Blair is BlueForest's central intelligence system. The business runs on tools that can't see each other — HubSpot (sales), Xero (money), Streamtime (time & jobs), plus our own apps. Blair's Librarian copies all of it, hourly, into one Postgres database (Supabase), where every record is stamped with the same join key: the Project ID (XXX-YYYY-NNN — Division + year + number, e.g. VID-2026-030). Because everything shares that key, questions that used to span three tabs and a spreadsheet become one query: quoted vs invoiced vs labor per project, revenue pace vs the True North goal, a client's full history.

The registry is the master list — one row per thing we track (a video project, an internal system, a campaign, a person), each with its Project ID. Divisions: VID client video · SYS internal systems · MKT marketing · EMP people. Exactly one piece of code is allowed to create Project IDs (mint_project_code in the database) — every app is a client of it, which is why there are never duplicate or malformed codes.

How data stays fresh: the Librarian clock runs hourly at :00 — Streamtime syncs, then triggers HubSpot and Xero. The footer of every page shows the last sync; the Overview's Librarian card shows per-source heartbeats. Nothing in Blair is hand-keyed except the weekly scorecard numbers and the monthly overhead bridge.

Overview — the daily read: revenue/sales/expense pace vs True North, the two key ratios (LCV:CAC, 30/30/20/20), mini summaries of every report, and open data findings. Sales — recent leads (the status dropdown writes back to HubSpot) and the open-quote funnel with days-since-last-contact. Systems — the SYS registry with build status, commands, and ideas; statuses are click-to-edit. My Time — hours by Project ID by week, billable split. Finance — invoiced YTD, receivables, per-project P&L, the overhead bridge (financial access required). Scorecard — the weekly grid, newest week on top; click any manual cell to enter a value. Video Status — the production board by lane; status notes and lanes are click-to-edit.

Editing rule of thumb: if it looks like a value on a grid or a badge, try clicking it — the Console is the editing surface now (the old Google Sheets are frozen). Edits write straight to the database and, where a source system owns the field (HubSpot lead status), to that system in the same click.

Asking questions: any Claude Code session in the Blair folder can answer business questions in plain English via the blair-analyst skill — it composes SQL against the same database, read-only by construction.

BuildStream is the content operation's tool — ideas → AV scripts → teleprompter → publish — and it's a native Blair tool: it writes its own tables (bs_*) in the same database, no connector needed. Its console lives at buildstream.blueforest.studio under the same Google sign-in.

The core rule: a Project ID = a message. Ideas carry no code (most ideas should die cheaply). A drafted script carries only its content number (BS-00N). When a script is approved for production — the commitment point — BuildStream mints the video its own MKT Project ID through Blair's minting authority, and from that moment the video exists in Blair like any client project: Streamtime hours, expenses, and any coded revenue roll up to it automatically. Blog posts, social cuts, and shorts made from that video are derivatives — they inherit the video's code, never their own.

Judging marketing as a whole: individual videos aren't profit centers; the content operation is. All content cost vs all content-derived revenue = the P&L summed over the MKT Division. Costs that belong to the whole operation rather than one video land on the standing overhead bucket (MKT-2026-001 · BuildStream — Content Overhead) so per-video numbers stay honest. Full rationale: Project IDs for Marketing Content (team GitHub).