Proof comparison, Chinese language conventions and quality evidence
Separate agreement with copy from correctness
A proof can reproduce the manuscript faithfully while preserving a factual error. Conversely, an unauthorized change can create a new discrepancy even when its wording sounds fluent. Compare against the controlling copy, read for internal consistency and use appropriate external evidence where needed. These functions complement one another. A method name is not a substitute for actually checking the line, reference, number, table or context.
Communicate an unambiguous correction
Use the applicable current proof-marking convention or an explicitly agreed digital annotation system. Mark the exact location and intended replacement, insertion, deletion or rearrangement without ambiguous crossing instructions. A textual description of a symbol is not a substitute for practicing the actual approved graphic mark. Where a proposed deletion is withdrawn, preserve a clear record of the intended retained text rather than leave two conflicting commands. Consolidate authorized changes into the master version without silently accepting every note from every contributor.
Apply language rules in context
Chinese punctuation, numeral use, quotation, title marks, parentheses and line-breaking choices depend on the syntactic function and actual governing style. An approximate numeral such as 三四 differs from an enumeration of two separate quantities. A quoted full utterance and a quotation integrated into another sentence can need different punctuation treatment. Avoid blanket rules such as banning a colon after every occurrence of a particular introductory word. Consult the actual current national standard and examples when the context is uncertain rather than infer a rule from a memorized fragment.
Verify changes and interpret a metric
After correction, check both the intended change and its surroundings for dropped characters, altered references, line movement and new errors. A quality-rate calculation requires the defined error weights, inspected denominator and unit. Multiplying a ratio by ten thousand gives the stated per-ten-thousand measure, not a percentage with the same numeric value. A favorable sample does not prove every page is perfect. Actual Chinese quality rules may define weights, repeated errors and consequences in detail; this bank labels hypothetical scoring inputs instead of inventing universal weights or treating a model result as official acceptance.