Every criterion an organisation uses to sequence work pushes difficulty later, which systematically schedules the hardest and least reversible activities into the window where reversal is no longer available.
Two pieces of advice sit at the centre of delivery practice. Both are correct. They cannot both be followed.
The first describes the shape of a project's life: influence over the outcome is greatest at the beginning and declines, while the cost of changing anything rises as commitments harden. The instruction that follows is to resolve what matters early, because that window does not reopen.
The second is a set of rules for putting work in order. Sequence by dependency. Level against the resources actually available. Let the team reach its stride before the demanding work arrives. Protect the early stages from a failure that would stop the project before it has momentum. Each rule is defensible, and every one of them defers difficulty.
The discipline that trains delivery professionals teaches both, in the same body of material, and never puts them on the same page. Composed, they give an uncomfortable conclusion: applying good sequencing rules to a project whose influence decays produces a schedule in which the hardest and least reversible work happens exactly when the enterprise can least change anything and pays most for trying. That is not the mark of a poor schedule but the output of a competent one.
The point of no return is rarely a date anybody writes down. It is the moment after which every remaining option involves paying: the marine works are in, the old platform is off, the crews are demobilised, the announcement has been made. What lands after it was decided months earlier, by people applying rules nobody asked them to justify.
The Strategic Context
Sequencing has an unusual property among consequential decisions: it is universally delegated. The curves describing influence and cost of change are executive knowledge, repeated in board papers. The ordering of work happens several levels below, in a scheduling tool whose defaults embody exactly the rules above. The composition is therefore missing not only from the teaching material but from the organisation chart.
The gap widens with the unit of management. On a project, deferral moves difficulty within one schedule, where an attentive manager might notice it. On a programme, the hardest work is integration between projects, deferred by construction: it cannot start until there is something to integrate. At portfolio level the effect compounds, because scarce specialists go first to the most predictable initiative and reach the most novel one afterwards, so the least understood work lands when the capacity to respond is committed elsewhere.
Consider a hypothetical desalination plant for a regional water authority. Dependency says process proving follows plant completion. Resource availability says commissioning engineers arrive when construction releases the site. Team maturity says the operating crew will be at its best by then. Protecting the early stages says prove the simple modules first. Four rules, applied conscientiously and independently, put the one activity capable of invalidating the design — sustained operation on real feed water at design recovery — after the intake is built, the membranes are bought and the offtake agreement is signed.
How much an enterprise should invest in defining work before it starts is the economics of front-end definition, owned by Article 28 in this collection. The question here is later: given whatever definition has been bought, what do the sequencing rules do with it?
What Leaders Misread About a Sequence That Looks Rational
Four readings keep the composition invisible.
The first treats sequencing as a technical activity rather than a risk decision. A schedule is approved as a plan — dates, resources, dependencies — and never signed as a decision about when the enterprise will find things out.
The second mistakes a consequence for a law. Effort peaks late in most projects, and this is presented as a natural characteristic of project life. Part of it is arithmetic: construction and integration consume more hours than definition. Part of it is the sequencing rules putting the work there. The two are never separated, so the curve is accepted rather than examined.
The third confuses the risk of early failure with risk. Sequencing to protect the early stages reduces the chance of an early stoppage, which is real value, but it does not reduce the probability that the difficult work fails. It relocates the discovery to the point where failure is most expensive, converting a cheap failure into a costly one and recording the change as prudence.
The fourth is the belief that gates compensate. A gate tests whether evidence from the preceding stage exists; it cannot produce evidence about work not yet done, so a gate after the last reversible moment can only ratify. Some enterprises worsen this by declaring certain outcomes unacceptable in principle, which strips an entire family of responses from the work that has been deferred; the capital consequences of that commitment are owned by [Related article: A Zero-Tolerance Commitment Is a Capital Decision], and are not examined here.
Reframing the Issue
Every sequencing rule answers one question: when is it easiest to do this work? None answers a second: when is it cheapest to be wrong about it?
Those questions have different answers, and only the first is asked. The easiest time to commission a plant is after it is built; the cheapest time to be wrong about whether it makes water at design recovery is before the intake is committed. The gap between the two answers is the exposure the schedule creates, created silently, because the rule that produced it was applied correctly.
A schedule is therefore a risk allocation across time: it decides which uncertainties are retired while retirement is cheap and which are carried into the region where the only response left is money. Nothing here requires abandoning the four rules, only that their combined effect become visible to someone who can buy an exception.
Four Rules That All Point the Same Way
Dependency ordering
Work is sequenced so each activity follows what it genuinely requires. The rule is sound, and the most reliable engine of deferral in the set, because difficulty and dependency are correlated: activities that integrate many others, resolve interfaces, or prove the whole rather than a part sit downstream of everything they integrate. Ordering strictly by dependency guarantees that the work carrying most information about whether the concept holds is performed last.
Resource availability
A schedule built without regard to available resources is a fiction, and levelling it against real availability is basic competence. The consequence is that work requiring the scarcest people moves to where those people are free — and they are scarce precisely because they can do the hardest work. Levelling pushes hard work toward the end of the queue with the regularity of a sorting algorithm.
The learning curve
Teams are understood to pass through stages before performing well, an idea usually credited to Bruce Tuckman, and the inference drawn is that demanding work should wait until the team has reached its stride. The inference is reasonable and the effect matches the others: the most demanding activity is scheduled for the period of greatest team maturity, which is the latest period, and the one in which the enterprise can least afford what the team discovers.
