∮

Sheaf Models for Scope Coherence

Local accounts, shared fields and the question of agreement

Project director view · An illustrated trial

Do these two package accounts disagree?

One package includes a utility diversion; another assumes it is complete before its work starts. Those statements may fit together. To tell, we need a shared boundary, a handover date and a rule for comparing them.

This trial explores a way of framing project scope: retain each local account, then ask exactly what must agree at its interfaces. The examples are authored illustrations; they do not calculate a result from your project data.

First compare the rule, then the values

Choose Scope boundary below, then Temporary works ownership. Read the interpretation against the two rows: different descriptions can agree; matching names can still leave acceptance unestablished.

Scroll the table sideways to compare all four columns.

Example overlap checks — supplied values, not live results
Shared fieldPackage APackage BQuestion to resolve
Possession windowFri 22:00–Sun 05:00Fri 20:00–Sun 04:00Same booked window, or separate requested work periods? Equality and containment require different checks.
Utility diversionIncluded in our scopeAssumed complete before we startThese can agree. Confirm the handover date, boundary and accountable owner.
Temporary works ownerPackage APackage AThe declarations agree; verify that the named owner has accepted the duty.

Choose a case to distinguish a disagreement from an unanswered question.

Mini-glossary
Local section
Data assigned to one part of the declared model.
Restriction map
A specified map from package data to the fields represented at its interface.
Global section
An assignment compatible with all restrictions in that model, not proof of the real project's completeness.
Sheaf
A structure for relating local data and compatible gluing. The graph version needs data spaces and restriction maps; this page provides illustrations, not a general implementation.
Using a check in a review

Checks on declared safety constraints can flag inconsistencies; they do not assess safety or replace engineering assurance.

Repeatable consistency checks to support integration judgement.

Declare the fields, comparison rule, evidence source, owner and escalation route for each interface. A passing check is one input to a decision; it is not a safety case or authority to approve a gate.

A possible trial has four steps: select a small project slice; agree and structure its overlap attributes; compare the findings with the normal review; record useful findings, false alarms and the effort required. These are activities, not a promised four-week timetable.

Practical questions

Does this require AI? No. Declared rules can be checked deterministically. AI-assisted extraction would still need source checking.

What effort is needed? Access to permitted source data, an integration owner and someone to implement and maintain the agreed rules. The effort is not established by this toy example.

What can go wrong? Missing interfaces can produce reassuring silence; inconsistent meanings can produce false alarms. Evaluate coverage and interpretation alongside the findings.

Then inspect a larger set of interfaces

Compare agreement and disagreement at the interfaces

Select an authored example. The graph, counts and statuses are supplied illustrations, not computed checks of entered project data. Click any package or overlap line for its declared facts. For an executable selected-interface checker, open Scope coherence.

One comparison. Choose Corridor date clash, then select the corridor release line. Compare the two dates in the detail panel. Return to Coherent baseline and select the same line. The supplied cases make the shared-field question visible; their counts are not a computed diagnosis.

coherent overlap watch / immature contradiction

Scenario detail

Choose a scenario or click a package / interface in the graph.

What this example illustrates

These supplied cases distinguish incompatible values from uncertain or missing shared facts. Whether they block a real decision depends on the rule, evidence and accountable owner.

The mathematical framing

Turn the mathematics into delivery language

The practical question is whether package plans agree at their shared interfaces.

Sheaf = disciplined stitching of local scope views

Each team can keep its own detailed view. The model asks whether those views agree when projected onto shared interfaces.

Overlap = shared boundary where truth must match

The overlap is where two packages refer to the same corridor, asset, handover, readiness condition, or quantity.

Consistency rule = what each package must agree on

Dates, quantities, retained assets, consent conditions, and readiness prerequisites are typical early rules.

Global section = compatibility within the stated model

Agreement on the declared overlap fields is a consistency result for that model. It does not establish completeness, factual accuracy or permission to commit.

Contradiction heatmap = where to intervene first

The output shows where incompatible claims sit, who owns them, and which cost gates they threaten.

Explain scope coherence in project terms

Lead with “hidden contradictions across interfaces” and only reveal the formalism underneath when useful.

