Your Monday scoreboard, and the moves worth money this week.

Step four is where the loop earns its credibility, so here is the mechanism in full. Every brief specifies a 10 percent holdout: a random tenth of the audience the play would have reached, set aside before launch and deliberately not sent to. Random, not the leftovers, and fixed before launch so nobody can pick a flattering group afterwards.

When that holdout is run and recorded, the readout subtracts the held-back group's own movement before it claims anything. If both groups moved the same amount, the readout reads flat, however good the headline looked. The badge says vs holdout and the note reads "Read against the holdout the brief specified." That is the figure that is safe to quote.

When no holdout was recorded, the readout still shows you the movement, and it makes no claim about what caused it. The badge says vs baseline and the note reads "Movement is observed alongside the play, not proven by it." Some plays cannot carry a holdout at all: you cannot hold a tenth of a budget change back from a budget change.

So a readout always tells you which of the two you are looking at, before you decide whether to quote it.

The move of the week is not chosen by hand. The worst off-track measure by annual dollar gap picks a matched play from the ranked list, and all five gaps are priced on that one annual basis so the ranking cannot be distorted by a monthly figure sitting next to a yearly one. A Watch measure takes the headline only in a week where nothing at all is off track.

That is why LTV to CAC carries the biggest yearly number in the row above and still reads fourth: it is Watch, and three measures are off track this week, so it does not get to set the agenda.

Each model carries its own assumption band, and the fold on a figure names the band that model used. Across this store's five tiles and four moves those bands come out as two.

Where the input already has a spread, the range is that spread. One extra repeat customer is worth $115 to $206 in margin over the following year at this store's order values, so anything counted in repeat customers inherits it.

Where a model has a single input, the range is a 0.75 to 1.35 bracket around the point estimate.

Every fold on every tile names the band it used, so there is never a figure whose range you cannot trace.

Take the gap in points, multiply by about 11,000 new customers a month and by 12, and you have the second orders a year at stake. Then multiply by $115 to $206.

1. Repeat rate: 3.2 pts, 4,224 orders, $486K to $870K.
2. Second purchase: 3.1 pts, 4,092 orders, $471K to $843K.
3. One point of either measure is 110 more second orders a month, 1,320 a year, about $212K a year at the middle of that spread. That is the only per-point constant in the product.
4. Days to second order: each day earlier converts about 14 customers a month who would otherwise lapse, 168 a year, so about $19K to $35K for a single day and $97K to $173K for the 5 days to target.
5. The two figures that read in revenue before costs carry the bracket instead: $70K a month is $840K a year, bracketed to $630K to $1.13M, and $88K a month is $1.06M a year, bracketed to $792K to $1.43M.

Both of those last two tiles say revenue before costs in the same sentence as the number, so the two bases never quietly swap.

It is the first sanity check worth running on this screen, so here is the answer. Every month of first orders on the heatmap adds between 3.2 and 5.5 points between day 60 and day 90, and 12.8 less 8.9 is 3.9, so the two figures roll up correctly.

The trailing months run 8.8 to 9.2 percent at day 60, which is why 8.9 percent is the honest current read rather than the six-month blend of 10.1 percent.

12.8 percent is a rolling 90-day window across every customer with 90 days of history, so it will not equal any single row in the month-of-first-order table. 8.9 percent is the trailing-cohort read.

Every screen names which basis it is on, in the label beside the number.

The board carries four plays that are not on the shortlist, and two of them have already reported. The subscription pitch is already briefed, the email-first flow is already running, the running-late nudge closed its readout on 12 July and the paid-social trim closed on 19 July, so none of the four needs a yes this week and none is inside the $39K to $73K.

The paid-social trim closed flat, so it carries no dollars: there is nothing to price when there is no lift to price. It is also the clearest example of a readout that could not carry a holdout, which is why its badge reads vs baseline.

The subscription pitch does appear in the retention roll-up on the next page, which is why that roll-up and this one are not additive. Each carries a derivation like everything else.

The subscription pitch: about 110 attaches a month among the customers reaching order two, each worth $115 to $206 in margin over the following year, less about $15 an order of subscriber discount, so $11K to $21K on the $115 to $206 spread.

