How the Record Plant Sausalito Archive Was Built

Every list tells a story, don't it? Anyone can post names and dates on a web page, and plenty do. I built this archive over many months, and the method kept changing under me. Even with the tools we have now, it is vexing work. This page tells that story, including the mistakes that taught me the most. Pushing a button still solves very little. But the tools arriving now, once you learn to steer them, change what one person can document.

The list starts with people

The archive rests on firsthand knowledge, never on casual guesswork. Martin Porter and David Goggin spent over ten years documenting this history. Their book Buzz Me In gathered interviews, photos, and primary accounts. The KSAN broadcast list comes from Raechel Donahue's own notebook, shared by Martin Porter. Tom Flye, the original chief engineer, anchors the studio session record. The 1993 Plant client newsletter, a framed original, named 25 sessions directly from the studio's own records. The Our Sausalito discography article contributed 89 entries, the largest single source in the archive.

A humble beginning

The first version held seventy-six names from earlier research files. I pulled them from old notes and scattered web lists. Those lists helped, but each one carried errors and gaps. Dates conflicted, names got misspelled, and studios got mixed up. A comprehensive list seemed far off. I needed a way to test every claim against real evidence if this was going to have a chance. There was a lot of cold, hard research ahead, a labor of love that could easily have stalled somewhere along the way.

One address, three namesakes

Every record in this archive has to resolve to one building: 2200 Bridgeway, Sausalito. Nothing else qualifies. That sounds simple until you read the credits.

Three studios shared the Record Plant name: New York, Los Angeles, and Sausalito. Liner notes often say "Record Plant" with no city. A Savoy Brown recording labeled Record Plant 1975 turned out to be the New York room, on a WLIR broadcast.

Sausalito also got credited as San Francisco. Discogs lists the Bob Marley Talkin' Blues sessions as "Record Plant, San Francisco." They happened at 2200 Bridgeway.

And Sausalito had other studios. Studio D, Harbour Sound, and White Rabbit all appear in the same liner notes as The Plant, along with Tres Virgos in nearby San Rafael. A Sausalito credit is not automatically a 2200 Bridgeway credit.

The mid-1980s rebrand adds one more wrinkle. Records from 1985 to 1987 can say Record Plant, The Plant, or Plant Recording Studios. All three mean the same rooms.

Building the engine

So I built custom AI tools wired straight into an Airtable database. Claude searches, disambiguates, collates (whips and purees), and compares at a scale that still startles me. Disambiguation means deciding when two listings describe the same event. A 1973 Montrose broadcast appeared under four different titles online. The tools match dates, venues, lineups, and reel numbers to merge them.

The division of labor is strict. The model retrieves, cross-references, and drafts. I authorize every change to the database. The rules further down exist to keep that line clean.

Not all sources are equal

Early on, I treated every upload as roughly equal evidence. That was the mistake. An Old and In The Way broadcast sat filed under Jerry Garcia rather than the band. A careless pass would have kept it that way.

I realized I needed to weigh provenance, not just collect links. That is when the etree standard changed how I work. etree.org is the community that set the rules for trading live recordings. Its culture demands documented lineage, from the source reel forward. You can see how a tape traveled from microphone to file. The Live Music Archive on archive.org carries much of that catalog. Adding etree lineage as a test was the biggest single leap in reliability.

What each status means

Every record carries one Verification Status. Here is what each one claims.

• Verified. Confirmed by a clear, credible source.
• From Primary Source. Confirmed from an original document or recording: a tape card, a station schedule, the studio's own client list.
• Unconfirmed. Plausible, but not independently corroborated.
• Needs Review. Evidence exists, and a specific question is still open.
• Conflicting Sources. Two credible sources disagree, and I have not reconciled them.

Sources are ranked before any status is assigned. Original documents and recordings outrank release databases. Release databases outrank aggregators and fan sites. A single secondhand credit never earns Verified on its own. Two records in the archive carry an explicit "not a Plant recording" finding for exactly that reason. One source claimed the credit. Wikipedia, AllMusic, and every Discogs pressing contradicted it.

The ladder every lead climbs

A new lead moves through fixed steps. None gets skipped.