Formal version, stripped down:

For each work package Pᵢ attach data F(Pᵢ)
  e.g. boundary, dates, quantities, design assumptions, asset states.

For each overlap Pᵢ ∩ Pⱼ define shared fields and restriction maps.
For the stated incidence model, a compatible family of sections sᵢ satisfies
  rᵢⱼ(sᵢ) = rⱼᵢ(sⱼ)
on every overlap.

In plain English:
the selected package fields agree under the declared maps.
This is not a test of every fact, resource or approval in the project.

The foray asks how useful local project accounts might compose without hiding disagreements. This primer illustrates that question; its authored cases are not a full sheaf implementation. Follow the source trail and interpretation.

Further reading: director briefings, lifecycle questions and a possible trial
Project director view

Framing the question for a project review

A sheaf-inspired model treats each work package as a local view and asks whether its declared interface facts agree with neighbouring packages. Explore what that question could add to a review before design, procurement or cost commitment.

Hidden contradictions across packages Interface rules made explicit Questions for approval gates Pilot on live PMO data
Problem

Fragmented WBS views create contradictions nobody owns

Package plans can each look sensible while the combined scope still contradicts itself at interfaces.

Method

Relate local data through a project graph

Attach local data to packages and define what must match on the overlaps between them.

Mechanism

Enforce agreement where packages overlap

If Civils and Utilities share a corridor, the corridor facts must reconcile rather than merely coexist in separate files.

Result

Locate the illustrated disagreement

The authored graph highlights selected package statements that cannot all hold under the stated interface rules.

Possible application

What would a trial on project records require?

Start narrow: one station box, corridor segment, reservoir zone, or package cluster with real interface pressure.

Lifecycle questions

Where does this help in a real infrastructure programme?

Use the tabs to retune the framing. The same logic can support option development, business-case confidence, procurement interface clarity, live delivery control, and assurance.

If trying this with project records

Design a bounded trial

Choose one interface-rich slice and permitted PMO artefacts. Compare the proposed checks with the existing review to find where they add information, miss a concern or produce a false alarm.

WBS and package list

Useful for defining nodes, owners, and decomposition — but not sufficient on its own for truth at overlaps.

Interface register

Usually the best starting point. Each interface becomes an overlap candidate with named owners and shared facts to test.

Master schedule and key dates

Dates are often the quickest contradictions to surface because different packages silently rely on different readiness assumptions.

Assumptions log and design basis

Assumptions are local truths until the model forces them onto shared interfaces.

Cost plan / estimate basis

Link costed items back to the same overlap facts so the output becomes decision-relevant rather than conceptually interesting.

Asset / GIS / BIM / drawing boundaries

Spatial or asset definitions help when the contradiction is literally a boundary or retained-asset problem.

Choose a slice with a known interface question

Choose 8–12 packages around one interface-rich area where the PMO already suspects coordination friction.

Extract the local truths each team already uses

Dates, quantities, assumptions, boundaries, access rules, and readiness conditions usually exist already — just scattered.

Define a small set of decision-relevant overlap rules

Start with same corridor release date, same retained asset owner, same quantity tolerance, same consent window, or same readiness prerequisite.

Run a contradiction review before the next commitment gate

The pilot succeeds if it changes a real decision: interface clarity, estimate assumption, stage-gate confidence, or escalation sequence.

UK public-sector framing: business cases, assurance and PMO data are one possible context for examining these interface questions. The trial does not depend on adopting that context.

Limits of this representation

What the model can and cannot establish

This primer illustrates how declared interface facts can be compared. A trial would need to establish whether those comparisons improve an actual review.

It does not replace project governance

It helps governance by making contradictions explicit. Someone still needs to decide, reconcile, own, and approve the resulting changes.

It only checks the rules you encode

A poor rule set gives false comfort. Start with a few commercially material overlaps rather than trying to encode the universe badly.

It is not a digital twin by itself

The model checks coherence of declared scope facts; it does not by itself simulate physical behaviour or operational performance.

Evaluate a bounded trial

A modest pilot on one difficult interface zone is more persuasive than a grand architecture that never touches a live decision.