Shotnest · Library
Find a changed number in similar passages
Gate 12 now and Gate 13 now share two of four distinct tokens, a 0.5 overlap; the comparison lists 12 and 13 as the only differing tokens.
How the overlap figure is built
The comparison tool normalizes both texts, collapses runs of whitespace and splits them on spaces. It then counts distinct tokens: gate, 12 and now on one side, gate, 13 and now on the other. Two tokens are shared and four distinct tokens appear in total, so the overlap is 2 ÷ 4 = 0.5. The tool also lists what appears on only one side, which here is 12 in the first text and 13 in the second.
The number that changed is the only part that matters to a traveller, yet it moves the overlap by the same amount as any other word. Reading the one-sided token lists is therefore more useful than reading the ratio.
Try the original example
- First text
- Gate 12 now
- Second text
- Gate 13 now
- Exactly equal
- No
- Equal after normalization
- No
- Token overlap
- 0.5
Compare two versions of a notice
- Paste the older text first and the newer one second.
- Read Exactly equal and Equal after normalization; both are No here.
- Go straight to Only in first and Only in second and look for digits.
- Decide which value is current from the issuer's own app or desk, not from the comparison.
- Only in first
- 12
- Only in second
- 13
- Word overlap in the app's duplicate check, digits ignored
- 1
Why the app's cleanup ignores this difference
Shotnest's duplicate check compares letter-only words longer than one character, so clock times and numbers do not count; for these two lines its word overlap is 1. A pair is only offered as duplicates when the pictures also have the same width and height and nearly the same image fingerprint, and the newest of each set is kept.
- Deleting the older of two gate screenshots because the words match, without checking which number the airline shows now.
- Reading 0.5 as half the meaning changed; one changed token in a three-word text is enough to halve the figure.
- Expecting numbers to weigh more than words; every distinct token counts once, whether it is a gate number or the word now.
Similar is not current
Neither the ratio nor the token lists know when each screenshot was taken or which one the issuer still stands behind. Keep the timestamp of each image and treat the comparison as a way to find the difference quickly.