Where the content comes from.

First-person accounts from people who were actually at the show — collected from public dead.net comment threads, Cornell alumni magazine archives, and direct submissions from attendees themselves. Every memory is quoted directly with quotation marksand attributed by name (or handle, if that's how the person posted).

These are the most distinctive material on the site and the piece we're most committed to growing. If you were at any of these shows, the contribute form on every show page goes straight to the queue. Submitted memories appear with your attribution, locked to the exact second they happened.

Cornell 5/8/77 is the only show currently with inline source URLs on every factual claim. Every sentence in show.md and songs.md ends in a bracketed [src: URL]tag pointing to the public source it came from — Wikipedia, dead.net, the Cornell alumni magazine, jerrygarcia.com, etc. There's also a separate SOURCES.md in the corpus listing every URL used.

Other shows are working toward this standard. Rolling out gradually as we have time to do it right.

show.md (overview, era, gear, world context) and songs.md(per-song notes) for Buffalo, Veneta, Winterland, Fillmore East, Nassau, Cow Palace, and San Bernardino were synthesized by Claude Sonnet 4.6 from a curated set of public sources — Wikipedia, dead.net show pages, jerrybase, archive.org tape comments, and contemporary reporting where available. Each show's show.md ends in a Sources section listing every URL the generator pulled from.

What this means honestly: the prose is grounded in real sources, but it's AI-written. We're not pretending otherwise. Most claims are accurate; some may be paraphrased in ways that drift from the original; a few may be wrong. If you spot an error, tell us — fixes go in the next deploy.

Every track on every show has a structured vocalist field(Bob, Jerry, Pigpen, Donna, Brent, etc., joined with "+" for shared leads) and an optional instruments fieldfor nonstandard instrumentation (Jerry on slide for Row Jimmy, Pigpen on Hammond B3, etc.). These were hand-set against canonical Grateful Dead lineups and Dead lore, then enforced via a build-time check so future shows can't deploy without them.

The chat is required to consult these tables before answering any question about who's singing or what someone is playing. It's the structural fix for the "Promised Land is Bob, not Jerry" class of mistake.

The AI "Tour Vet" chat operates under three rules:

1. Cite every claim. Every factual statement in a chat response ends in a small bracketed source — [show.md], [songs.md], [witness: <attribution>], [VOCALIST TABLE], [whisper: <id>]. Hover any pill to see the full source.
2. Quote when possible. When the corpus contains an answer in its own words, the chat quotes it directly with quotation marks instead of paraphrasing.
3. Refuse when silent.If the corpus doesn't support an answer, the chat says "I don't have that in the corpus" rather than falling back on general training-data knowledge of the Dead.

The chat doesn't replace listening or reading. It's a docent over a fixed corpus — useful for surfacing what the corpus already contains, transparent about its limits.

Wrong fact in a show note. Misattributed memory. Vocalist field that says Jerry where it should say Bob. Whisper that describes a moment that isn't actually there. Anything.

Use the feedback link below or in the footer of any show page. It goes straight to the maintainer. The witness memory already submitted by Soglassy at Swing Aud '77 is in the corpus because someone reported it; the same path is open for corrections.