Choosing Between Monolithic and Progressive Strategies by Scale

Matching the strategy to the scale

This is the evaluate-level capstone of the context-strategy task statement. Having the two strategies, the cost dynamics, and the implementation, you now decide which strategy fits a given situation and, crucially, recognise that the right answer can change over time. The rule is scale-driven: a small, stable, predictable set favours monolithic for its simplicity and single-round-trip latency; a large, growing, or highly variable set favours progressive discovery to control token cost and cut tool-selection distraction. The evaluation is reading the catalogue's size and trajectory and choosing accordingly.

The added dimension at this level is time. The decision is not made once and frozen; it must be revisited as the catalogue grows past the point where the original choice was justified. That temporal aspect is what makes this the capstone rather than a repeat of the foundation.

Small and stable favours monolithic

For a small, stable, predictable set, monolithic loading is the right choice, and defaulting to progressive here is a mistake. When the whole set fits cheaply in context and rarely changes, loading it all up front is simple to build and reason about, and it costs a single round trip per request with no discovery step. Progressive discovery's benefits, lower token cost and reduced selection distraction, barely apply to a set that is already small, while its cost, the discovery round trip and the added complexity, is pure overhead. So the simplicity of monolithic is a genuine advantage exactly when the set is small and stable.

The judgement is to recognise when 'small and stable' holds and not to over-engineer it with a discovery mechanism the scale does not warrant. Applying progressive discovery uniformly, even to a small fixed catalogue, is one of the traps.

Large, growing, or variable favours progressive

For a large, growing, or highly variable set, progressive discovery is the right choice, because that is where monolithic loading's costs bite. A large set makes the full-set-times-volume token cost substantial, and it enlarges the tool-selection candidate space, raising wrong-tool risk. A variable set means most of what is loaded is irrelevant to any given request, so carrying it all is especially wasteful. Progressive discovery addresses both by exposing only the relevant subset per step, trading one round trip for a leaner, more focused context. The bigger and more variable the set, the more that trade favours progressive.

So the two strategies map cleanly onto the scale axis: monolithic for the small-stable end, progressive for the large-variable end. The evaluation names which end the described catalogue sits at and picks accordingly.

Revisit as the catalogue grows

The capstone's distinctive point is that the right strategy changes as the catalogue changes. A set that was small and stable at launch, correctly served by monolithic loading, can grow over time until monolithic's token cost and selection distraction cross the point where progressive discovery would serve better. If the decision is treated as a one-time launch choice and never revisited, the system keeps paying monolithic's growing cost long after it stopped being the right call. So the discipline is to re-evaluate the context strategy as the catalogue grows, much as the tool-set audit re-evaluates the tool set as scope evolves.

The trap is fixing the choice once and never reassessing it. The credited answer treats the monolithic-versus-progressive decision as one to revisit whenever the catalogue's size or variability crosses the threshold where the original justification no longer holds.

What the exam trips candidates on

Two traps. The first is choosing a strategy once at launch and never reassessing it as the tool or document catalogue grows, ignoring that the right choice changes with scale. The second is applying progressive discovery uniformly even to a small, fixed catalogue where it only adds unnecessary complexity. The credited answer matches strategy to current scale and revisits the decision as the catalogue grows.

Common misreadings to avoid

How this shows up on the exam

Expect two catalogues at opposite scales, or one catalogue that has outgrown its launch-time strategy. The reliable reading is monolithic for small-stable sets, progressive for large-variable ones, and revisiting the choice as the catalogue grows. This capstone integrates monolithic vs progressive strategies, the token cost of monolithic loading, and implementing progressive discovery, and shares its revisit-as-scope-evolves discipline with the tool-set audit workflow.