How do you measure logistics behavior? And do you trust it?

Logistics measures outcomes with precision. It does not follow that logistics measures the behavior of the decisions that produced them.

Spend, cost per shipment, cost per unit, service, utilization, transit, claims, and carrier mix describe the realized network. Executives need those measures. The error begins when an outcome is asked to establish whether the network chose well.

The ratio is not defective. It is incomplete for a behavioral claim. It collapses the freight entering the network, the economics attached to that freight, the valid options, and the selections made inside those options into one realized scalar.

Two populations can report the same cost per unit while selecting materially different options. Two periods can report different cost per unit while selecting at nearly the same position inside their available choices. A cheaper network can choose worse. A more expensive network can choose better. A flat line can conceal material drift.

Outcome movement is evidence that the realized network changed. It is not, by itself, evidence that decision behavior improved or deteriorated.

A decision is defined by the factual option and the complete set of valid options that surrounded it. Anything less is a partial account of the decision.

Let an exact analytical population contain rows r. Each row carries its own context: lane, contract applicability, time, physical facts, service requirements, governed Postoverlay economics, and available Intelligence evidence. Validity is local to that row's governing facts.

“Every valid option” does not mean every imaginable carrier, service, or rate. It means every option that can be proven for the row from governed source truth. No invented carrier. No fabricated contract. No synthetic rate. No option admitted because it makes an optimization look better.

Completeness is not an aesthetic preference. Omit one valid option and the rank ordering can change. Omit a class of valid options and the network can appear to choose well only because the comparison field was artificially narrowed. The option field must be earned before behavior can be claimed.

Materializing the option field is necessary. It is not sufficient. Every option must be ordered under every decision pole the assessment declares.

A Rank Basis defines one pole of optimality: base economics, Postoverlay landed cost, a named Intelligence measure, an Intelligence composite where valid, or the Raindex composite. Different poles answer different questions. Rank 1 under landed cost is not Rank 1 under transit performance, and neither is universally best outside its basis.

For option o in row r, dense rank under pole b is written as $\rho_b(o\mid\mathcal{D}_r)$. Equal values receive equal ranks; the next distinct value receives the next rank. Ties remain visible because they describe the topology of the valid field.

The method is deterministic because the objects and operations are explicit. It does not infer choice quality from a trend line.

INPUT  exact row population U, declared poles B

for each row r in U:
    D_r <- materialize_every_provably_valid_option(r)
    assert factual_option(r) in D_r

    for each pole b in B:
        rank every option o in D_r under b using dense rank
        record factual rank rho_b(f_r | D_r)
        record field depth max_o rho_b(o | D_r)

    R_r <- factual rank vector across all poles

OUTPUT exact-population behavior fingerprint Phi(U)

The row-level vector is the atomic behavioral object. It can be inspected alone or aggregated over any governed population without changing its definition. The complete network, one mode, one carrier population, one geography, one context intersection, or one row all resolve from the same comparison contract.

Average Rank is not sufficient without the field around it. A factual Average Rank of 2.2 inside an average field depth of 20 describes a different position from 2.2 inside a depth of 3. The factual rank tells us where the network chose. Counterfactual depth tells us what that position means.

An economic pole can also yield a basis-specific Counterfactual Floor when every row has complete, comparable economics. That floor is set-derived analytical evidence. It is not realized savings, a guarantee, or permission to introduce an option absent from the field.

Outcome charts repeatedly invite three narratives: improvement, stability, and deterioration. None is a behavioral conclusion until the decision field is computed.

1. Dip
   Cheaper does not prove better.

   A favorable mix can reduce cost while the factual choice moves backward across the same decision poles.

2. Plateau
   Flat does not prove stable.

   Opposing movements can cancel inside the aggregate while row-level choice behavior drifts materially.

3. Spike
   More expensive does not prove worse.

   The counterfactual field can become more expensive while the factual choice improves inside it.

The difficult part is not drawing a counterfactual chart. The difficult part is earning the right to call the counterfactual field real.

A behavior measure is trustworthy only when the source world, option population, ordering contract, and analytical identity are inspectable. A polished score does not repair an incomplete field.

1. 01
   Complete option materialization

   Every option that is provably valid for the row is present. Missing options can change the rank field.

2. 02
   One governed source world

   The factual option and every counterfactual option resolve from the same contracts, context, time, and evidence.

3. 03
   Stable option identity

   Each option preserves the carrier, contract, matrix identity, economics, evidence, and lineage that make it real.

4. 04
   Explicit poles and direction

   Every ordering states its Rank Basis and whether lower or higher values are preferred.

5. 05
   Preserved ties

   Dense rank keeps equal values equal. A tie is evidence, not noise to be broken for a cleaner chart.

6. 06
   Exact population and version

   The calculation binds one declared scope and exact version. It never substitutes remembered or latest state.

7. 07
   Visible weighting and coverage

   Unweighted and volume-weighted views answer different questions; missing evidence remains visible.

8. 08
   Fail-closed semantics

   If validity, comparability, or evidence cannot be proven, the field is incomplete and the behavioral claim is withheld.

TRUSTED(BEHAVIOR) =
    complete_option_field
    AND factual_membership
    AND explicit_rank_poles
    AND exact_scope_and_version
    AND preserved_ties
    AND declared_weighting
    AND visible_evidence_coverage

The assessment changes when outcomes and decision behavior are treated as separate, equally inspectable objects.

1. Outcome
   What happened?

   Observed cost, service, utilization, transit, claims, and mix.

2. Decision field
   What was valid?

   Every option that can be proven for the exact row from governed source truth.

3. Factual position
   What was selected?

   The factual option located under every declared decision pole.

4. Behavior
   How did the network choose?

   The aggregate pattern of factual rank positions across an exact population.

The executive question is no longer only, “What happened to cost?” It becomes: What was possible for each row? Where did the factual choice sit across every relevant pole? Which populations changed position? Was the world harder, were the decisions worse, or did both move?

If an assessment cannot show the complete valid option field, the factual choice, the declared decision poles, and the exact population being compared, it is not measuring logistics behavior. It is interpreting outcomes.

An outcome is observed. Decision behavior is computed. Trust begins with the complete decision field.