• Discovery. The lead surfaces from a tape card, an aircheck, a broadcast schedule, an archive listing, a book, or an interview.
• Duplicate check. The lead is matched against existing records by title, artist, date, catalog number, and label. This runs first, not last.
• Preview. The proposed record is written out field by field, each claim tied to its source.
• Approval. Nothing is created or changed without my explicit go.
• One record at a time. The core record is created alone. Tracks, people, sources, and albums are linked afterward, in a separate reviewed pass.
• Read-back. Every change is re-read from the live database before the work counts as done.

Leads that fail are not thrown away. The record keeps the date, the sources checked, and the reason. A search that came back empty is a finding. Writing it down keeps the next pass from repeating it.

Where I look now

Different kinds of evidence live in different places, and each needs its own search.

Release credits. Discogs, AllMusic, MusicBrainz, and label pages prove a studio was involved. They are weak on which studio, and nearly useless on dates, since most give a year only.

Studio documents. Tape cards, session logs, the studio's 1990 to 1991 client list, and its 1993 client newsletter. These are primary. They carry reel numbers, tape speed, track counts, and per-take dates. They also record sessions that were never released and appear in no database anywhere.

Broadcast records. Raechel Donahue's handwritten notebook is the arbiter for the KSAN in-studio run from 1973 to 1978. Station schedules and dated airchecks give exact days. That is why broadcast dates here are precise and most session dates are not.

Audio. The etree world sits at the top: db.etree.org and the Live Music Archive. archive.org is the backbone. It hosts the Live Music Archive, mirrors the etree catalog, and preserves the recordings so links do not rot. Next come the Guitars101 collector trees, then YouTube, official reissues, setlist.fm, and newspaper listings. For each event I save the best source and a backup.

People. Direct contact with people who were in the room. That includes the KFOG promotion director who produced the Live From The Plant series. An interview list ranks surviving studio staff from the original era first.

The blind spot, and the fix

Many Internet Archive items carry a bare date as the title and no artist field at all. The artist and venue appear only in the description or the file names. Searching by artist name cannot find them. One Widespread Panic broadcast was missed roughly five times for exactly this reason.

Now I search by venue and location, then read each item's own description instead of trusting its title. A second trap sits inside that fix. Filtering searches to the audio category silently excludes etree, where the live broadcast tapes concentrate. Two of four finds in one sweep surfaced only because I ran the venue and location searches too.

Recording the texture

A bare confirmation was never enough for a real census. Each record captures what kind of recording it is: audience, soundboard, FM broadcast, video, or official release. I note quality too: lossless or lossy, complete or partial. A soundboard tape and a phone recording tell different stories. Those details let a listener judge the source before pressing play.

Audio that actually plays

A listen link here means playable audio, not a citation. Only two hosts qualify: archive.org and YouTube. Both play inside the archive's own player.

Reference sites never go in that slot. That means Discogs, MusicBrainz, AllMusic, Wikipedia, setlist.fm, and article pages. Store pages are out too, even on a permitted host. An auto-generated YouTube release upload counts as a store link.

Some good audio still fails the test. One fan site hosts a complete broadcast but offers no way to embed it. I logged it, rejected it as a host, and found an equivalent Internet Archive item instead. One vendor came off the list entirely. Shows are listed there, but the audio could not be reached in any form.

A link that loads is not the same as a link that plays the right show. Links are checked live, and I listen when identity is in question. One broadcast went in only after listening confirmed its trade-cassette lineage was good enough.

Catching the fakes

The web is full of confident uploads that quietly lie. One Butterfield item on the Internet Archive spliced in non-KSAN Boston tracks. A casual listener would never notice the seam. I flag splices, mislabels, and rebroadcasts directly in the record, so the archive does not inherit other people's errors.

When sources disagree

Conflicting evidence is held, never quietly resolved. The record keeps both claims, names what would settle the question, and stays flagged.

One 1974 KSAN broadcast has three dates. The record and a commercial listing say October 23. A sound-archive operator says September 11. A grey-market release says December 30. The notebook is the arbiter, and the date has not been changed.

A Commander Cody pair looked like one event logged twice. I caught the merge direction before executing it. The row I nearly deleted was the correctly dated one. Re-dating then showed they were two separate events.

