An undocumented decomposition heuristic sets both the cost of running a control system and the earliest moment at which any problem inside it can surface, and almost no enterprise has decided either number deliberately.
Somebody sits down with a scope of work and decides how far to break it up. Two levels or five; packages of a fortnight or a quarter. It takes an afternoon, it is minuted nowhere, and it is treated as a formality performed before governance begins.
That afternoon fixes two numbers for the life of the initiative. The first is what the control system costs to run: every package must be described, estimated, owned, coded, scheduled, reported and formally closed. The second is the earliest date on which anything going wrong inside a package becomes evidence rather than opinion, because until a package finishes the only account of its condition comes from the person responsible for it. Neither number is written down, and the choice behind both rests on a rule of thumb whose discipline cannot agree what kind of rule it is.
The consequence is direct. When a programme reports trouble later than the business could have acted, the instinct is to question the reporting culture, the delivery team's candour or the assurance function. Frequently none is at fault: the information could not have arrived earlier, because the unit of measurement was larger than the window in which anything could still be done.
The Strategic Context
The heuristic sizes the lowest unit of work — the package carrying an estimate, an owner and a deliverable. In one body of teaching material for the project control discipline it appears three times in three incompatible statuses: as one of two rules called load-bearing walls of the discipline, expressly not guidelines and not preferences; as an informal heuristic from a single industry's practice; and as a construction rule at once qualified as a heuristic rather than an absolute, with the note that capital projects commonly work to a band five times wider at both ends.
A rule cannot be all three at once. That the three survive together in one corpus is the finding: the number setting an enterprise's detection time is not derived, not defended, not stable. It is inherited.
Then it is broken. One worked exemplar declares its structure compliant and states the range of its packages. Three of them carry no effort estimate at all — documentation deliverables, field left empty. Compliance with a numeric rule is asserted over units holding no number to test, in a document that elsewhere calls documentation the most under-estimated work there is.
Why Granularity Looks Like an Administrative Preference
Decomposition depth is treated as planning taste, argued between a planner who wants precision and a delivery lead who wants to be left alone. It never reaches an executive agenda, and the framing survives because the discipline itself describes both failure modes administratively — too fine and overhead swamps the project, too coarse and problems appear only at a package's end.
Neither outcome is administrative. The first is a running cost with a real annual figure. The second is a latency, and latency decides whether the enterprise still holds an option when the news arrives. An organisation that has never converted its decomposition rule into those figures has made not a planning choice but a capital allocation and a risk decision, silently, on a planner's authority.
Reframing the Issue
Granularity is the exchange rate between the cost of looking and the cost of finding out late, and every enterprise trades at a rate it never quoted.
The useful question is not how large a package should be, but how long the business can afford not to know. That interval — the recovery window — belongs to the business, not the work: the period in which parts can be expedited, a shift added, a subcontractor replaced, a slot met. Once it closes, information arrives as history.
Detection here means the interval before a problem inside a unit of work can register at all. It is not the latency of an indicator — how long a measure takes to move once the thing it measures has moved — treated separately in this series, nor depth of visibility, how far down a chain an enterprise sees. Those defects compound with this one; none substitutes for it.
The Two Numbers a Single Choice Sets
The cost of control
Halve the package, double the count. Each carries a recurring administrative cost: a dictionary entry with acceptance criteria, an estimate with a basis, an owner, a code, a schedule activity, a cost account, a status return every cycle, a closure. That per-package cost is knowable — a controls function can price it in an afternoon — and almost no enterprise has.
The surfacing interval
A package's elapsed duration is its effort divided by the people on it, and until it completes its condition is self-reported. Add the wait for the next reporting cycle and the lag before that report reaches someone with authority to respond. The sum is the earliest moment a problem inside the package can produce evidence rather than opinion. If it exceeds the recovery window, the control system is decorative for that branch, whatever it costs.
