Unverified recording of an online conversation

If the claims are accurate: why this page exists

The recording is of a conversation between two people who appear to be part of the Doginals ecosystem. In it, the speaker labelled “Speaker A” below says he is a paid agent of Bark Meta, says he was involved, and says the October 26 2026 incident was planned in advance.

If those statements were accurate, they would be significant: they would contradict the reading, documented elsewhere in this archive, that the event is explained by a design flaw alone (see technical explainer and what we cannot know). They have not been verified. The speaker may be mistaken, exaggerating, joking or trolling (he says himself at several points that he is known as a troll), and the identity assignments in the next section have not been verified. This archive therefore neither adopts nor dismisses the statements. It publishes the recording and a transcript so that readers can check them against the public record.

The publisher is anonymous and publishes no contact address; see the corrections policy.

Who is speaking (identifications unverified)

• Speaker A is identified as YOSO (@JosephCoat). Speaker A is the speaker who says he was involved and who says he is a paid agent of Bark Meta. Everything this page attributes to Speaker A is what the speaker says in the audio. It is an unverified claim, not a finding about anyone.
• Speaker B is identified as BDoge (@BDoge101), the participant who questions Speaker A.
• “Jon” in the conversation is a person who says he lost a Doginal Dog in the incident. This is his own statement; he is not further identified on this page.
• Unverified. All of these identifications are unverified and are not confirmed by those named. This archive has not checked the statement about the lost Doginal Dog.
• Limits. These identifications are unverified. They have not been confirmed by any of those named, and this archive has not matched either voice to any person or account. An X account with a given name or handle existing does not show that the person speaking is that account’s owner. The labels “Speaker A” and “Speaker B” come from automatic voice clustering.
• Corrections. Anyone named above, or anyone else mentioned, may point to a public, verifiable source that shows an assignment or a statement attributed to them is wrong; see the corrections policy.

Listen

Compressed copy, 3 h 06 min 58 s, mono, Opus at 16 kbps (16 kHz), 21.0 MB. Compression is lossy, but speech is close to the quality of the full-rate audio: in a 10-minute test, an automatic transcript of this copy differed from a transcript of the uncompressed audio by about the same amount as two transcripts of the same uncompressed audio differ from each other (13.4% against a 12.7% baseline). Names and overlapping speech can still be unclear. The transcript is a machine transcript and contains errors too.

Some older Safari versions cannot play Ogg Opus; if the player does not start, download the file and use a media player. Link to a moment with #t=SECONDS, for example #t=1464 (00:24:24).

Summary of what is said (claims, not facts)

Labels “Speaker A” and “Speaker B” come from automatic voice clustering. The names shown next to them are unverified identifications (see who is speaking). In the context of the conversation, Speaker A is the speaker who says he works for Bark Meta and Speaker B is the person questioning him. Names, handles and project names are written as they were transcribed from the audio and may be misspelled. Each line below describes what a speaker says; the timestamps link to the moment in the audio.

• Speaker A says he is paid by “Bark” and mentions other people by name, including “Shibo”, “Chief” and “Zero” (examples at 00:04:06, 01:57:20, 02:02:11); he says Bark “sends me out” on missions, and describes what these people did or did not do as his own account. He says his job is to record, disrupt and infiltrate X Spaces and chats (01:50:27).
• Speaker A says the event was “planned prior”, that the plan had been for a later date (he mentions December and 2028), and that it was brought forward (00:24:24, 01:41:46, 01:46:36). Asked by Speaker B whether it was brought forward because of Speaker B and a person called “Jon” (see who is speaking; unverified), the next line is labelled Speaker A and answers “Yes” (02:35:30); that was a leading question and the label for that line is uncertain.
• Speaker A says there are about nine people involved, that there is one shared password rather than a proper multisignature setup, that he does not “call the shots”, and that his share is 25% (01:46:36, 01:35:28, 01:12:47).
• Speaker A says dogs were sent to people with large followings as a tactic and that some could be taken back (01:07:31, 02:19:23). He is transcribed saying that he told Shibo “just use what we already planned initially. Take them, rip the dogs out. Make sure that this shit is fucking solid and tight. We have it all in the fucking conditions on the site. Everything is done. Like there’s no, there will be no legal recourse” (01:57:20). These are the speaker’s own unverified claims, and nothing here has been checked against the site’s actual terms.
• Around 00:56:06 lines labelled as Speaker A (uncertain label) say “we’re gonna touch your friends” and “we will push it to them and take everything if you want to choose this option because I don’t want to threaten you”, in reply to Speaker B saying he will keep posting. The wording is garbled and the speaker split is uncertain.
• Later in the conversation Speaker A is transcribed saying he “didn’t do anything” and is “not a part of it” (02:54:27), and is transcribed asking him to “cut off” and later to “delete” the recording (03:03:28, 03:06:13).
• Throughout, Speaker A says he is a troll and that he has lied to other people; the recording therefore contains both statements that may be serious and statements that may be bluffing.

Comparison with the public record (what can and cannot be checked)

• The numbers “269” and “$300,000” are first said by Speaker B and later repeated by Speaker A. The on-chain record in this archive shows 264 inputs; see scale and value.
• Speaker B recites the times of the transactions (02:45:07). They match the publicly documented block confirmation times on the timeline, so they are not independent evidence.
• No wallet address, transaction ID or web address is spoken in the recording, so nothing in it can be matched directly to a transaction.
• Statements about who is paid, who holds keys, who is on the team, and what was planned are statements by one speaker whose identity is unverified, and cannot be verified from the audio.

Transcript (searchable)

Machine transcript (faster-whisper “small” model), expect a word error rate on the order of 10 to 15% and more on names and overlapping speech. Italic speaker labels with a question mark are low-confidence. The names shown next to the speaker labels are unverified identifications; the claims in each line are the speaker’s own. Greyed lines starting with “~” are low-confidence passages that had to be re-transcribed. Click a timestamp to play from that point.

Downloads and file hashes

• transcript.txt (plain text with timestamps)
• transcript.srt (subtitles)
• transcript.json (structured; start/end seconds, speaker label, uncertainty flags, speaker identification notes)
• recording-compressed.opus (21.0 MB, Ogg Opus, mono, 16 kbps, 16 kHz)
• SHA256SUMS.txt for the files on this page

The hashes let a third party confirm that a file they hold is the same file. They prove nothing about when or how the recording was made.

On-chain inscription

A hash-anchored inscription of the original recording or its hash may be added so that its existence at a point in time can be shown. Link: [on-chain inscription link: to be added]. An inscription would show only that the data existed at that time, not that the recording is authentic or that anything said in it is true.

Method and limits

• Transcript: automatic speech recognition by a small model; some stretches of speech were initially dropped by voice-activity filtering and were re-run (marked “~”). A few quiet stretches may still be incomplete.
• Speaker labels: automatic voice clustering into two groups. The names attached to the two groups are unverified identifications; no voice match to any named person or account was done, and those named have not confirmed them. About a third of lines have a low-confidence label, and one line can contain words from both speakers.
• Names: names, handles and project names appear as the software transcribed them and may be misspelled or mis-heard. They are the speakers’ words, not findings about those persons or projects. Private identifiers such as phone numbers, e-mail addresses, home addresses, government IDs and private keys would have been redacted; an automatic scan found none, and the scan cannot guarantee that none exist.
• Authenticity: not established. The file may be edited, incomplete, staged or misattributed.

To correct an error or a name, see the corrections policy.