The Measurement Desk
Skills · Subagent · Command · PDF
.claude/the-measurement-desk/zip
skills/06
├ attribution-reconciler
├ cfo-one-pager
├ experiment-designer
├ lift-readout
├ measurement-plan
├ mmm-brief
agents/01
├ results-verifier
commands/01
├ /experiment
FIELD-GUIDE.pdf9pp
English
One payment · No account · No subscription
- ✓ Download link in your inbox in seconds
- ✓ Secure card checkout · VAT handled
- ✓ 14-day money back, no questions
- ✓ Every future revision included
- ✓ Runs in Claude Code and claude.ai
Built for Claude first, our recommended agent. Skills follow the open Agent Skills standard (SKILL.md), so they also run in any agent that reads it; subagents and slash commands are Claude Code formats.
Cheaper inside a bundle
The Measurement Desk
What the numbers are worth, before finance asks
six skills·one subagent·one command·9-page guide
Every figure in your marketing reporting graded from asserted to experimental, so the quarter reaches finance in four columns: Spent, Shown, Inferred, Unknown.
Executive summary
- The problem
- Platform, last-click and house numbers each claim the same sales, and the quarter goes to finance as if all of them were measured. Most of it is allocation, correct by its own convention and silent on cause, and nobody marks which is which until finance does.
- What it does
- Six skills and a read-only results-verifier grade every figure from asserted to experimental, reconcile the attribution views without summing them, freeze test designs before spend, and re-derive any claimed lift from the raw data.
- Why take it
- When the budget is challenged, you have to say which numbers you proved. This hands finance a quarter that says it, the Unknown column included.
For marketing leads and analysts who have to say what their numbers are worth.
In the first week you will
- 01Run /experiment on the test you were about to commission and get a frozen design or the arithmetic showing it cannot be powered
- 02Put the platform, last-click and house numbers side by side with attribution-reconciler and read the overclaim as a figure
- 03Build the quarter with cfo-one-pager, the Unknown column stated as a figure rather than omitted
Contents of the Kit
- 01Six installable agent skills: measurement-plan, experiment-designer, attribution-reconciler, lift-readout, mmm-brief, cfo-one-pager
- 02The results-verifier subagent: re-derives a claimed lift from the supplied test data, reports every discrepancy, and never fills a missing cell with a guess. Read-only
- 03An /experiment slash command that turns one sentence into a frozen test design
- 04The Evidence Grade as a working scale: five grades from asserted to experimental, applied to every figure the kit emits, with what each grade licenses you to claim
- 05Blank templates for the measurement plan, the frozen test design and the four-column finance page
- 06A 9-page Field Guide on separating what you observed from what you allocated from what you caused, and plugin manifest included
One skill, as it ships
SKILL.md
measurement-plan
- Use when
- Use when a marketing team is setting its measurement approach for a year or a quarter, when someone asks what should be measured and how, when a new channel is added to the mix, or when a finance review asks what each number is worth. Grades current state from what is actually reported, never from what the stack could theoretically do.
- Requires
- The channel list · What is reported today per channel
- Will not
- Never invent a benchmark, an industry average, a typical conversion rate or a typical return · Never promote a grade because a meeting needs it there · Observed, attributed and incremental never share a column or a sentence
- Returns
- Measurement plan · {organisation or brand} · {period}
One of six skills in the kit · All of them install the same way
The contents, in full · six skills, one subagent, one command
- 01measurement-planOne row per channel: the decision the measurement informs, the method that answers it, the Evidence Grade it holds today, the grade worth reaching, the cadence, the owner and the review date. It closes with the two or three tests worth funding this year, ranked by the decision value they would settle rather than by spend.
- 02experiment-designerA frozen design document for a randomised holdout, a geo experiment, a switchback or an observational model with a control group: one falsifiable question, the primary metric chosen before launch, the minimum detectable effect supplied by you, and the contamination risks, invalidation list and reading rule fixed before anything is spent. When the available volume and duration cannot see that effect, the output is the sentence "this test cannot answer this question", not a smaller test.
- 03attribution-reconcilerThe same conversions under the platform windows, under last click and under your own model, side by side in three columns that are never summed. It runs the double-count check against your order ledger, reports the overclaim as a figure, and computes the incrementality-to-attribution ratio where you have a test result, with ratios that expire.
- 04lift-readoutA finished test read against the nine-line standard: design, test window, sample per arm, the pre-period parallel check, the primary metric, the point estimate, the confidence interval, what the test cannot tell you, and the grade. It ends in what the result licenses, in one sentence, and what it does not.
- 05mmm-briefThe brief for a mix-model vendor or an in-house team, written backwards from decisions: the questions the model must answer, the data inventory with its true granularity, the experiments available to calibrate it, the acceptance criteria, and the questions the model will not be asked to settle. It flags the conditions under which no model will help.
- 06cfo-one-pagerThe quarter in four columns. Spent, the money. Shown, the effects demonstrated at grade four or five. Inferred, the estimates at grade three with their assumptions. Unknown, the spend on which no causal evidence exists, stated as a figure rather than omitted, and followed by the tests that would move money out of it.
- 07results-verifier (subagent)Hand it the raw test data behind a claimed lift, from your own experiment or a vendor's study, and it re-derives the arithmetic: arm sizes, baseline, point estimate, interval where the data supports one. It reports its own numbers next to the claimed ones and lists every discrepancy, every missing input and every assumption it had to be given.
- 08/experimentThe slash command that turns one sentence about the test you are considering into a frozen experiment design.
A retail media network reports nine times return on a campaign. A properly designed holdout on the same campaign reports one point four. Neither number is a lie: they answer different questions, and the gap is the bill for sales the brand already had. The Measurement Manual argues that almost every number a marketing team reports is an allocation rather than a measurement. This kit is that discipline as tools, and nothing in it promotes a figure a grade because a meeting needs it there.
Pairs with
The Measurement Manual is the argument this kit runs: the three registers, the Evidence Grade, the four designs and the four-column quarter. Think Like a CMO explains why the master gauge on most dashboards is broken, and The Third Wave of Digital Marketing treats incrementality as the honest layer the retail media funnel is missing. The Analytics Desk decides what gets measured before it is built and stops where attribution begins; this kit starts there. The Reporting Engine reports the numbers correctly; this one says what they are worth. For a single retail media placement against one network, the Retail Media Desk’s incrementality-test-designer stays the closer tool, and this kit takes over at portfolio level.
Before you buy
01Do I need Claude Code?
No. Every skill uploads to claude.ai as a zip. Subagents and slash commands are the Claude Code extras.
02I am not a programmer.
Install is copying folders into a .claude directory. The Field Guide in each kit assumes nothing beyond that.
03Will it work in my agent?
Skills follow the open Agent Skills standard (SKILL.md), so any agent that reads it can run them. Built for Claude first, our recommended agent.
04Do the tools invent numbers?
Never. Every figure is supplied by you, confirmed by you, or labeled as an assumption in the output.
05How does it arrive?
Instantly. Pay by card and the download link is in your inbox within seconds, and on the confirmation screen.
06Is this a subscription?
No. One payment, lifetime access to your download, every future revision of what you bought included.
07What if it is not useful to me?
Write to info@conclusionmp.com within 14 days of purchase and you get a full refund. No questions, no forms.
Ready when you are
The Measurement Desk, $12.
Instant download, one payment, every future revision included. Or take The Complete Operation, thirty-one items for $99, and this kit works out at $3.19.