GOONBOT TRACE / 01The other side ↗01 PUBLIC POST
A BoredSexy post
The investigation began with a public video post shared in a Discord thread. The question was simple: where had the footage come from?
FIELD NOTE 001 OCT 13, 2026 / VERIFIED TRACE
Every frame
leaves a trail.
One public video post. A trail through the media behind it. A visual match in an older gallery. This is what source tracing looks like when the evidence—not a guess—leads.
Follow the trace ↘A real investigation, with all imagery abstracted.
01 / POST
02 / FRAME
03 / MATCH
SOURCE SIGNAL
FOUND ↗
FOUND ↗
01 / THE CASE ONE POST, SIX CLUES
The October
13 trace.
Read the sequence as it happened. No faces or source links reproduced here.
02 MEDIA LAYER
The CDN video
Page data pointed to the post’s hosted video. The media supplied a concrete artifact to inspect, rather than a caption to speculate about.
03 VISUAL EVIDENCE
Frames across the clip
Still frames were sampled across the timeline. A distinctive scene became the visual query; the clip itself carried no visible digital source watermark.
04 ADJACENT EVIDENCE
Sibling thumbnails
Other uploads from the same account formed a contact sheet. Several bore Chinese-language leak-site and channel watermarks—a clue about the wider upload batch, not proof of this clip’s original host.
05 SEARCH
Reverse image search
A sampled frame went through visual search. After one index produced no useful match, another surfaced a visually matching gallery result.
06 PROVENANCE
A matching leak gallery
The result led to an older gallery with a matching image from the same shoot. That established a provenance lead and contradicted an AI-only assumption; it did not establish the clip’s first uploader or original publication.
CASE BASIS · Discord thread 1559498432857768007 · session 20261013_173004_8117dbe5 · 13 October 2026. Source imagery and identifying names withheld.
02 / THE ENGINE FROM ONE TRACE TO MANY
Make the search
systematic.
The case was a manual trace. At scale, the same evidence chain becomes a controlled visual-indexing pipeline.
⌗
SCRFD
Locate face regions in candidate frames. Detection is an input to comparison, not an identity claim.
◈
AdaFace
Encode permitted reference and candidate crops as embeddings, robust to varied image quality.
↗
FAISS
Search nearby vectors, then review candidates against source context, timestamps, and provenance.
03 / THE BOUNDARY EVIDENCE WITHOUT EXPOSURE
Find the origin.
Protect the person.
We don't need to republish sensitive material to show how a trace works. This demonstration uses abstract artwork, omits performer identities and leak destinations, and distinguishes an observed match from an unproven first source.
See Vanish: the rights-holder-first side of the engine ↗