Methodology · Construction note 07 · widget M.1

Numbers that must not be added, and numbers that must not travel alone

Two opposite disciplines govern how figures go together here. Some numbers may never be added, since the addition smuggles in an assumption nobody has. Others may never appear alone, since on their own they read backwards.

Panel v1.9 Data anchor 21 Aug 2026 Widget M.1 · Manpower reproduction

The rule this widget carries

Rule: before combining figures, ask what the combination assumes that you do not know; before publishing one alone, ask whether it reverses meaning without its partner.

§1

The composite that would have been easy to build.

Event	Cost	Composition
Putting a man in the field	₽3.4–3.6 mn	Signing bonuses of about ₽2.05 mn — ₽400,000 federal plus regional payments averaging ₽1.65 mn — and six months of pay at ₽1.3–1.5 mn
Death	₽9–11 mn	The federal lump sum, insurance, and regional payments to the family
Wounding with disability	up to ₽3 mn + pension	Lump sum for injury, plus a pension for which no series has been collected

The obvious next step is one figure: cost per casualty, or cost per man-year of war. Quotable, comparable and wrong.

Any sum implies a ratio of deaths to woundings to men who serve and return, and that ratio is exactly what is not known = the composite would look like a measurement while resting on an assumption never stated, and being a single number it would be quoted without the assumption forever after.

What is published instead is the three figures side by side, each attached to the event it arises from. A reader who has a distribution can combine them and take responsibility for it. The panel does not supply one it lacks.

§2

An assumption stays visible only while the figure stands alone.

The first figure contains a choice: six months of pay rather than twelve or twenty-four. The horizon is justified by turnover — recruitment on the order of 360,000–400,000 a year against a force of about a million makes a half-year contract a conservative unit of account — and it is declared as an approximation rather than a measurement of average time in service.

Inside a composite the horizon would have become invisible: a cost-per-casualty number carries the same assumption, and nothing in the number tells the reader it is there. The discipline is not that assumptions are forbidden. It is that each figure should carry at most one, in view.

§3

Three summary quantities that answer different questions, and no index over them.

Quantity	Question	What it counts
The claim against revenue	How large the task is	A magnitude
Conversion mechanisms spent, of those known	How many moves have been used	Decisions — the input
The manifestation matrix	How far the manifestation has gone	Consequences — the output

A magnitude, a counter of decisions and a counter of consequences share no unit, thus averaging them produces a single figure that moves for three unrelated reasons and conceals which one moved. That is the opposite of what the panel is for.

The same refusal applies inside the matrix. A scalar of cells filled per quarter is not published, since cells are not equidistant and are opened by gates of different kinds. Speed is measured only within a circuit, in months between entry into adjacent depths, where the comparison is between like things. The order of circuits relative to one another is not established at all: the reader chooses the circuit and the depth, and the panel names neither.

§4

The mirror rule: figures that may not travel alone.

The price of a contract, with the manpower balance. Signing bonuses at an all-time high mean the recruitment mechanism still works and the regime is paying for it — rung 1 of the quiet mobilisation ladder. The terminal reading arrives when the price stops being raised, not when it rises = the same figure means opposite things on different rungs, so it is read only together with the balance of recruitment against losses.
Debt service to revenue, with the real deposit rate. At a suppressed rate and accelerating inflation the debt burden line can improve at exactly the moment the bill is being shifted onto banks and depositors.
The debt service ratio, with the placement figure. Placements that do not happen create no coupon obligations, so the ratio improves precisely because the market is closed.

Two real rates are handled the same way, and the instinct to pick one is what the rule exists against. The policy real rate is positive, the household real deposit rate is negative, both are correct and they describe different agents. The panel publishes both and declines to choose, stating only which of them sets the zone and why — a household does not see the policy rate and takes no decisions on it.

§5

Composites that do get built, and the two rules holding them.

The claim on the state perimeter is a composite of thirteen lines, so the discipline is not a blanket prohibition. It is admissible since the lines share a unit and a period, and since two rules apply.

Rebuilt from components at every version, never carried. A headline figure carried across four versions without a derived composition is the case this rule exists to prevent: when the composition was finally built, one component turned out to be roughly a third of what had been implied.
The disclosure table is never abridged. All thirteen lines are shown, since the reader adds up what is visible — a shortened table whose visible rows do not sum to the stated total is an error a calculator finds in two minutes.

The test between the two cases is short enough to keep in the head: components sharing a unit and a period may be summed, and the sum must be shown as its parts; components requiring an unknown weight to combine may not be summed at all.

§6

What the refusal costs.

It costs the headline. A single number travels, three numbers with their events named do not, and a reader who wants one figure will either leave or build their own. The panel accepts that, and the acceptance carries its own risk: a set of figures with no stated reason for staying apart is an invitation to combine them, and the combination will then be attributed to the panel.

Thus the reason is published beside the refusal in every case, rather than the refusal alone. Someone who builds the index anyway has been told what it conceals, and the responsibility moves with the number.

Both disciplines apply to any dashboard built from heterogeneous material.

the procedure, and the three ways it goes wrong

Country risk, supply chain exposure, portfolio stress, programme performance. Etc.

For every proposed composite, write down what it assumes: weights, a distribution of outcomes, a common unit, a shared period. Then ask whether you know any of them or are inventing them to make the sum possible.
If a weight has to be invented, publish the components side by side, each labelled with the event or the case it arises from, so a reader who does have the weights can combine them and own the result.
Where you do combine, show the sum as its parts and rebuild it each release. A composite carried forward is a figure nobody has checked since the version in which it was first computed.
Never abridge the table supporting a headline number. Readers add what they can see, and a table that does not foot is the cheapest possible way to lose the argument.
Find the figures that reverse meaning alone. The test: can this number improve because the underlying situation got worse — activity stopped, a market closed, a payer went silent? If yes, it has a partner.
Where two measures of the same thing are both correct for different agents, publish both. Choosing one hides the gap, and the gap is usually the finding.
Publish the reason for every refusal to aggregate, not just the refusal. It converts a gap in your product into a statement about the subject.

Three failure modes.

The composite that hides an assumption. The assumption is stated once, in the methodology, and then travels nowhere with the number. Within a month the figure is being quoted as a measurement, including by you.
The half of a pair that gets quoted. Any number that improves for the wrong reason will be extracted and used against you, most often by someone acting in good faith who found it on your own page.
The index built to make the work quotable. It solves a distribution problem — nobody reads three numbers — by creating a measurement problem, and the measurement problem is permanent while the distribution problem is not.

A number that cannot be added is telling you something about the world. An index built over it is telling you something about its author.

Construction notes

01 · Calibrating a scale on administrative action — widget 1.3
02 · Ordering by payer, and the rung nobody announced — widget 2.4
03 · Choosing the direction of a ratio, and who controls the scale — widget M.2
04 · Evidence weighted by regime, and a contaminated denominator — widget 1.2
05 · Scales that read consequence, not volume — widget 2.5 · M.3
06 · What may enter a financing identity — widget 1.1
07 · Numbers that must not be added, and numbers that must not travel alone — widget M.1

© Dmytro Voronenko, PhD, 2026 · ORCID 0009-0003-5118-302X · CC BY 4.0 · Panel v1.9 DOI: 10.5281/zenodo.22068421
Panel: https://incidence.world · Updates: https://notes.incidence.world

Construction note 07 · applies to panel v1.9, data anchor 21 Aug 2026 · figures are as of that anchor and are not restated when the panel updates.