A missing number looks exactly like a low one. One event writer, one number per stage and a total beside each count is what makes a screen worth reading.
Twelve numbers in eleven places. Each stage can be counted somewhere, and no two of the counts agree, because each tool counts a different week. At a thirty-person software company the screen is a slide, rebuilt by hand on the morning of the meeting.
By hand today
“own the metrics, dashboards, and analysis that inform weekly pipeline reviews”
A job advert for a revenue operations role, HSI, read 8 September 2026
That is one line from an advert, and it is a whole job. Somebody is hired to go and get the numbers, because the numbers do not arrive anywhere on their own. The posting platform has one count, the form tool another, the CRM a third, and the calendar a fourth. Each is right about itself and none of them is about the engine.
So the screen gets made by hand, from four exports and a memory. It takes the morning, it is out of date by the afternoon, and last month’s version cannot be compared with it because the definitions moved in between.
The failure underneath is worse and quieter. A number can be missing entirely and look like a low number. ArkOne ran a paid campaign with no working analytics at all until a fix landed late, recorded at plans/linkedin-ads-build-day/LEARNINGS.md line 47. Nothing on the screen said the measurement was absent. It simply read as quiet.
What it costs you each week
Sixty-four dollars a month for the cheapest paid dashboard tier, with every data source billed on top of it. Drawing the screen is not the expensive part. Keeping the sources agreeing is.
Source: a third-party reading of Databox plan pricing, read 8 September 2026
Treat that figure as a third-party reading rather than a quote from the vendor, because that is what it is. The shape matters more than the exact number: a low monthly price for the canvas, then a charge per connected source, which is the part that grows as the engine grows. Self-hosted dashboards are free and connect to databases, not to your posting account or your ad platform, so the joins are still yours to write.
The cost you are already paying is the morning. Add up the time spent assembling the weekly screen, then ask which stages were on it. If capture, sequences and bookings each have a number but nothing joins them, the assembly is the job you are paying for.
The build
Claude Code gives you one writer and one screen. Every stage reports through one function, so one place can count all of them. Every number carries its total, and exactly two of them raise an alert.
Four rules make the screen possible, and they belong upstream of it. Name every stage. Give each stage exactly one number. Put all of them in one place. Alert on two.
The first rule is the one that saves you. Every event in the engine is written by one function, not by whichever tool happened to be open. On the ArkOne site every event goes through a single track() call with one writer to the data layer, at docs/analytics.md lines 10 to 11, and the tag container itself is versioned in the repository at line 35, so a definition that changes is a change somebody can read in a diff. Copy that shape and build your own. There is no demo to download.
Work in the folder that holds your site, your CRM access and the logs from the earlier chapters. Type:
List every stage of my engine, from the sentence about who I sell to
through to the forecast. Give each one exactly one number for this week
and show the total it came from. If a stage has no machine-readable
number, print it as missing and never as zero. Show last week beside
this week. Read events from one writer only and tell me if any stage
reports through a different path. Write it all to one page, weekly.md,
and append one row per week to weekly.csv.
What comes back is one page with a row per stage, each row carrying this week, last week and the total behind it. Underneath, a short list of stages reporting as missing. That list is the point of the first run.
Two habits keep the screen honest. Count the thing you can act on rather than the flattering aggregate: ArkOne records landing-page clicks rather than a blended click-through rate, because the blended figure moved for reasons nobody could act on, at LEARNINGS.md line 25. And show a total beside every count. That rule is written into the platform’s own read helpers, at src/lib/pagination.js lines 4 to 7 and 21 to 22, because a count without its total hides everything it left out.
Keep the weekly rows. A CSV appended once a week, with a few months kept, is what turns a screen into a trend, at docs/seo-runbook.md line 134. Analytics products expire their event-level history on their own schedule, so the copy you keep is the one you will still have next year.
What it feeds
A review that changes something. A screen nobody has to build on Monday is a review that can open on the decision, which is what the next chapter runs. It works only if nobody spends the first half of it agreeing what the numbers are.
Three numbers prove this stage. One number per stage, each with its total. Stages reporting as missing, which should reach zero. And the number of places the screen was assembled from, which should be one.
What stays with a person
Which two numbers wake you up. Everything can be shown. Only two things should be allowed to interrupt a day.
An alert on twelve numbers is an alert on nothing. Pick two: usually one at the top of the engine, where supply dries up first, and one at the point where you know money is lost. Write down why each was chosen and the level that trips it.
The other judgement is what a number means. A screen can show that replies fell. Only a person can say whether the market went quiet in August or the message stopped landing.
The tools you already pay for
Nothing new to buy.
One event writer is what every stage reports through.
CRM gives the counts per stage, each with a total.
A text file holds weekly.md now, and weekly.csv is kept.
Your analytics container is versioned with the site.
No dashboard product and no operations hire.
Twelve numbers, one screen. Then the review opens on the decision.
Ask: Which stage of your engine has no number at all this week?
This is chapter 11 of twelve, after forecasting without a spreadsheet; chapter 12 runs the review. The book, Building a real GTM engine with Claude Code and Codex, end to end, runs the whole engine in order, one chapter per stage. A free starter edition for firms under fifty people is coming through the 100k programme.