Identity errors get the same care. Raechel Donahue, whose notebook anchors the broadcast list, is not Tom Donahue, the DJ heard introducing broadcasts on air. Mixing them up would corrupt both the sourcing and the on-air credit.

The rules that keep mistakes from spreading

Each of these rules was written after a specific failure.

• Count live, every time. No number on this site comes from memory or an old export.
• Never sweep. A good rule applied in bulk is still a bulk change. One bulk cleanup of audio links stripped six records and lost three source addresses for good. Now I find a replacement first and swap one record at a time.
• A finding is not an action. A wrong-venue link stayed live for two days because the fix was written into the notes and never carried out. In another case, two duplicate pairs had been flagged ten weeks before anyone acted.
• Be honest about dates. A date of January 1 means the source gives a year and nothing more. It is a placeholder, not a claim. Broadcast dates are not affected.
• Keep internal and public apart. Working notes never render on the site. Public summaries live in their own field.

When one instance of a defect turns up, the response is an audit of that pattern across the whole archive, not a single fix.

How the pieces fit

The archive runs on one Airtable database. Seven tables describe the history. The Recordings table holds one row per recording event, whether a show, a session, or a broadcast. The others cover artists, people, albums, tracks, sources, and rights.

The People table has a field I care about most: Studio Connection. It records what each person did at 2200 Bridgeway specifically. That is the archive's original contribution, kept separate from general biography. The Tracks table carries tape-card detail such as reel number, tape speed, and track count. Track count is diagnostic. A two-track reel usually means a mix or reference copy, not a master.

Separate tables run the process. Unverified leads wait in a quarantine table until they clear the bar. Thirty rows moved there in one audited pass. Another table logs every cleanup run. A third tracks articles, and it reports live status from the published page alone. An audit once found 51 articles marked published, with only 9 actually live. Another holds session state, so work carries over cleanly between sessions.

The database is the only place changes are made. An import step copies it to the store this website reads. Nothing reaches this page until that import runs, so a fix confirmed in the database is not yet a fix a visitor can see.

What you can see

The page shows you what I could confirm, entry by entry. A linked entry means I located an actual recording and checked it. An entry without a link is attested but not yet located. Those come from the notebook and primary accounts, and they are waiting on a tape. The Type label tells you whether an entry is a broadcast or a session.

What this archive cannot tell you yet

A reference that hides its weaknesses is not a reference. Here are this one's.

• Date precision is uneven. Broadcast dates are exact. Most studio session records carry a year-only placeholder.
• Provenance text varies in depth. A minority of records carry a hand-written statement of what happened at 2200 Bridgeway. The rest carry a line built from structured fields. That line is accurate but generic.
• Open questions stay visible. Records marked Needs Review, Conflicting Sources, or Unconfirmed appear with that status attached.
• Some shows have no playable source. Several KFOG broadcasts are documented, but no permitted host carries them. Repeated searches are logged in the records.
• Coverage stops where the evidence stops. Tape cards remain in private hands. Some reels are known to exist and have not been digitized.

A list that feeds itself

The archive keeps growing because research keeps surfacing new shows. One verified setlist often names a second broadcast I hadn't logged. A reissue's liner notes can point to two more dates. Each event's footprint names its neighbors, so the work feeds itself. I park every new lead in a discovery queue so nothing slips. Then I work the queue down, one full pass at a time. That's why the count climbed from seventy-six to more than four hundred.

Where the census stands

As of October 2026, the archive holds 454 records, 452 of them published to the public archive: 303 studio sessions, 142 live broadcasts, 5 mix-only credits, and 2 rehearsals. 61 recording events link to playable audio streams (65 recording URLs are on file, but 4 are Wolfgang's Vault links with no playable stream). Of the published entries, 394 are verified against independent or primary sources (87%), 37 remain unconfirmed, 18 need review, and 2 have conflicting sources.

Open to challenge

I built this page so the method stays open to scrutiny. You can check my sources and challenge any call I made. Every located recording links to the best source I found. If you spot an error, the contact form goes straight to me. A real census improves in public, and this one is built to.