I pulled the frames from both posts into the folder where I keep anything I need to check before I believe it, and then I sat looking at a bench that cannot answer the question. Two people I follow reached the same conclusion without any of it: a Japanese account(@_halkyofu_) wrote that the footage feels a bit AI to him and that the timing of its arrival is suspicious, and @UAPWatchers asked outright whether he was the only one skeptical of the qtecqot videos. Both of them got there by eye. I wanted to get there by file, and the file is where the trouble starts.
The material is three YouTube uploads from a channel called qtecqot, presented as recovery footage of the entity nicknamed Skinny Bob plus a walkthrough of a craft interior. First upload 25th May, 2026, then 15th June, then one in the last week of July. The X account with the same handle was created in April 2026 and had around 28 followers when the clips started circulating. That much is checkable and settled.
What sits on the bench
- yt-dlp, to get the native file at native resolution and frame rate, because whether the uploads are genuinely 4:3 or a 16:9 crop wearing a vintage costume is the cheapest tell available(I rejected screen-grabbing off the X player, which re-encodes into 16:9 and destroys the exact measurement I want).
- ExifTool, for container and encoder metadata(rejected: commercial "is this AI" scoring APIs, which hand you a confidence percentage and no method, and which I would not put my name behind).
- A C2PA manifest check through contentcredentials.org or a local viewer(rejected: trusting YouTube's synthetic-content label on its own, since that label is self-declared by the uploader).
- Google's SynthID Detector, which only speaks for Google-family output(rejected: reading a negative as proof of human origin, something Google's own help page tells you not to do).
- ffmpeg frame extraction, for the grain test and the specular-highlight test(rejected: eyeballing at player resolution, which is where every one of these arguments goes to die).
- Reverse image search on keyframes, hunting a commercial film-damage overlay(rejected: nothing, this is the one check that actually broke the 2011 case).
What the bench is for, and where it stops working
The reason I keep these tools together is the 2011 originals. Four black-and-white clips went up from an anonymous account, ivan0135, in early to mid May 2011, framed as leaked film of a captured Grey-type entity said to be from Zeta Reticuli, covering roughly 1942 to 1969. Crowdsourced work collected at skinnybob.info and on Metabunk did real damage to the archival story: the film-damage and scratch layer was matched to a commercially sold stock overlay with copies online by about 2009, the on-screen timecode font was identified as Consolas, a typeface released in 2006~2007, and the degradation stack was consistent with Boris FX Sapphire plug-ins for After Effects. The detail I find most persuasive is the ordering. The aging effects also degrade the timecode overlay, which means they were applied last, on top of everything, the opposite of what a genuine filmed-then-transferred artifact does. Filter intensity also drops across the four uploads, so it is a taste decision by an author, not a property of a physical reel.
That is not the same as a solved case, and I will not pretend it is. Nobody has ever produced the underlying footage, the set, the puppet, the studio, or the maker. Fifteen years, no confession, no book, no monetisation, no documentary deal. Opinion splits along professional lines in a way that is itself interesting: people from CGI backgrounds read it as CGI, people from practical-effects backgrounds read it as animatronics or stop-motion. Both camps agree it was made, they disagree on how. The honest verdict is that it was processed with modern tools, its origin is unknown, and nobody has claimed it. One correction while I am here: the KGB attribution the Japanese post repeats as fact is an accretion. The clips referenced air force files, and the KGB frame appears to have grown out of the uploader's Russian-looking handle and the retellings after it.
Against that, qtecqot's uploads are a different evidentiary situation entirely, and the pipeline meets them at its weakest point. A YouTube re-upload strips most container metadata before I ever touch it, so ExifTool has little to read. C2PA is a provenance record, not a detector: a manifest that survives settles the question, an absent manifest settles nothing at all. SynthID covers Google's own generators and was extended into Search, Chrome, Lens and Gemini around Google I/O in May 2026, with Google saying it has watermarked over 100 billion images and videos, but open-weight video models embed nothing, and the consumer Sora product was retired on 26th April, 2026, right before qtecqot's upload window opens. As such, "it's Sora" is a lazy attribution, and Veo, Kling, Runway or a local open model are the more plausible candidates, all of which hand me a file with no signature in it.
So the tests that remain are physics tests on pixels, and the frames themselves tell you where to aim.
The other frame worth freezing is the landing site. Two figures at distance, a conifer treeline, a pair of dark cones repeated near a light cliff face, the underside rim of a very large disc curving across the top. Small repeated background objects, morphing foliage and gait plausibility at that scale are where current models still lose coherence over time, and none of that requires a watermark to evaluate.
What makes the aesthetic so effective is the thing the 2011 hoax handed to everyone who came after it. MIT Technology Review has flagged the fake-surveillance look as its own genre of AI slop, and detection people are blunt about the mechanism: when you ask for footage that looks like an old camera or a cheap security feed, the blur, the grain and the low resolution hide the artifacts that would otherwise expose it. "Grainy 1960s 8mm classified film" may be the most forgiving prompt style in existence, and it is precisely the style Skinny Bob established. Both of the bookmarked frames carry the burned-in lower-left timecode with a black block over the leading date characters. Real archival redaction over exactly the field that would let you date the reel is convenient for a maker and useless to a verifier.
Laundering by adjacency
The failure mode is not the clips, it is the packaging. @mario1973p, a community creator, cut the 2011 ivan0135 material and the 2026 qtecqot material into one chronological supercut, and most people, including the Japanese poster, almost certainly met the new footage inside that edit. Fifteen years of unresolved credibility transfers sideways onto uploads that are two months old, purely by sitting next to them on a timeline. What I find worth noting is that mario1973p himself says the recent material is AI-generated video mimicking the Skinny Bob film. He built the compilation and still says half of it is synthetic, which is testimony against his own work, though it is a well-informed community judgment and not a forensic report, and I have seen no frame-level analysis of these three files published by anyone.
The channel's own text is the part that decides it for me. The descriptions run as an alternate-reality game: numbered continuation releases out of eight, "source anonymity is maintained", "failsafe contract is preserved", and a status line saying 0135 location and status unclear as of 2026/04/21, incapacitated. That is a self-sealing device. It writes the new channel into the old mythos and supplies, in advance, the reason the original uploader can never confirm anything. @UAPWatchers reports qtecqot claiming prior contact with Ivan, and I could find no corroboration for that claim anywhere, which leaves an anonymous account vouching for itself via another anonymous account. Ray Santilli ran the same structure in 1995 with the Alien Autopsy film: degraded medium as alibi, anonymous ex-military cameraman as firewall, then a 2006 concession that what aired was a recreation. The template is old. What is NEW is that the latex and the lighting rig have been replaced by a text box.
My stake in this is not neutral and I am not going to pretend otherwise. I take sightings and the disclosure fight seriously, and 2026 has been an unusually real year for it: the PURSUE releases began on 8th May, 2026 with 162 declassified files covering more than 400 incidents, with further tranches on 22nd May and 10th July. Those are documents with registry numbers and named custody, which is what actual declassification looks like, and they are the record I want people arguing about. Synthetic leak footage arriving in the same window does the opposite of disclosure, it gives every official denial a free excuse and buries the paper under vibes. The EU AI Act's Article 50 marking obligations start on 2nd August, 2026 and YouTube's altered-content disclosure has been enforced since January 2026, so there is at least a rule to point at now, whether or not anyone points it.
One question for whoever runs qtecqot, since it is the only thing that would move this: will you post the three original files with container metadata intact, alongside whatever record you have of the contact with ivan0135 you say happened?