Decisions log

Four months of building without a public launch. The team agreed to a hard July 1 cutoff — whatever is functional by end of June ships, everything else gets hidden (not deleted) and rolled out month by month. Feature inclusion requires yes to two questions: does it make HH's core service delivery better, and does it help a client own their marketing?

Build, Deliverables, Creative Studio, Tasks, and Inbox are hidden from the nav at launch. The pages and routes remain intact so features can be re-enabled without a redeploy. Tasks hidden specifically because HH/MM internal accounts have ~5,000 tasks and the panel is unusable at that volume.

To get a real lived experience of the Burrow before clients use it, HH's own Meta and Google Ads campaigns will run through the Autopilot feature. MM static ads nominated as the second test account. Both accounts use existing hardcoded client ID overrides (HH: c63e6cc5, MM: 858637a4).

No new tools have been delivered to clients in six months despite the team having ~20 production-ready agent skills. Packaging them as a plugin means clients get a single install. Active clients receive monthly plugin updates; updates stop when they leave. Part of a quarterly coaching initiative to help clients set up their own Claude knowledge bases.

Working with SAZ Digital to produce a 3-minute Sales Video Letter and 8–12 low-fi studio videos at ~$4k initial cost and $1.5k per subsequent batch. Static image ads run alongside. Lead qualification handled via a multi-stage form on the landing page. Target: $400/lead at ~50% conversion rate.

Organic social needs content pillars defined, a proper production process, and potentially a new page before it's worth pushing. Status quo on current organic output until October. Paid ads are the marketing focus for Q3.

Regional audiences lack access to in-person workshops and have different buying context. Capital city audiences are already exposed to competitors (Brandon et al.) running boardroom-style ads. HH's differentiator in that market is live data, one-on-one coaching, and duration parity at the same price point.

Blog topic determines the messaging for social and email that cycle. Keeps production coherent and reduces context-switching. Standalone service promotions and timely content slot in around the blog calendar. Social-first formats (low-fi video, Twitter screenshots) supplement but don't replace the blog as the source of record.

Email is Teddi's domain and she manages the blog pipeline. Josh takes organic social because he's defining the content pillars and driving the video production with SAZ. The split reflects who owns the output, not who contributes ideas.

Landing pages need social proof and the team has strong existing client relationships. Asking two scripted questions during an already-scheduled meeting is the path of least resistance. Josh to supply the two questions; Teddi to run the collection process in upcoming client meetings.

The Burrow portal's /api/hilda/chat calls Anthropic directly with MCP tools wired into the portal route. No NHH proxy. Two-stack parity (portal and NHH each maintaining tool wiring) is accepted as a known cost. NHH's /api/chat is orphaned and on the cleanup list along with the portal's streamHilda(). Drive and Sheets tools are excluded from the portal's tool set.

No build step, no bundler. Pure HTML/CSS/JS so migration into the portal is a copy job, not a rewrite.

HH resolves to c63e6cc5-333e-4033-af9f-4be578e9e26a and MM to 858637a4-7e23-4a73-9b72-5e5ea983f493 via BURROW_CLIENT_ID_OVERRIDES in NHH burrow-proposals.js (PR #125). All other clients resolve dynamically from the DB.

NHH/Cowork/Burrow delineation, training moved to a 3-session structure, LinkedIn and Content training retired, consulting and support separated, one-off advisory repriced as a lead tool, MM Autopilot June test month. Full detail on Byron Workshop.

The deliverables table has a unique constraint on (client_id, type, period_label). All inserts use INSERT ... ON CONFLICT ... DO UPDATE SET content = EXCLUDED.content, updated_at = NOW() so re-ingesting a report never duplicates rows.