✿ Unified morning report

You scheduled the six August emails on the afternoon of July 30. The cockpit kept showing all six as unsent drafts, because it was reading a receipt — a file written at the moment we pushed the emails and never touched again. The moment you press schedule, that file is out of date, and every page built on it starts telling you the opposite of what is true.

The risk was concrete: a page that says draft next to an email sending today invites you to go schedule it a second time, which is how you get the same email twice.

Now the status column comes from a read-only look at Klaviyo itself. Tonight's look, at 1:15 AM: five of the six sit in Queued without Recipients (Klaviyo's scheduled state) and the August 1 anniversary thank-you is in Sending. If that live look ever fails or goes stale, the page says status not read rather than guessing.

Read-only, by construction: the code that talks to Klaviyo can only ask questions — there is no create, schedule, or send anywhere in it.

• the live-status feed
• the cockpit generator

The 21 emails for August 13 through September 30 were pushed to Klaviyo on July 31. The cockpit showed none of them, and told you a new batch of writing was overdue.

Two small reasons. The page looked for receipt files with batch-push in the name, and the new receipt is called batch23-push — one hyphen away from being seen. And the two receipts spell the send date differently (suggested_send in one, send plus time in the other), so even if it had found the file, every date would have come back blank.

It now reads every push receipt and both spellings. It also separates two clocks that were being confused: booked (what you have actually scheduled — through August 11) and written (what exists as drafts — through September 30). The next batch of writing is not due until September 16. Scheduling is what is due now.

• your scheduling tick-list

You asked for a daily read of which ads have hit their gates. It now runs as part of the night shift and appears inside the cockpit's Ads drop-down: every ad that spent money in the last seven days, biggest spender first, with what it spent in three days, in seven days, and over its whole life.

Each ad is scored against the rulebook you approved, nothing invented: $54 spent with zero sales is a stop, $81 with under one sale is a stop, $150 with 150 clicks to the site and under 70 cents back per dollar is a stop, and $250 is the only place a real verdict is allowed (1.87 back per dollar or better means scale, 1.45 to 1.87 means run, below that means kill). The two catalog ad sets are judged only from July 27, when their evidence clock was reset.

Tonight's read, on data through July 31: 128 ads spent money this week. 11 owe a decision. 5 tripped a hard stop — all five have spent past $54 with not one sale between them ($59, $59, $63, $72, and $118). Two ads are past the $250 gate: I-Like-Silence at $292 with $1.76 back per dollar, which the rulebook calls RUN, and the flagship combo ad set at $5,964 with $2.09 back per dollar, which it calls SCALE.

This is a reading, not an action. Nothing was paused, changed, or spent. The decisions are yours at the August 5 sitting.

• the scan
• tonight's numbers

The cockpit and the state-of-play page both said: create five ad sets by hand at $8/day, named 73 through 77. Three things wrong with that sentence. Your instruction on July 30 was "$100/day, 5 x $20" — you explicitly rejected $8. Slot 73 is not available, because it is the Jimothy-03 audition you switched on that same day. And that leaves four to create, not five.

Both pages now read: four ad sets, $20 a day, paused, names 74 through 77.

The reason two separate pages made the same mistake is that they were both quoting an old planning file that the override never reached. That file is now stamped superseded at the top, with a pointer to the real numbers, so nothing quotes it again.

• the override you signed

None of these would have cost you money on their own; together they are how a set of documents stops being trustworthy.

They are written, fact-checked, and sitting in Klaviyo as drafts. Three of them (August 15, August 21, September 18) are marked amber because they wait on the A/B walkthrough we do together — August 15 first.

Open the tick-list in your browser; each line is one email, and ticking it grays it out so you can stop and come back.

• the tick-list

The oldest one still blocking work is number 16: whether I may create a safety file inside this workspace's protected settings folder. It closes a gap a reviewer found; nothing is at risk while it waits.

Numbers 11, 12 and 13 are the ten minutes in Ads Manager and the two catalog-experiment choices. Those unlock the catalog work whenever you have the time.

• the ledger

Push receipts have two different shapes. Backlog NS-016 assigns the schema fix to Iris for August 7, with a schema note and parser fixture as closure evidence.

The email front door only scopes the August 1–12 calendar window. Backlog NS-017 assigns the receipt-wide scope fix to Iris for August 3, closed only by an exact 27-ID comparison test.

• the night-shift backlog

Closed as backlog NS-015: the full .site tree now contains the cockpit and existing review boards, and scripts/morning-pipeline.ps1 refreshes that tree in the safe order. Deployment remains a separate authorization gate.

• the cockpit page (regenerated)

This is a read-only reading from the published threshold artifact. Tim still makes every decision and activates or pauses only in Meta.

Meta-attributed results from igl.db, opened in SQLite read-only mode.

The count comes from the newest canonical CS run report on disk. That run is marked incomplete, so blocked evidence stays visible on the board. Nothing sends itself; Tim reviews and sends replies.

Tim will comment on the review board card

One line of code changes it

I walk you through it click by click

Thresholds get written against your number

Part of today's ads discussion

Closes a latent gap a reviewer found; nothing is at risk meanwhile