Research
Handing a Task Off Mid-Flight Instead of Restarting It Hits Comparable Accuracy at 22-34.6% of Full-Agent Cost
Commitment-Frontier Residual Completion formulates model-to-model handoff as commitment-constrained residual completion: it freezes a residual contract from accepted progress, closes the successor's continuation into an evidence-linked graph, and admits execution only when the remainder is covered, with live receipts discharging obligations. Across five environments and two same-provider model pairs, CFRC matches strong full-task agents on macro accuracy at 22.0% to 34.6% of their inference cost, with cross-provider results showing the transfer holds. The three enforced invariants are target-before-proposal, whole-proposal-before-authority, and live-evidence-before-success.
↳ Follow the thread