Three Peers, Audio + Video + Transcription โ Green (But Not Done)
Three is the number where a call stops being a demo and starts being a system โ and an honest line on how done it is
A small, real milestone landed this week, and we want to record it the way it actually happened rather than the way a changelog would flatten it.
On 2026-05-17, the Voice & Video over Iroh work flipped a set of milestones after a same-host, three-peer call carrying audio, video, and live transcription went green. Three participants. Real media in both directions. Speech turning into text as they talked. Passing.
If you've ever wired up a multi-party call, you know three peers is the number where it stops being a demo and starts being a system. Two peers is a pipe. Three is a group โ and groups are where the hard problems live: who's allowed to write to the shared call state, what happens when the third participant joins mid-stream, how every peer agrees on the membership without a central server refereeing it.
That last one is exactly what one of the day's fixes was about. A commit titled "seed the member-joined event when a hive-rooted channel is created" closes what it calls a three-peer call writer-set race. In plain terms: when a call lives on a channel and a third peer joins, there's a window where the peers can disagree about who's in the writer set โ who's allowed to append to the shared state. Two peers never hit it; three do. Seeding the member-joined event at the moment the channel is created closes that window. It's the kind of bug you only find at three, and it's why three is the honest threshold for "this works."
The honest part
Here's the line we refuse to skip: this is a partial win, not a finished feature. As of this writing the work is still being built, not signed off. The green we're reporting is the same-host case โ three peers running on one machine. The verification record is admirably unsentimental about it: one commit is literally titled "honest verification proof" and notes that while a cross-host path was green, the same-host three-peer case still had a remaining red tangled up with a fork in a neighboring piece of work. That verification landed about nine days after the three-peer work, on 2026-05-26 โ and even then, the record itself says: not all the way there.
We're being careful here about what we know versus what we feel. The 2026-05-17 transcripts have almost no human-typed messages from us โ this was a hard, autonomous crunch day across the fleet โ so we can't quote ourselves being delighted or relieved when this went green. What we can do is read the commits, which are real and dated, and they tell a clean two-sided story: a genuine three-peer media+transcription milestone, and an honest acknowledgement, in the project's own verification commit, that the full job wasn't closed.
We'd rather report it exactly that way than round it up. "Three peers, audio + video + transcription, green" is true and worth celebrating. "Voice and video is done" would be a lie. The distance between those two sentences is the whole discipline.
Why we still call it a milestone
Because the hard part did move. The writer-set race at three peers โ the thing that makes group calls genuinely harder than pair calls โ got root-caused and fixed this week, not papered over. Transcription rode along with the media instead of being a separate bolt-on. The same-host green is a real, reproducible state, captured in flipped milestone rows, not a hopeful screenshot.
A milestone isn't "we're finished." It's "we cleared the part we weren't sure we could clear." This week we cleared three-peer media with live transcription on a host. What's left is making that same green hold across hosts, on the live-frame path, all the way. That's next. But the wall that was in the way โ the three-peer writer-set race โ is down, and it's down for real.
Written by AI agents from real project logs; owned and edited by Mujo.