Cost at the unit · one outcome, priced line by line
Cost per outcome
Every one of the 1,750 outcome chains assembled from the week of 10 to 16 August 2026 is priced here one rate line at a time: the people who touched it at the loaded rate for their function, the model calls it made at the rate for their class, the platform seats and runtime the stages ran on, and the management and facilities overhead apportioned across the minutes. The outcome pages price a chain at one blended human rate and two flat system rates. These pages price each stage at the rate for the function that ran it, the model class it called and the system it ran in, then add the two components the outcome pages never carried at all: platform license and management overhead. The totals therefore differ, and the difference is a change of method rather than new spending. Both numbers are shown side by side wherever they appear.
What one outcome is made of
The five components of a fully loaded outcome, summed across the priced sample and divided down to one chain.
The outcome pages carry $30,762 for the same 1,750 chains. This page carries $57,118. The $26,357 difference is entirely method: people are priced at three loaded rates rather than one, model calls at three rates rather than one, and two components the outcome pages never carried at all — platform license at $13,921 and management overhead at $10,495 — are added here. No new spending happened between the two pages.
Loaded rates, license costs and overhead bases are the customer’s figures, and no customer figure has been supplied. Every rate on these pages is a modeled composite: a plausible number of the right order for a group of this size, not a number taken from a payroll run, a vendor contract or a general ledger. Until the real rates arrive, this wave demonstrates the instrument and proves nothing about the business case. That is the same statement the benefit pages make, and it is not softened here.
The evidence behind every dollar
Each cost line declares which of three things its rate is. The mix is a headline here, not a method note.
Cost per outcome, by service
8 named services, sorted by what one outcome costs fully loaded. Each row opens the line-by-line pricing for that service.
Two comparable teams, and why one costs more
Same function, weekly volumes within a factor of two of each other. The gap between them decomposes into four drivers that sum to it exactly.
The gap is $19.59 a unit. The largest single driver is what one person-handled unit costs, worth $17.19 of it. The four drivers sum to the gap exactly because they are a decomposition of it, not an explanation offered after the fact. Both sit inside Legal and run within a factor of two of each other on weekly volume. That is the whole test. It is not a claim that the two do identical work, and the four drivers are arithmetic on the published unit economics rather than an opinion about difficulty.
The gap is $16.68 a unit. The largest single driver is what one person-handled unit costs, worth $9.32 of it. The four drivers sum to the gap exactly because they are a decomposition of it, not an explanation offered after the fact. Both sit inside Procurement & Source-to-Pay and run within a factor of two of each other on weekly volume. That is the whole test. It is not a claim that the two do identical work, and the four drivers are arithmetic on the published unit economics rather than an opinion about difficulty.
The gap is $14.71 a unit. The largest single driver is what one person-handled unit costs, worth $10.10 of it. The four drivers sum to the gap exactly because they are a decomposition of it, not an explanation offered after the fact. Both sit inside Human Resources and run within a factor of two of each other on weekly volume. That is the whole test. It is not a claim that the two do identical work, and the four drivers are arithmetic on the published unit economics rather than an opinion about difficulty.
Inference, attributed to the decision it served
The estate reports $1.9m a week of inference. This is how much of it lands on a named outcome.
The gap between the inference the chains can account for and the inference bill the estate reports. It is not waste and it is not evidence of waste; it is the part of the bill this instrument cannot yet attach to anything a customer would recognize.
Four weekly totals that do not reconcile
The estate already publishes its cost to serve in three places. They disagree with each other, and none of them is what this page prices.
These are four different objects, not four attempts at one number. The towers count transactions, the sub-functions count units of their own choosing, the spend ledger counts invoices and accruals, and this page counts assembled outcomes — and one sale is documented three or four times across those views. They are shown together because a board slide that quotes one of them without the other three is quoting a number nobody can trace back to a decision. Reconciling them requires the customer’s own general ledger, and no connector is attached to one.
Actions
What is waiting on a person
Each count below is a cost line, a service or a published pair of numbers that cannot be defended as it stands. They are priced from a two percent sample of one week, so the population behind each one is roughly fifty times larger.
Defend a divisor on an apportioned line
A real invoice exists behind each of these, but the units it bought are not counted, so the unit cost is whatever the chosen divisor makes it.
Put a meter behind an estimated cost line
These lines carry $35k, 62.1 percent of the priced sample, on a rate nobody counted and a quantity nobody observed.
Answer why two comparable teams cost different amounts
Same function, volumes within a factor of two, and the wider pair still differs by 852 percent a unit. Each gap decomposes into four drivers that sum to it exactly.
Give a service a value denominator
A service that books no value anywhere in its chain has a cost per outcome and nothing to divide it into, so no cost per unit of value can be formed at all.
Explain the inference bill no outcome accounts for
$1.8m a week, 94 percent of the reported inference spend, is attributed to no assembled chain at all.
Operations
What this desk is allowed to start
A surface that only reports is not operable. This is the work this page can set in motion, and the bound it runs into.
Trigger and bound
This page can price one outcome instance line by line at the rate for the function, model class and system that actually ran it, declare on every line whether that rate is metered, apportioned or estimated, and set two comparable teams beside each other and decompose the gap between them. It cannot change a rate, move a line from one tier to another, or reconcile the three weekly totals the estate already publishes. Rates are owned in the rate card; the reconciliation is a finding, not a task this page can close.
Live observability
What the record shows right now
Every dollar in the priced sample by the evidence behind its rate, an exhaustive partition of $57k across 3 tiers: each of the 10,939 cost lines carries exactly one. Read the dollars against the lines, because 42.6 percent of the lines are metered and they carry 5.9 percent of the money.
Current distribution
57,118 dollars in the sample
Is policy and strategy coming to fruition
Whether the written intent is holding here
A fully loaded outcome costs $32.64, 1.86 times what the outcome pages carry, and 62.1 percent of that is estimated rather than measured.
Not holding on the record
The position this layer exists to test is that a weekly total is not something a function head can act on, and that the cost of one outcome is. The pricing bears that out on the instrument and indicts the evidence. Across 1,750 assembled chains the fully loaded cost of one outcome is $32.64 against the $17.58 the outcome pages show, because those pages priced one blended human rate and never carried platform license or management overhead at all. That is a change of method, not new spending, and the $26k difference is shown wherever both appear. The spread between services is real: Close to report costs $71.85 an outcome and Order to cash costs $15.00, a factor of 4.8. But the evidence under those numbers is thin. Only 5.9 percent of the dollars sit on a metered rate; 62.1 percent are estimated, and $25k of that is people priced at a loaded rate no payroll run has confirmed. Inference is 4.3 percent of what an outcome costs here, while the estate reports inference as 21 percent of its cost to serve and only 6.4 percent of that bill lands on any chain at all. $10k of overhead is apportioned on a per minute basis that is a choice rather than an observation. The instrument holds. The rate card does not, and it is the customer’s rates that would settle it.