They are private to Brett's Claude account — he shares them from the page's share menu. No login for you to create, no password to manage.
e6018e24, throw it away.
Three files in the repo recorded that as the live URL. It returns “artifact not found.” The 785ceb33 link above is the real one. All three files were corrected on 26 Aug — see the last section.
YouChews sells on four surfaces at once — our own store, Amazon, Walmart, and whatever paid traffic we buy on Meta and Google. Each of those platforms tells us a flattering story about itself.
Meta claims a sale if the buyer merely saw an ad. Amazon reports its own ad spend but not the organic sales riding alongside it. Google reports conversions but no cart activity at all. Add those numbers together and you get a company that looks twice its real size.
Mission Control exists to be the one page that is not lying. It pulls every platform into a single set of numbers computed the same way, so a decision gets made against the business rather than against whichever dashboard happened to be open. The Daily Scorecard is the same discipline in report form.
The number that matters most is MER — all revenue divided by all ad spend. Break-even for our store is 1.57×. Below that, a sale does not pay for the ad that caused it.
Working. Twelve data tables pull cleanly from Windsor.ai over a plain REST key — no connector logins, no OAuth to babysit. The pull takes about a minute, and a self-test refuses to publish if the numbers contradict each other.
Three real problems, and they are the job.
It was armed on 24 Aug as a 2-hourly job. That job was session-scoped — it lived inside one chat window and died when the window closed. Nobody noticed for two days. The data on this machine was 23 hours stale when I picked this up, and the live page was showing numbers four days old.
Underneath is a known, stubborn cause: this repo lives in ~/Downloads, and macOS forbids scheduled jobs from reading that folder. They fail silently with a permission error. The same wall killed the 5:30am report — which is why that report has been run by hand for several days rather than firing on its own.
The fix is already decided and half-done: the new Mac Mini gets the repo at ~/Beacon, not ~/Downloads. That single move makes scheduled jobs legal for the first time. Until then, treat both jobs as manual.
This is the “integrate the daily tracker into our KPIs” ask, and it is more specific than it sounds. Two feeds are pulled every single refresh and then dropped:
Also missing from the board though present in the daily tracker: Amazon units and order items, GA4 bounce rate, and the new-vs-repeat customer split at platform level.
Every tile is a dead end. You can see that cost per order is $12.26, and you cannot ask which campaign, which creative, which day. Answering that means opening Meta and re-deriving it by hand — which is exactly the work the dashboard was built to end.
“We need to do a better job of integrating all the data that's in the daily tracker into our KPI's and I need to have a better way to drill into the tiles. I want to see things as funnels and wanna have a rock solid foundation with all the data connectors and we'll build from there.”
Read that as: foundation → integration → drill-down → funnels. Brett named the foundation last but means it first — the rest is worthless on a feed that silently goes stale.
Make the refresh survive a closed laptop. Move the repo to ~/Beacon on the Mini, re-arm both jobs there, and prove it by leaving them alone for 48 hours. Then add the check that would have caught this: the page must display its own staleness loudly, so an old number announces itself instead of being read as today's.
Klaviyo into a real email band. Walmart into its own tile, and into profit once its fees are calibrated. Then a written map — every metric on the page, which system it came from, how it is computed — so two people never mean two things by “revenue.”
Every tile answers “compared to what?” and “which one?” Click cost per order, get it broken out by campaign and by creative. This is where Beacon's fingerprinting eventually plugs in: performance attributed to elements, not just ad IDs.
Today there is exactly one funnel and it covers the store only. Brett wants funnels as a way of seeing, per platform and per campaign: saw it → clicked → added to cart → started checkout → bought, with the drop-off named at each step.
On 23 Aug the funnel read “161% of carts carried on” — because cart-adds were counted by Google Analytics and orders by Shopify, two systems with different lag. A funnel must take every step from one source, or it will print nonsense. The self-test now checks for this; do not remove that check.
Brett is non-technical by choice. He does not run commands and does not want to. Plain English, define every term, no jargon — that standard is not a courtesy, it is the working agreement.
You are remote, so the split is:
| Kind of work | Whose |
|---|---|
| Hands on a machine, a delivery, a signature in a room | Brett's |
| Files, jobs and settings on the Mac | The Claude sessions' |
| Systems, connectors, vendors, research, follow-through | Yours |
The scarce thing is not execution — the Mac sessions are fast. The scarce thing is someone holding the thread across days, noticing that a job armed on Tuesday quietly stopped on Wednesday. That is the actual job here, and it is why this project is yours.
Two house rules that will save you time:
785ceb33 was dead and e6018e24 was its live replacement. A fetch proves the opposite. Likely cause: the 24 Aug republish ran under Brett's second Claude account, landing it in a gallery the main account cannot see.armed: NO, with the reason.The lesson worth keeping: everything above was written down confidently, and none of it was checked. Fetch the link before you send it.