Thesis
v2's union overshoots the budget on 8.5% of instances. The sequential variant (BM25 → EGO on the remainder) is described in v2 but never measured; in v3 it and calibrated fusion are the main rows of the table.
Implementation
Two pipeline calls: the second with budget = B − used₁ and the first pass's selected paths excluded from the candidate set. Trivial over the existing pipeline.
Honesty requirement
Two passes = double cost. Report wall-clock p50/p90 as a table column, not a footnote — this is an argument of the table.
Cross-refs: #241 (--budget is not a hard cap: cores bypass it) and #238 (per-file budget share) both distort "budget-compliant" claims and are prerequisites.
Q-class. Matrix cell 10.
Thesis
v2's
unionovershoots the budget on 8.5% of instances. The sequential variant (BM25 → EGO on the remainder) is described in v2 but never measured; in v3 it and calibrated fusion are the main rows of the table.Implementation
Two pipeline calls: the second with
budget = B − used₁and the first pass's selected paths excluded from the candidate set. Trivial over the existing pipeline.Honesty requirement
Two passes = double cost. Report wall-clock p50/p90 as a table column, not a footnote — this is an argument of the table.
Cross-refs: #241 (
--budgetis not a hard cap: cores bypass it) and #238 (per-file budget share) both distort "budget-compliant" claims and are prerequisites.Q-class. Matrix cell 10.