FPF.Preface:11 - When A Scale Claim Calls For Comparison

FPF’s Bitter-Lesson Preference in E.2:6 uses the empirical Bitter Lesson to motivate examination of claimed advantages from more compute, data or search capacity in computational work. When that scale or generalization claim is current, or a separately declared local generality policy applies, begin with C.19.1’s cheap scale-claim probe. State the receiving task, usable budget range, conditions and evidence needed for the claim.

Compare admissible options by their usable performance and uncertainty over that range. A steeper improvement curve can still stay below the required performance. When neither option dominates under the applicable comparison, there is no scale-based preference; a separate local policy must state its own basis and cost if it chooses greater generality.

An ordinary bounded procedure can be used on its task-specific grounds. For example, a fixed procedure for summing a known finite list needs correct arithmetic and an adequate task budget; the existence of a general learner does not require a scale audit or an adaptation waiver. Independently applicable assurance and oversight still apply.

Positive procedures, prohibitions and autonomous sequencing follow the actual task, control requirements and authorized policy. Minimal prescription and autonomy do not follow from a scale comparison. Adaptation likewise needs its own objective, evidence, permission, change authority and safeguards. Use E.2:6–7 for the exact activation, comparison, heuristic-debt and precedence conditions.