Back to composing work packagesAll experiments in this enquiry

Project wiring diagrams · Essai 06

The same package,
a different assembly.

Six shelter units use the same package design. Is repeating the box enough to make the project modular?

Each unit needs a connection to its site and a crew to assemble it. The units can open separately, or the customer can require all of them before accepting anything. Hold each package’s internal work fixed; change the boundary around it.

A fictional, finite experiment · no project forecast

End

Make the consequences of composing reusable work packages inspectable.

Way

Separate the package, interface, resource instances and acceptance rule. Measure architecture before outcomes.

Means

Compare exact failure states and a small crew schedule. Challenge each apparent improvement with a control.

Before looking at performance

Three measurements, not one success score.

The architecture profile below is declared in advance. It is our operational definition for this toy assembly, not a validated general measure of modularity. The three numbers remain separate; neither cost nor output appears in them.

Standard interfaces

Share of units using the same site-connection contract. More is more standardised.

Shared crew pairs

Share of all unit pairs assigned the same crew by the displayed scheduling rule. More means more resource coupling.

Common exposure pairs

Share of unit pairs in one failure domain: 0% for independent states; 100% for an all-unit common state.

These dimensions do not measure interchangeability, maintainability, learning, supply-chain diversity or engineering fit. A successful outcome does not reclassify the architecture.

Try the construction

Keep the internals. Change the connections.

All units retain two time slots of internal work, eight cost units and a 90% marginal success chance. The invented numbers isolate mechanisms; they are not calibrated to construction data.

Assembly conditions

Standard connections and crews cannot exceed the number of units. Setup is charged only if at least one standard connection is used.

The opening example is ready.

Calculations are temporary in this tab. Nothing is saved or sent; reload restores the opening example. There is no record, file or History entry. Correct a value and calculate again.

Last calculated assembly

The current assemblyThe calculated connections and acceptance rule.

What is inside every unit?

Every unit is the same declared process: a crew performs two time slots of internal work at cost eight, then connects the unit. A standard connection adds one slot and one cost unit; a bespoke adapter adds three slots and three cost units. All adapters succeed by assumption. Changing the interface therefore changes effort here, not reliability.

The crew is held for the whole operation and returned at the end. Two crews with the same skill are two instances; sharing a crew does not copy it. No learning, rework, travel or failure-driven rescheduling is modelled.

Expected accepted units
Chance of no accepted output
Chance of accepting all units
Planned completion
Planned cost

All possible accepted quantities

Exact probabilities, not a sampled simulation. The same 90% success chance applies to each unit under both exposure choices.

One crew cannot be in two places

Unit order is fixed. Each goes to the earliest free crew, breaking ties by crew number. This is a declared greedy schedule, not an optimiser.

Calculated crew assignments
UnitCrewStartFinish

Force a failure. Inspect the boundary.

These are interventions, not another random draw. They use the last calculated assembly. A local intervention under common-only exposure is deliberately outside that model's probability support.

Controls that can overturn a simple story

Which claim survives the comparison?

Standardisation has an entry cost

Keep six units and three crews; set setup cost to 24. Compare six standard interfaces with zero. The standard assembly finishes sooner but costs more. This is a trade-off under the stated prices, not a free improvement.

More crews buy time

Compare one crew with six. Output probabilities stay fixed; planned time falls and capacity cost rises. Identical package types do not establish parallel resource availability.

Common exposure moves both tails

Under separate acceptance, common exposure preserves the expected total but raises the chance of getting nothing. Under all-or-none acceptance, it raises the expected accepted quantity relative to independent outcomes. A single “risk is lower” score would hide that reversal.

Under the picture

The composition rule supplies the consequence.

The diagram is an acyclic dependency view. Unit boxes retain identical internals; interface badges, crew assignments and the acceptance node carry the changed meaning. Dashed exposure links describe statistical dependence, not physical flows. They are not resource wires or two-way information arrows.

xᵢ ∈ {0,1} = success of unit i Contained output: Y = Σ xᵢ Complementary output: Y = n × ∏ xᵢ Independent exposure: P(x) = ∏ 0.9ˣⁱ × 0.1⁽¹⁻ˣⁱ⁾ Common exposure: P(all succeed) = 0.9; P(all fail) = 0.1

The two output functions accept the same vector and preserve different properties. The sum preserves each unit's contribution; the product makes the weakest outcome decisive. Our binary all-or-none rule is an intentionally severe limiting illustration of complementarity, not Kremer's complete production model.

Exact enumeration and a separate binomial formula agree for 2–8 units. Resource checks reject overlapping use of the same crew. These checks establish the finite model's arithmetic and conservation rules; they do not validate the assumed numbers, the architecture profile or engineering performance.

Source → concept → construction → limitation

Why these sources meet here.

Modularity as independence within common rules

Flyvbjerg, Budzier and Christodoulou, Towards a Theory of Modular Natives, draft 4.0, pp6–7, motivates asking whether repetition and independence coincide. Pages15 and27 explicitly identify the lack of a valid measure of degrees of modularity across the comparison project types. It is a preprint, not peer reviewed. Our declared profile responds to that measurement problem locally; it does not repair or replicate the empirical study, or establish a general scaling law.

Tasks whose outcomes are complementary

Michael Kremer, The O-Ring Theory of Economic Development (1993), pp551–552, introduces production in which mistakes across tasks can sharply reduce the value of the whole. It supplies the contrasting mechanism. The binary product, shelter scenario, common-exposure comparison and cost assumptions are our construction; no conclusion about worker matching, wages or development is reproduced here.

What remains open: would a measured interface or resource profile predict containment in real work, after controlling for technology, selection, repeated learning and coordination costs? This model makes that question sharper; it supplies no empirical answer.

Compare the earlier repeated-course essai, which follows conserved materials and accumulated wall state, and the drawing that forks, which changes revision agreement. Here the new object is the consequence of repetition under different assembly and exposure rules.