Huddle · today's agenda

What shipped — and the two decisions we need

Sunday, August 2, 2026 · 8:00 am · Paul · Tim · Melissa

A big build landed since Friday. Most of Tim's punch list is closed, the clone sites are real stores, and the engine now writes from research instead of templates. Today is mostly show and decide — two decisions need Tim's call.

The headline: the “generic output” problem is fixed. The engine reads your real interior pages and a live industry knowledge base, and writes from both. Judge it yourself on the magnet clones before we talk.
Part 1 — What shipped (10 min, screen share)
The clone sites are real stores. ferrospring.com and codedspring.com are now full distributor storefronts — the complete 38-product Polymagnet line, product pages, a real FAQ, shipping/returns/contact, and live Stripe checkout. Same standard as polymagnetics.com.
Content is grounded, not boilerplate. The engine follows internal links, reads up to 5 real pages, and pulls the site's actual sections and FAQ questions. Briefs now name your real categories (Demos & Kits, Alignment, Latch, Spring, Detent) and cross-link your real pages.
New: Industry Intelligence step. A ★ menu item showing the competitive field (K&J, Eclipse, Bunting, Master Magnetics, Dexter, Apex), the source worth citing (Correlated Magnetics Research), the behaviors buyers care about, and where the industry is heading. It also grounds every brief. Refreshes weekly.
New: one-click video page. 20 hand-picked, verified-embeddable videos on your subject, published as a /videos/ page. Live now at ferrospring.com/videos/.
Navigation is fixed. A ⌂ Main button on every screen, “My Domains” instead of “Portfolio,” you land on your dashboard instead of the sales page, a back-link on reports, and a labelled menu on narrow windows.
Roll back anything. Approving an on-page fix ships it live — and there's now a ↩ Roll back button that restores the page exactly. Verified byte-for-byte.
Cleaner and safer. Stray entries purged (error pages, blocked sites, an old prospect list) — the site list went from 425 to 258 real sites. Unreadable pages can no longer become “sites” at all. Plan-based site limits added; Tim and Melissa are uncapped as testers.
Also: wider auto-growing edit box · compact domain list with a detail popout · real Operations metrics · a weekly readiness digest · a full written spec at wholereach.com/specification.
Part 2 — Tim tests it live (15 min)
1

Run the engine on a magnet clone, together

the real test

Log in at automarketingengine.com/account/login, run ferrospring.com or codedspring.com, open ★ Industry Intel, then read a content brief end to end.

The question on the table: does it read like it actually knows coded magnets now — or is it still generic? That answer decides the next two weeks.

Part 3 — Decisions we need today
2

Is grounded-brief quality enough, or do we buy AI writing? Tim's call

decision 1 — the content ceiling

Today the engine produces a brief: the structure, the real FAQ questions, the competitor angle, the keyword and the outline — grounded in real research, assembled deterministically (no AI writing costs). The next step up is turning an approved brief into a finished long-form draft, which means a metered AI cost per article.

What we need from Tim: read one brief and say whether a writer could run with it as-is. If yes, we stay deterministic and cheap. If no, we scope the metered draft tier.

3

What do we point at the real polymagnet.com? Tim's call

decision 2 — the write path

The real polymagnet.com is on Shopify and rate-limits our full audit, which is exactly why we built the clones. Approving a fix ships live on the clones; it can't touch the real store. Options: keep proving on clones · add a Shopify app/API write path · or run read-only audits on the real store and apply fixes by hand.

What we need from Tim: which of those is worth building first.

Part 4 — Where we go next (10 min)
4

The written specification is done

read before or after

Full blueprint at wholereach.com/specification — what exists today (the 51-check scoring model, the deliverable catalog, the setup flow) clearly separated from what we'd build next. Skim Part F (making it easier to use) and Part G (getting recommended by AI assistants) and tell us what you'd reorder.

5

Top candidates for the next build

rank these with us
  • Approval at scale — batch review and trust tiers. Today there are 2,557 items waiting on approval across the network; the queue has become the bottleneck.
  • Deeper competitor analysis — score the competitors too, and show the gap side by side.
  • Measured results — connect analytics so we can prove traffic movement, not just readiness.
  • Faster first win — a new user ships one real fix in 90 seconds instead of meeting 18 menu items.

Parking lot

This page mirrors wholereach.com/huddle/ — edit either one and the other follows automatically.

All huddles

Sun Aug 2 · currentWhat shipped & the two decisions Sat Aug 1The upgrade roadmap Fri Jul 31Does it actually work? Tue Jul 14Kickoff ArchiveMeeting notes & transcripts