Protection of the early stages
Early failure is disproportionately damaging: funding is withdrawn, sponsors lose authority, and the project is cancelled before it has produced anything. Sequencing to establish momentum — the straightforward modules, the low-risk sites, the friendly first customer — answers that political reality, and puts the work most likely to fail furthest from the moment at which failure would have been survivable.
Take a second hypothetical: a publisher migrating peer review and production to a new platform. Dependency places archive migration after the platform is built. Resource availability delays it further, because the few people who understand the legacy metadata are the ones running the current issue cycle. Team maturity argues for learning on new submissions first, and protecting the early stages for piloting on a small title with a forgiving readership. All four are correct, and between them they schedule the flagship archives, with their identifiers and indexing relationships, into the final window, after the legacy system is decommissioned and the vendor's specialists have gone.
That last detail matters. Where the deferred work is also work the enterprise has contracted out, deferral and dependence compound: it reaches the hardest moment holding neither the knowledge nor the people. What happens when a profit motive is interposed between an enterprise and one of its functions is owned by [Related article: When You Put a Margin Between the Enterprise and a Function]; this article stays with the sequencing decision that put the enterprise there.
Decision Framework
The instrument is the sequencing consequence check, run once at baseline approval and again at every re-baseline. It has six steps and takes a competent planner a week.
First, establish the last reversible moment for each major commitment: the date after which undoing costs more than completing. It is an economic point, not a gate — derive it by asking what each commitment would cost to reverse the day after it is made.
Second, list every activity scheduled after that moment which could still invalidate a design assumption, a rate assumption or an acceptance criterion. These are the post-reversal proofs; on most projects the list is short and unwelcome.
Third, name the sequencing rule that put each one there. If nobody can name it, the placement was inherited rather than decided, which is itself the finding.
Fourth, ask the substitution question: what is the cheapest partial proof of the same assumption that could run before the last reversible moment? A rig, a single train, a fortnight of real feed water, one high-value title migrated end to end. The check never demands that the activity move, only that a proof be bought.
Fifth, price the difference between the partial proof and the cost of being wrong afterwards. The second figure usually sits in the investment case, and its size generally ends the argument.
Sixth, apply the rule: any post-reversal proof without a funded partial proof earlier requires explicit acceptance by a named executive at baseline approval. A risk register entry is not acceptance; it is a record that somebody noticed.
| Sequencing rule | What it defers | Partial proof to buy early |
|---|---|---|
| Dependency ordering | Whole-system proving | A subscale demonstration of the interface |
| Resource availability | Work needing the scarcest specialists | Specialist days bought early against the most exposed assumption |
| Learning curve | The most demanding activity | A rehearsal on a representative sample, with the actual team |
| Protection of early stages | The work most likely to fail | A deliberately hard first case, run small |
The governance test is one line: a schedule baseline arrives with its list of post-reversal proofs and a name against each, or it does not arrive. This checks the criteria, not reversibility itself; reversibility as an asset with a decay schedule, and the inventory an enterprise would need of it, is owned by [Related article: Reversibility Is an Asset That Decays].
From Strategy to Execution
Immediately, run the check on the largest delivery in flight and circulate the list of post-reversal proofs to the sponsor. It requires no new data, and the list changes the conversation faster than any argument about it.
Over the next two quarters, make the check a condition of baseline approval and fund partial proofs from a named budget line the project manager cannot reallocate. Unfunded early proofs are cut first, because they buy information rather than progress, and only one of those shows in a status report. Change the planner's brief so the sequencing rationale for each major activity is recorded when the schedule is built, while the reasoning exists.
Over a year and beyond, treat the capacity to prove things early as an enterprise asset: rigs, pilot cells, sandboxes, representative data sets, specialists who can be pulled forward. Measure delivery on when uncertainty was retired as well as on when work was finished, since an organisation reporting only the second will keep producing schedules that defer the first.
Signals to Monitor
Watch for commissioning, integration or migration dated in the last tenth of successive schedules; that pattern is the composition operating. Watch for "we will know when we test it" in risk responses, which names a post-reversal proof without calling it one. Watch for specialists planned to arrive after the last reversible moment by design rather than accident. Watch for contingency drawn late and heavily, which shows uncertainty was carried rather than retired. And watch for late change requests clustering in one area of scope, which is almost always the area whose proof was deferred.
Questions for the Leadership Team
- On our largest delivery, what is the last date on which reversing costs less than completing, and who computed it?
- Which activities after that date could still invalidate a design assumption, a rate assumption or an acceptance criterion?
- For each of those, which sequencing rule put it there, and who applied it?
- What would the cheapest partial proof cost for each, and is it funded where the project cannot quietly reallocate it?
- On our last three completed projects, when was the largest uncertainty retired, relative to the point at which change became expensive?
- Who signs our schedule baselines as a risk decision rather than approving them as a plan?
Closing Perspective
A schedule is also a statement about when the enterprise will learn things, and the rules producing it are optimised entirely for the other reading: when work gets done. Nothing in a planner's tooling, training or brief asks what the ordering does to the cost of being wrong. The failure is never bad judgement; it is the absence of a question.
An enterprise that does not ask it will keep discovering its most expensive facts at the moment it can least act on them, and will keep recording those discoveries as misfortune rather than as the output of its own method. The choice at baseline approval is narrow and permanent: buy proof while proof is cheap, or buy remedies at the only price still on offer.
About EraNorth Insights
EraNorth Insights publishes practical analysis on strategy, projects, operations, transformation and decision intelligence for professional and organisational use. About EraNorth.