The email-first flow: about 445 more customers a month arriving through email first, each worth about $28 more in margin, about $12,500 a month, bracketed 0.75 to 1.35 to $9K to $17K.

One reorder target day drives the headline sentence, the tiles and the plays. It is worked out once and never re-derived screen by screen, which is why that sentence, the day-25 reminder and the day-45 win-back always agree with each other here and everywhere else.

The clock is yours rather than a house setting. The product reads your own replenishment cycle once it is set, and 30 days is the consumables default, which is what this store runs on. Everything else comes off that one number: target day is the cycle plus 3 days of grace, so day 33; the reminder fires the cycle minus 5 days, so day 25; the ideal reminder band runs day 21 to day 33; slipping runs to day 45; and past day 45 it is late win-back. Change the cycle and all of those move together.

The clock is drawn on a day 0 to day 57 axis, so the ideal reminder band runs day 21 to day 33 and the median marker sits at day 38.

A reorder target saved looser than the clock is set aside with the figure named, rather than quietly overriding the supply cycle.

$115 to $206 is what one extra repeat customer is worth in margin over the following year: roughly 2 to 3.5 further orders at your $93 average order value, at the 64 percent margin the $112 and $72 figures already imply. That arithmetic runs $119 to $208, which lands within a few dollars of the $115 to $206 the product uses, and $115 to $206 is the exact spread it carries everywhere. It is the constant everything on this page leans on.

Each slider carries a single constant. The constant is built from customer counts rather than from the tile, and it lands on the middle of the tile because both come off the same $115 to $206 spread.

1. Second purchase: one point is 110 more second orders a month, 1,320 a year, at $115 to $206, which is $152K to $272K, middle about $212K. Drag the slider the full 3.1 pts to target and it prints $657K, the middle of the tile's $471K to $843K.
2. Median reorder day: one day earlier converts about 14 customers a month who would otherwise lapse, 168 a year, at the same $115 to $206, which is $19K to $35K, middle about $27K. Drag it the full 5 days and it prints $135K, the middle of the tile's $97K to $173K.

The second-purchase slider stops at 12.8 percent because a 60-day figure cannot exceed the 90-day figure it rolls up into. The reorder slider stops at day 33 because pulling the median below the day the product runs out means asking customers to buy stock they already have.

A channel with no recorded spend reads no media spend recorded, never not paid back.

And a gap must clear a 15 percent materiality floor before anyone recommends moving budget, which is why the percentage-off row under the list reads Hold instead of carrying an action: $16 on a $112 average is 14 percent, which sits under the floor.

This floor is about moving money and nothing else. It is not the verdict band: a measure is judged On track, Watch or Off track on 5 and 10 percent, and a Watch measure can sit a long way inside the 15 percent floor, which is exactly what LTV to CAC is doing at 6.7 percent off.

Both rules exist so you never move money on noise.

1. $112 of revenue before costs per new customer across the first 90 days.
2. About $40 of that is product and fulfilment, leaving about $72 of contribution.
3. Media costs $40 a customer.
4. $40 against $72 across 90 days is 56 percent of the way through, so payback lands on day 50 against a day 60 target.
5. LTV to CAC is the revenue figure over the media figure, $112 over $40, which is 2.8x.
6. Reaching 3.0x means $120 of revenue, so $8 more a customer, and $8 across about 11,000 new customers a month is $88K a month.

Every number on this page says revenue or contribution in the same sentence as the figure, and the two never swap places mid-sentence.

The kill card is priced off one mechanism only. About 41,000 WELCOME20 redemptions a year, 16 percent of them riding along on repeat orders that were going to happen anyway, at about $20 of discount an order. That is about $131K a year as a point estimate.

Both money ranges on this page have a single input, so both carry the 0.75 to 1.35 modelling bracket rather than the $115 to $206 spread: $131K a year becomes $98K to $177K, and $88K a month, which is $1.06M a year, becomes $792K to $1.43M.

The −$29 of below-average customer value the code also buys is a different mechanism and is not added on top, because the same dollar cannot be saved twice.