Verchehrskadett (quantum-structural-lanes)
Change the circuit's information storage, reconstruction, cleanup or representation. A promising paper is a hypothesis, not a gain.
Start with the actual board
Confirm benchmark, active owner and contract. These methods concern reversible quantum circuits; do not transfer them to another kind of benchmark without a concrete mechanism. Maintaining these skills does not change the campaign.
Read the shared contract and benchmark facts. Start with the five-line post; reserve the full card for an integrated candidate.
Run record mining before papers: recover overlooked measured rows, process-blocked work and mechanisms present in one architecture but absent in another. Respect existing ownership.
Route by the measured bottleneck
| Bottleneck | Route | Decisive question |
|---|---|---|
| Unused result, blocked verifier or missing cross-family lever | quantum-record-mining |
Is useful evidence already available? |
| One-qubit frontier cliff | Record mining, then peak-owner census | Which architecture stops here, and what binds one qubit lower? |
| Peak allocation inside an already dirty-safe gadget | quantum-dirty-borrowing |
Is there a legal donor interval? |
| Proven last-use cleanup | quantum-mbundo-lifetimes |
Can conditional phase repair replace that suffix? |
| Structured garbage support | quantum-syndrome-uncompute |
Can a smaller correction basis preserve the required channel? |
| Expensive walk or payload replay | quantum-replay-arithmetic |
Can state and consumer be redesigned together? |
| Stored branch history | quantum-divstep-reconstruction |
Does successor state determine the needed reverse action? |
| Material coherent table lookup | quantum-dense-qrom |
Does the native ISA and actual table support the construction? |
| Long compatible arithmetic region | quantum-encoded-arithmetic |
Can conversion be amortized across its consumers? |
| A trusted independent evaluator | quantum-verified-architecture-search |
Can search change an architectural cost term? |
A match selects an audit, not an implementation. Read the full relevant construction/proof and map every premise to actual source. Identify already implemented mechanisms and price missing primitives before ranking.
With parallel work authorized, separate ownership by mechanism and files; do not create one duplicate lane per paper. Keep deterministic MBU distinct from probabilistic syndrome cleanup, and replay separate from competing edits to the same backend.
Measure changes alone before composition, then recompute dependencies and complete-path costs. Do not add component percentages. Apply the shared stop rules and promotion gate; report measured gains, scoped negatives and missing receipts.