Source
납품일: 2026년 10월 15일 대금 지급: 검수 후 30일 이내 운송비는 공급자가 부담한다.
A polished revision can still change a deadline or who pays a cost. Use line comparison to locate changed sentences and word comparison to inspect their numbers and wording.
We changed dates, payment terms and shipping responsibility in a synthetic supply note. Both inputs are provided so you can repeat the comparison.
Comparison level is a zoom setting, not a quality ranking
Line comparison locates changed passages
It quickly narrows a long document, but one changed character can cause the whole line to appear modified.
Word comparison reveals the expression inside a line
It isolates small edits, but counts a removed segment and its replacement as separate change segments.
Whitespace and line breaks need a separate decision
Filters reduce copied formatting noise, but spaces can be meaningful in Markdown, code, and identifiers.
The responsible party changed, not only the numbers
A synthetic supply note changes day 15→30, payment 30→45 and shipping responsibility from supplier to buyer, then adds a warranty clause. Most text stays the same while the meaning changes.
Below are the source, revision and actual changed spans calculated with the tool’s comparison library. Use lines to locate changed sentences, then inspect words for numbers and responsibility. Locating a change is not approving it.
Revision
납품일: 2026년 10월 30일 대금 지급: 검수 후 45일 이내 운송비는 구매자가 부담한다. 하자보증기간은 24개월이다.
Actual word-diff spans — minus removed, plus added
- 15 + 30 - 30 + 45 - 공급 + 구매 + 하자보증기간은 24개월이다.
- Download file · TXTSource TXT for repeat comparison
- Download file · TXTRevision TXT
Measured result: the counts described different kinds of change
Changing 5 to 6 produced one modified line in line mode. Word mode showed two segments: one removal for 5 and one addition for 6. Two does not mean twice as much editing; it describes the two sides of a single replacement.
The meeting-report test produced four line differences and six word-level segments. The four line differences were three modified lines and one added line. The six word segments were two removals and four additions covering the date, weekday, output format, and new review note.
| Test | Actual edit | Line result | Word result |
|---|---|---|---|
| Replace one number | September 5 → September 6 | 1 difference · 1 modified line | 2 segments · 1 removed + 1 added |
| Edit meeting report | Edit 3 lines + add 1 line | 4 differences · 3 modified + 1 added | 6 segments · 2 removed + 4 added |
| Reflow only | Place the same words on 1 line and 3 lines | 3 differences · 1 modified + 2 added | 4 segments · 2 spaces removed + 2 newlines added |
Line comparison is a map of the document
Line comparison treats everything between two newline characters as one unit. If only one digit changes, that line is still a modified line. The highlight is broad, but additions and omissions are easy to locate across dozens of paragraphs.
The GNU Diffutils manual likewise describes traditional diff as comparing files line by line and reporting groups of differing lines. This view fits meeting notes, configuration lists, subtitles, and other material where each line represents structure.
- Meeting notes and checklists: locate an added or removed item
- Configuration and logs: identify which line changed
- Long reports: narrow the document to modified passages first
Word comparison zooms into a changed line
Word comparison can treat words, punctuation, spaces, and newlines as tokens. It reveals the exact text that disappeared and appeared, making it useful when one number, date, name, negative term, or obligation can change the meaning.
Do not read the segment count as the number of words edited. A continuous removal and a continuous addition are counted separately, so one replacement such as 5 to 6 produces two displayed segments.
- Contracts and policies: inspect may, must, except, and similar terms
- AI revisions: verify changed numbers and wording against the source
- Emails and reports: find typos and short phrase replacements
Why line breaks alone still produce differences
The third test replaced two spaces with two newline characters without changing any words. Line mode reported one modified line and two added lines. Word mode reported two removed-space segments and two added-newline segments. The wording was identical, but its line structure was different data.
Text copied from a PDF or webpage may therefore benefit from joining artificial wrapped lines before comparison. Do not do that to Markdown lists, source code, subtitles, or other content where line position carries meaning.
Use whitespace, case, and blank-line filters as a second view
A separate test changed the case of Q3-REPORT, added spaces around one value, and inserted one blank line. With no filters, line mode reported three differences. With whitespace, case, and blank lines all ignored, the result fell to zero.
A filter does not repair the source; it only changes what the comparison considers equal. Review the unfiltered result first, and use filters only when spaces or case are cosmetic. They can be real data in paths, passwords, code, Markdown, and product identifiers.
Choose the order that matches the document
For a long document, start with lines to locate changed passages and then switch to words for important passages. For one or two short sentences, word mode is a reasonable starting point. If copied formatting overwhelms the result, inspect a second view with selected whitespace filters.
The key is not to treat the change count as a score. Four line differences and six word segments use different units and cannot be compared directly. Read the highlighted source and revision to decide what the deletion and addition mean.
| What you need to check | Start with | Why |
|---|---|---|
| Added or removed paragraphs and items | Line mode | Locates changed structure quickly |
| Numbers, dates, names, and short phrases | Word mode | Shows the exact replaced segments |
| Whitespace noise after copying | Lines + selected filters | Reduces cosmetic marks without hiding all context |
| Markdown, code, and configuration | Lines, filters with caution | Line breaks, spaces, and case may be meaningful |
Scope and limits of this experiment
The reported counts are reproducible for the listed strings and the comparison behavior tested on September 3, 2026. Sentence length, repeated text, moved passages, and a different comparison program can group changes differently.
Word-level does not guarantee perfect human word boundaries in every language. This engine tokenizes words, punctuation, whitespace, and newlines, while other programs may use characters or sentences. The highlighted content matters more than the headline count.
Test conditions and reproduction
Three revision patterns were measured with the same comparison engine
- 1
The test was run on September 3, 2026. Line mode treats the content between line breaks as one unit. Word mode tokenizes words, punctuation, newlines, and runs of non-newline whitespace.
- 2
Test one changed only the 5 in ‘The deadline is September 5.’ to 6.
- 3
Test two changed a date, weekday, and output format in a six-line meeting report, then inserted one review-note line.
- 4
Test three changed no words and only reflowed one line into three lines.
- 5
Line results counted added, removed, or modified rows. Word results counted each continuous removed or added highlight as one change segment.
Raw reproducibility records (JSON)
The main results are explained above. These raw records contain processing conditions and file hashes. Opening a link displays JSON text, not an illustrated article.
Official documentation used for this analysis
Use lines to locate a change and words to judge its meaning
Line and word modes produced different counts for the same revision. In the meeting-report test, three edited lines and one inserted line became four line differences but six word-level segments. That is a difference in grouping, not a ranking of accuracy.
Start a long review with lines, then zoom into important passages with words to verify numbers, dates, conditions, and wording. Use whitespace filters only when formatting is truly cosmetic, and preserve line structure when it carries meaning.