paper, use paper-deep-dive and create the requested card type only as one component of that larger package.

Use research-doc-workflow for platform selection, hierarchy, image/caption representation, and post-write verification. After the platform is resolved, also use the matching adapter: feishu-doc-workflow for Feishu/Lark, notion-doc-workflow for Notion, or obsidian-doc-workflow for Obsidian/vault Markdown.

For any reader-facing Chinese prose in the card, also use chinese-technical-writing. This card skill adds paper-card-specific structure; the broader Chinese-first terminology standard lives there.