Why one size cannot be right for a whole plan
Recovery windows differ sharply between branches of one initiative, which implies deliberately uneven decomposition — fine where the window is short, coarse where it is long. The discipline's own doctrine forbids that, treating uneven depth as a defect and as evidence of lopsided planning attention. Here it suppresses the right answer, and an enterprise following it uniformly buys precision where it is worthless and forgoes it where it decides everything.
Decision Framework
The surfacing-interval calculation is run once per branch of the plan, before the structure is baselined.
Establish four figures per branch. The recovery window: how many days after a problem begins the enterprise can still change the outcome, stated by whoever carries the consequence, not by the project. The current surfacing interval, computed as above from observed durations and lags. The annual cost of carrying one package. And the daily exposure — the branch's burn rate plus downstream work already begun on the assumption it is sound.
Then three tests. Where the surfacing interval exceeds the recovery window, resize until it does not; that is not negotiable against overhead, because the alternative is paying for a control system nobody can act on. Where the interval is materially shorter, packages are too small and the surplus precision should be sold back as overhead. Where a package carries no effort estimate it is not a package, and no compliance claim may be made over it. Record the outcome as a written sizing decision naming the window, who supplied it and the date, so it can be revisited when the business changes rather than inherited as a number of unknown origin.
Two boundaries. Early surfacing is worth buying only if slack remains to absorb what surfaces, and whether any does is governed by a shared resource consumed without record — not covered here, because it belongs to [Related article: Float Is the Only Resource With No Owner]. And this prices detection for work already in the plan; whether anything is authorised to test that the plan holds all the work belongs to [Related article: Nothing Is Authorised to Test Whether the Plan Is Complete].
From Strategy to Execution
Immediate. Take the branch of your largest initiative with the shortest recovery window, measure its surfacing interval across the last three cycles, and put the two numbers side by side in front of the sponsor.
Medium term. Require every plan submitted for approval to state its sizing decision and the window behind it. Price one package, publish the figure, and use it in every argument about detail.
Long term. Move decomposition depth out of the planning function's discretion and into governance, as a branch-level decision with a named business owner. The separate question of the duration at which the work costs least overall — set by an overhead recovery rate chosen inside finance, not by the work — belongs to [Related article: The Cheapest Duration You Only Find When You Are Late].
Signals to Monitor
Watch for problems that arrive with a package's completion rather than during it; for packages with no estimate in a plan declared compliant; for uniform decomposition depth across branches of obviously different urgency; for status returns frozen at one percentage then moving sharply; and for recovery actions that expired before the report reached anyone able to authorise them.
Questions for the Leadership Team
- For each major branch of our largest initiative, how many days after a problem begins can we still change the outcome, and who supplied that number?
- What is our measured interval between a problem starting inside a package and its appearance in front of somebody able to act?
- What does one work package cost us per year to carry, and who has priced it?
- Which current plans contain packages with no effort estimate, and on what basis was compliance claimed over them?
- Where did our planning standard's sizing rule come from, and when was it last tested against a recovery window rather than precedent?
- In the last three initiatives that reported trouble late, was the information withheld, or could it not have existed earlier?
Closing Perspective
A chilled-goods processor rebuilding a pasteurisation and filling line has a recovery window set outside the project, by the intake calendar and a retailer's promotional slot. Miss it and the loss is a season of contracted volume, not a schedule variance; packages on that branch must be short enough that trouble surfaces while expediting or temporary co-packing remain available. A facilities services provider mobilising a contract sits at the other extreme: service levels recover next month, the work is repetitive, fine packages buy only administration. Both are hypothetical, and both ordinary.
The two should not decompose alike, and under the conventional rule they would. Whoever last set that rule in your organisation decided how late your business is willing to find out. If nobody can say who it was, the decision still stands — unowned, which is the one condition under which it will never be revisited.
About EraNorth Insights
EraNorth Insights publishes practical analysis on strategy, projects, operations, transformation and decision intelligence for professional and organisational use. About EraNorth.