[RFC] Info Request | Scene matching: understanding how stash decides if a scene matches or not

[Dark-Obsidian]

Scope

I am finding that a number of my scenes don't match when scraping in scene tagger, but then do show when searching... however when reviewing the match found by search, it shows that multiple phashes and ohashes matched the scene... so why did this not match when scraping?

Examples

Here is a recent example from Scene Tagger...

• Stash did NOT match the scene when scraping
• It DID find the scene when searching
• As you can see: The filename 100% matches the scene name... To be clear, I fully understand that scrape <> search, even so it just seems a bit... 'disappointing' that stash didn't suggest the scene when doing "Scrape All" and I had to manually "Search" the videos 1-by-1, or use a 3rd party extension like stg-annon's userscipt to do this in bulk.
• After searching for the scene, I then find out that the below... ✅ Duration matches ✅ Multiple pHashes match (dist: 5 + 7)

... given these multiple match points, I am curious why stash didn't match the scene when scraping, and only matched it when searching

Request / Discussion topic

• It would be useful to understand more about how stash matches a scene please... or more accurately, how it "decides a scene is a match or not", if someone can elaborate on this please that would be great
• Are there minimum number of data points that are needed for a match (e.g. a pHash < 5)?
• Are there any data points that veto/void a match (e.g. if scene duration + stashDB duration differ by more than X%)?

Reference Reading

None at this time.

[Dark-Obsidian]

Search can also be mis-leading / incorrect...

On an adjacent topic, I am also a bit perplexed with "Search" some times...

• I tried scraping a scene called Dangerous Games but no matches were found
• Searching found 10(+) matches, but none of them with the name dangerous games and from reviewing the suggested scenes info I could tell none of the 10 were my scene (see search results and end of comment)..

Leading to wasted time...

• As there was no scene just called Dangerous Games in the search results, I took this that the scene was not in stashDB, so...
• I googled for the scene and actors, found it was a Randy Blue video
• Searched Randy Blue for the video, and got the scene page
• Began manually populating the title, performers, tags, studio, url, release date, etc
• Manually added two new/empty performer entries (as they didn't exist in my stash previously)
• Submitted the scene to stashDB, only to THEN get a warning that it already existed... ... and yes, the existing entry on stashDB is the scene I wanted all along! 🙄

Ultimately...

When I search for Dangerous Games...

• Why does stash NOT include the scene Dangerous Games in the results?
• And yet DOES include scenes like: Morning Foot Worship and JOI Game!?

---

Appendix

Search results -- pasted here to avoid messing up flow with large image in middle...

[Flashy78]

I believe StashDB matches phashes with distance of 8, so your scene should have been found. If you Scrape by Fragment (not Scrape All), does it work?

Text search on StashDB (which is a different Github repo over at Stashbox) is not perfect. There's an existing PR to allow you to use quotations to search multiple words as a phrase instead of individual words, which could help in your case. But also there's just a 10 result limit in order to keep things performant, and those two words are very common.

I'd suggest jumping on Discord where folks can help you in real time to debug your issue.