False positives with then→than

[hippietrail]

Describe the bug
In some contexts a correct "then" is flagged and incorrect "than" is suggested.

To Reproduce
Steps to reproduce the behavior:

1. Paste in "And if just one of those is set incorrectly or it has the tiniest bit of dirt inside then that will wreak havoc with the engine's running ability."
2. Notice that this "then" connects a consequence of an earlier "if" with a result.
3. Hover over the flagged "then" and see it wants to replace it with "than", which won't make sense in this context.
4. Another example is "So let’s check it out then."

Expected behavior
No error flagged.

Screenshots

Platform
VS Code, Windsurf

[hippietrail]

I seem to still be getting these:

Generated contexts for this fragment:

• He was always a top student in school but then his argument is that grades don't define intelligence.
• She loved the atmosphere of the school but then his argument is that it lacks proper resources for students.
• The teacher praised the efforts of the school but then his argument is that the curriculum needs to be updated.
• They were excited about the new program at school but then his argument is that it won't be effective without proper training.
• The community supported the school but then his argument is that funding is still a major issue.

The original was from unscripted rambling YouTube transcript with run-on sentences etc. But here are some equivalent generated examples that all get the same erroneous flag:

• So then these resistors are connected up in parallel to reduce the overall resistance.
• So then these resistors are connected up to ensure the current flows properly.
• So then these resistors are connected up to achieve the desired voltage drop.
• So then these resistors are connected up to demonstrate the principles of series and parallel circuits.
• So then these resistors are connected up to optimize the circuit's performance.

Generated contexts for this fragment:

• Yes so then sorry you didn't receive the memo about the meeting changes.
• Yes so then sorry you had to wait so long for a response from our team.
• Yes so then sorry you felt left out during the discussion; we value your input.
• Yes so then sorry you missed the deadline; we can discuss an extension.
• Yes so then sorry you encountered issues with the software; let me help you troubleshoot.

[elijah-potter]

I'm happy to continue looking at this. Would it be possible to get the full text sentences that are causing the errors in the screenshots?

[hippietrail]

I'm happy to continue looking at this. Would it be possible to get the full text sentences that are causing the errors in the screenshots?

They're often partial sentences from YouTube transcripts processed to turn each line into a sentence to prevent the 'this sentence has too many words' lint overwhelming all the little problems I'm looking for. Often full of other distracting mistakes or even controversial topics.

I could start adding YouTube URLs with times to these, or I can find non-distracting equivalents / clean up each example. I often clean them up before adding them here to some degree already, but not always.

LLMs seem to be quite good at generating contexts for sentence fragments. I'll edit some of those in...