occurrence check with scope: sentence reports the alert at the wrong occurrence when a counted token repeats in a later block #1147

Check for existing issues

• Completed

Environment

• OS: Windows (11)
• Install method: Chocolatey
• Vale version (vale -v): vale version 3.17.1

Describe the bug / provide steps to reproduce it

An occurrence rule scoped to sentence (the pattern Microsoft's SentenceLength uses) counts words correctly but reports the alert's line/column at the wrong occurrence of the counted token. When a word from the over-long sentence appears again as the start of a later block (a list item or paragraph), the alert anchors to that later match instead of to the sentence. You navigate to the reported line, find a short unrelated sentence there, and it looks like a false positive. Reproducible with a stock occurrence/scope: sentence rule and no third-party package, so it is core behavior.

Steps to reproduce:

1. .vale.ini:

   StylesPath = styles
   MinAlertLevel = suggestion
   [*.md]
   BasedOnStyles = MyStyle

2. styles/MyStyle/SentenceLength.yml (identical to Microsoft's SentenceLength):

   extends: occurrence
   message: "Try to keep sentences short (< 30 words)."
   scope: sentence
   level: warning
   max: 30
   token: '\b(\w+)\b'

3. repro.md — the 33-word sentence is on lines 1–4 (its second sentence starts with Replace); line 6 is a short bullet that also starts with Replace:

   The `format` option is read automatically (the loader resolves it at startup
   level), so leave it as-is. Replace `<preset>` with a bare name (for example
   `fast`) to select that preset's defaults (colors, spacing, and margins with no
   other changes), or a full path (for example `./theme.json`) to load your own.
   
   - **Replace the renderer too.** Write the output yourself in a custom format.

4. Run vale repro.md.

Expected: the alert points at the long sentence, whose second sentence begins on line 2.

Actual:

repro.md:6:5:MyStyle.SentenceLength:Try to keep sentences short (< 30 words).

It is reported at line 6 — the bullet, a 12-word sentence that is not over-long. Deterministic across runs.

Confirming the cause: change the bullet's first word from Replace to any word not in the long sentence (e.g. Swap), and the alert correctly moves to repro.md:2:28 (the long sentence). The position is being resolved to a later match of a token shared with the sentence.