Order beats pace. Every phase after the first improves one existing decision.

The organising property

Phases two and three add no new obligations, only better inputs.

Once the first phase is running, the operator has one recurring output to act on. Everything installed afterwards feeds that same output rather than creating a parallel one: cycle capacity turns an estimate of remaining room into a figure, volatility tooling establishes whether conditions support the intended management model, execution diagnostics explain why expectancy moved, and the structural and benchmark layers separate variance from decay before the operator responds to either. This is why the stack does not become proportionally heavier as it grows — the weekly decision count stays at one, and its accuracy rises. That property is unusual enough to be worth checking against intuition: most systems become heavier as they are completed, and this one becomes heavier only in reading time rather than in decisions to be made.

Why the floor comes first

The later phases have nothing to sharpen until a decision exists.

The four-surface loop is not merely the cheapest starting point, it is the prerequisite for the rest being useful at all. Volatility context improves a management model only if positions are being sized under an authorised pool. Execution diagnostics explain movement in an expectancy figure that has to be computed from an aggregated record. Benchmark comparison classifies a live path that must first be governed enough to resemble the modelled one. Installed before the loop closes, each of these produces a genuine, accurate, unusable reading — correct about an operation that is not yet being run the way the reading assumes.

The failure of the reverse order

Analytics-first adoption produces excellent diagnosis and unchanged behaviour.

The natural instinct is to start with whatever answers the question that felt most urgent at purchase, and that question is almost never about deployment. The result is an operator who can state their capture efficiency, their branch attribution, and their position against the envelope, while sizing exactly as they did before. Nothing about that outcome is a system fault — the instruments installed are doing their jobs precisely. What was never installed is the part that says no, and the operator reasonably concludes the system did not work, having run the half of it that was never going to change anything.

Where the phases can legitimately be reordered

The rule is about dependencies, not about a fixed calendar.

The sequence is derived from what each layer needs, which means it bends where the dependency is genuinely absent. An operator running trend branches with runners has a stronger case for pulling volatility tooling forward, because trail coefficient decides between capturing a fat-tail move and exiting into ordinary noise — that is a management decision affecting outcomes now, not an explanation of past ones. Equally, an operator whose expectancy is not in doubt can defer the scorecard. What does not bend is the floor: no reordering makes the analytical layers work before there is a governed decision for them to improve.

• Phase 1 establishes the decision; phases 2 and 3 only improve its accuracy.
• Analytics installed early are correct and describe an operation not yet being run.
• The order is a dependency graph, not a calendar — it bends where the dependency is absent.

The attention argument

A sustained partial stack beats an intermittent complete one.

Adoption pace is bounded by attention rather than by understanding, and exceeding that bound has an asymmetric cost. A stack that consumes more weekly time than the operator's circumstances support does not degrade gracefully — it gets abandoned during exactly the difficult weeks that produce the most informative evidence, which leaves gaps in the record precisely where gaps do the most damage. The phase structure exists so the commitment can be sized honestly at the outset and grown when there is evidence it is being sustained, rather than assumed at the maximum and quietly defaulted on. Sizing the commitment honestly at the outset is therefore a technical decision about record coverage, not a concession about ambition.

The key idea

There is one right first move, and it is the least interesting one available.

Everything about the sixteen-capability picture invites starting somewhere impressive, and the correct first move is a journal, a weekly aggregation, a gate and a throttle — four surfaces that produce a tier, a pool and a directive and nothing worth looking at. An operator who installs those and adds nothing for two months is further ahead than one who installed twelve surfaces in a fortnight, because the first has a closed loop generating clean evidence and the second has a great deal of accurate information and the same deployment behaviour they arrived with.