SuperAPI AI Directory

paper-quality-gate

Testing & quality

View repository

Updated 10-16 08:46

About this skill

Paper quality gate

Internal deterministic gate used after bounded LaTeX sanitization and immediately before compilation. Input is a JSON object containing paper_contract, length_gate, citation_gate, and manuscript_package. The preceding sanitizer may normalize safe punctuation and replace detected forecast magnitudes with explicit placeholders, but this gate remains an independent fail-closed audit of the resulting artifact.

The command exits non-zero when an upstream LLM gate blocks, when a required verdict marker is absent, when the manuscript artifact cannot be read, or when an evidence-free manuscript presents empirical findings as completed results. It also rejects concrete English or Chinese predicted outcome numbers (for example, an anticipated 12% improvement) when evidence is absent. Experimental setup parameters remain allowed; a predefined metric target such as 50% accuracy is also allowed when the unknown time-to-target remains nonnumeric. Explicitly conditional claims and planned significance thresholds are not treated as observed findings. Unknown outcomes must use TBD / 待实验确定 instead of invented point estimates.

Every figure/table caption is audited independently when evidence is absent. Categorical outcome claims such as "the proposed method achieves lower cost" or "the proposed method maintains the lowest communication cost" are blocked. Neutral planned placeholders and explicit hypotheses/future tests remain valid, including captions that state concrete setup values.

The gate also rejects literal Unicode Greek math glyphs, Unicode en dashes, and single Latin-style em dashes in the manuscript. Paired Chinese em dashes (——) remain valid native-language punctuation. Authors should use named LaTeX math macros such as \(\alpha\) / \(\varepsilon\) and LaTeX -- / --- punctuation, avoiding silent missing-glyph degradation in XeLaTeX.