Tom Proctor

Building a historic site and archive from scratch for the legendary Record Plant studio in Sausalito. The role spans journalistic research and writing, photography, web design, data governance, and quality, all in one. It also includes Record Plant Radio, the streaming player built on the archive. Built mostly on Notion, Claude, and Airtable.

Relational database and research

• Nine-table relational Airtable archive of 450+ historic recording events (1972-2008), spanning broadcasts, sessions, albums, artists, tracks, sources, people, and rights.
• Claude-over-MCP pipeline reconciling entries against nine public archives, with logged provenance and audit-tracked cleanup runs.
• 87% of recording events verified against independent or primary sources, with 37 held as unconfirmed rather than resolved by assumption.
• Defined a reusable model for AI-assisted archival reconciliation.

Public site, journal articles, and streaming radio

The public archive site brings together provenance-verified recordings, original research, journal articles, and photography. Record Plant Radio presents the archive through streaming radio players; the site structure, layouts, articles, original photography, and publishing workflow are part of the same project.

The 2200 Studios archive is a relational Airtable database documenting every verifiable recording event at a single building — 2200 Bridgeway, Sausalito, home of Record Plant Sausalito (1972–1986) and The Plant Studios (1986–2008). Fourteen tables, nine of them core, describe hundreds of recording events: studio sessions, live radio broadcasts, albums, artists, tracks, sources, people, and rights — published to a public archive at 2200studios.us.

The technical manual documents how the archive was built and how it stays trustworthy: the schema's growth from a flat list into a nine-table relational model, a tiered evidence system that qualifies every claim before it becomes a record, dated census passes, conflict-holding instead of conflict-hiding, and the write governance — approval gates, live-read rules, no bulk sweeps — that keeps LLM-assisted work from corrupting the data.

The archive project demonstrates the consulting-relevant craft: relational data architecture, evidence governance, information architecture, and building systems where AI handles retrieval and drafting while humans authorize every state change. The supporting manual is written for technical peers — database, archive, and IA practitioners — it is honest about what is verified, what is approximate, and what is still open.

• Fourteen-table Airtable base (nine core), 48-field Recordings hub, field-ID-keyed writes with reciprocal link fields.
• Tiered source ranking and seven-value verification status; disproven records retained as archive material, never deleted.
• Approval-gated pipeline — preview, approve, write, read-back — with certification-gated skills and a standing no-bulk-write rule.
• Single-record session state shared across tools, so Claude and ChatGPT sessions read the same project memory.

The Box Framework is the guiding concept behind every radio on this page. It is the methodology and thinking that produced Record Plant Radio, the KFOG player, and the radios to come — a way of building audio radios from the Internet Archive that treats the player as one repeatable system, not a string of one-off projects.

The point of view is the same one behind my Brand-Grade AI work: the human sets the constraints up front — the brief, the rights approach, the voice — and the machine produces inside them. Quality is governed, not hoped for. The Box Framework applies that discipline to audio; the whitepaper applies it to enterprise content.

The model

A radio is three separate pieces: the engine, the catalog, and the skin. The KFOG radio proved the engine carries over unchanged. Only the data, the filter buttons, the dial readout, the details panel, and the copy changed.

The Radio Brief

The Radio Brief is the only thing you write to start a new radio. It takes about ten minutes. It covers:

• The name and web address.
• The archive searches, and what to include and leave out.
• What one selection is — a broadcast, an episode, or a game.
• Up to four filter buttons.
• The dial readout.
• The skin.
• The rights approach.
• What's out of scope.

The eight-step pipeline

Sweep, shape, describe, determine rights, load, clone, skin, verify.

New in v3.0

Part L, the Radio Kit — the fast path for building any audio radio from the Internet Archive.