← Composing work packagesEssai 08 · 18 November 2026

Warranted reuse / receiving contracts

What must the next package accept?

Two work packages can offer exactly the same eventual outcomes, yet make very different demands on their neighbours. The difference is which options they keep open until the handover.

A fictional refuge design package can be handed over in review formats A, B or C. The receiving team is equipped to accept a chosen set of these formats. Its capabilities stay available until a matching handover occurs. Internal preparation is outside that team's control.

All options stay open

Every preparation leaves ABC available.

Any one format suffices.

Some pair stays open

Preparation leaves AB, AC or BC.

Any two formats suffice.

Choice is made early

Preparation leaves A, B or C alone.

All three are needed.

Each has possible handovers A, B and C. A list of possible outputs cannot tell the receiving team what it must be ready for.

Test the receiving promise ↓

Continue to Essai 09 — What must remain possible next?

Derive a contract, then challenge a replacement

Choose the possible ready sets: each row is one stable result of internal preparation. ABC means all three remain available together; three separate rows A, B, C mean an uncontrollable commitment has already occurred.

Live calculation only. Choices are not saved, sent or turned into project records. Reload or Reset restores the example. No real review formats, costs or engineering approvals are inferred.

Original package
Proposed replacement

Untick to test incomplete knowledge. A known refusing outcome still defeats the guarantee; a set with no known refusal is then inconclusive.

Receiving capabilities kept available

The receiver agrees any common format. It does not choose what the supplier has already ruled out.

Illustrative setup cost

Additive units, paid once per capability. Zero is allowed. These are invented comparison weights, not a cost estimate.

Use the same receiving capability for another package

Capabilities are reusable between these two completed handovers. This does not share consumable resources or promise simultaneous capacity. For both guarantees, the receiving set must meet every ready set from both packages.

The leap: turn behaviour into a receiving obligation

The earlier boundary lab preserves what a finite count model can do. The feedback lab distinguishes possible acceptance from a policy that guarantees it. Here the question changes: what is the least the neighbouring package must offer, and which replacements preserve that promise for every neighbour?

For a ready family F and receiving set E, success is guaranteed precisely when E meets every member of F. Such a set is a hitting set. Retain only the inclusion-minimal hitting sets, called the blocker b(F). They are alternative sufficient receiving contracts: choose any one, or a superset. Minimal means no capability can be removed; cheapest depends on the weights.

Readiness and receiving requirements determine one another at this resolution: b(b(F)) = min(F). Removing a ready set that contains another preserves every guarantee question, because the smaller set is harder to meet. It can lose possible outputs. For example, {A, AB} and {A} have the same guaranteed receiving contract, although only the former can hand over B. The lab keeps possible outputs visible separately.

Replacement is therefore a comparison of promises, not of identical diagrams. A replacement is safe for this particular guarantee if it accepts every receiving set already sufficient for the original. When it does not, the lab constructs the cheapest old recipient that can fail, and the exact new ready state that refuses it. Adding a second handover composes these obligations by taking the blocker of the union of their ready families.

Why these results hold

Operational meaning

The package starts with one token. An internal preparation transition chooses one ready state R∈F. For each a∈R there is a handover transition labelled a, consuming that ready token and producing Done. A receiver with set E synchronizes on labels in E. There are no other moves or loops. If R∩E is empty the run is stuck; otherwise every maximal continuation reaches Done in one step. Thus every maximal run succeeds exactly when ∀R∈F, R∩E≠∅. Infinite delay and unfair scheduling are outside this finite maximal-run model.

The exact contract

Every sufficient E contains a minimal sufficient subset, since the alphabet is finite. Hence b(F) determines all sufficient recipients. Superset ready states impose no stronger condition than the minimal states. For the converse roundtrip, a set R that contains no member of min(F) has a complement meeting every member of F. That complement contains a member of b(F) disjoint from R. Therefore R meets every blocker exactly when it contains an original minimal ready set. Taking minimal elements proves b(b(F))=min(F). A family containing the empty ready set ∅ records an actual deadlocked preparation, so no receiver can succeed in every run. A family with no listed states is instead an absent model: the interface reports unknown. Although the mathematical convention is b(∅)={∅}, the app does not turn missing information into vacuous success.

All-context replacement

G preserves F's guarantee iff each g∈G contains some f∈min(F). Sufficiency follows because any old recipient meets f and therefore g. For necessity, if no f is contained in g, the complement of g meets every f but misses g: it is a distinguishing old recipient. The implementation also enumerates all eight recipients to rank the cheapest concrete witness. If F was impossible for every recipient, preservation is vacuous; the page says so.

Composition and limits

A reusable receiver must pass each separate handover: H(F∪G)=H(F)∩H(G). Minimum sufficient sets are b(F∪G). This is an exact conjunction of one-shot guarantees, not a general composition theorem for interacting Petri nets. Internal preparation, complete readiness, agreed boundary meanings and willingness to take any shared format are assumptions. If those meanings are incompatible the model does not apply; if evidence is missing the result is conditional. Passing a count or readiness contract does not establish project suitability.

Sources, original construction and verification

Baez and Master, Open Petri Nets, sections 4 and 6, supplies the distinction between execution and existence of a reachable outcome. Our synchronizing one-shot receiver test is a separate construction; we do not identify it with their open-place composition.

De Nicola and Hennessy, Testing Equivalences for Processes (1984), section 1.2, distinguishes passing some run from passing every run and orders processes by the observers whose tests they pass. We use a finite, terminating, one-handover class of observers, not the full CCS theory.

Edmonds and Fulkerson, Bottleneck Extrema (1970), section 2, gives minimal-transversal blockers and their involution on clutters. The established duality supplies the lossless guarantee contract; this application to receiving work packages and the example weights are our construction.

An independent test enumerates maximal executions in a labelled transition system and checks every nonempty readiness family on this three-mode alphabet against every receiving set. Further tests compare every ordered pair of families for replacement, check blocker roundtrips, serial composition and known counterexamples. These are finite model checks and a general mathematical argument, not field validation.

Executable checks · Calculation source · Retained boundary lab · Retained feedback lab

What this changes for reuse

A recurring block now comes with a computable menu of sufficient receiving capabilities. An apparently cheaper internal commitment can shift cost onto the next package; preserving an extra option can remove that downstream burden. The result lets us expose and compare that transfer before choosing a block. It supplies no probability, deadline or permission to accept a real design.