Which strategy bundle best addresses token scarcity?

Dedup and quality filtering are not housekeeping; they are effective-budget multipliers. Repeated low-signal text consumes steps that could have been spent learning new structure. When scarcity is real, every redundant shard has opportunity cost.

Teams that continuously refresh dedup policy and quality thresholds often recover more useful learning than teams that merely increase crawl volume. The point is not fewer tokens in absolute terms; it is higher signal per optimizer step.

In practice, this section is where interviewers test decision quality. A strong answer links curation for signal density, diversification for breadth, and governed synthetic augmentation to one observable symptom and one corrective action. You can cite novelty-adjusted source utility, duplication rates, and domain-stratified regressions as the monitoring surface, then explain how the team decides whether to continue, rollback, or retune. Grounding the explanation in measurable signals prevents the conversation from becoming generic theory and shows that you can operate under uncertainty with finite compute budgets.

A useful teaching pattern is to add a concrete scenario: more crawl tokens but weaker performance on high-value long-tail domains. After naming the scenario, state the failure boundary (chasing raw token volume while overlap and low-signal content consume budget) and the operational response (portfolio planning that reweights sources by marginal learning value). This structure demonstrates ownership thinking: you are not only describing what the concept means, you are showing how to keep a production run safe when this concept becomes the deciding factor.