compute_split's largest-remainder dust distribution has a dedicated at-scale test in contracts/escrow/src/test.rs (test_large_split_distributes_dust_by_largest_remainder, exercising 60 recipients with a hand-computed O(n²) reference implementation cross-check). refund_remaining_budget's structurally identical dust-distribution logic (in contracts/milestones/src/lib.rs) is only exercised by test_multi_sponsor_refund_rounds_dust_by_largest_remainder in contracts/milestones/src/test.rs, which uses exactly 3 contributors — nowhere near the MAX_SPONSORS = 20 worst case, and nowhere near the scale that would meaningfully exercise the tie-breaking behavior (multiple contributors sharing an identical fractional remainder) the way the escrow-side test deliberately does. Add a milestones-side large-scale dust test mirroring escrow's, ideally at or near MAX_SPONSORS contributors with deliberately colliding remainders.
compute_split's largest-remainder dust distribution has a dedicated at-scale test incontracts/escrow/src/test.rs(test_large_split_distributes_dust_by_largest_remainder, exercising 60 recipients with a hand-computed O(n²) reference implementation cross-check).refund_remaining_budget's structurally identical dust-distribution logic (incontracts/milestones/src/lib.rs) is only exercised bytest_multi_sponsor_refund_rounds_dust_by_largest_remainderincontracts/milestones/src/test.rs, which uses exactly 3 contributors — nowhere near theMAX_SPONSORS= 20 worst case, and nowhere near the scale that would meaningfully exercise the tie-breaking behavior (multiple contributors sharing an identical fractional remainder) the way the escrow-side test deliberately does. Add a milestones-side large-scale dust test mirroring escrow's, ideally at or nearMAX_SPONSORScontributors with deliberately colliding remainders.