Goldwäscher (quantum-record-mining)
Run before paper exploration. This is read-only triage; it does not reopen lanes or submit rows.
- Snapshot the official board and mechanically extract Q×T mentions from the authorized ledgers, lane mailboxes, result files and notes. Use
scripts/frontier_mine.py; inspect missed formats and preserve source/line provenance. Mentions are leads, not measurements. - Compare each lead with
C(q)=min(T_i:Q_i<=q), using benchmark semantics. A row dominated by a lower-width result has no new frontier value. Refresh before any promotion. - Search
verifier|HOLD|blocked|killed|closed|parked. Separate process blockers (missing review/delivery) from correctness/economic rejection. Recover primary receipts and current ownership; a blocked result is not permission to duplicate it. - Build a compact lever × architecture matrix from notes and active source: present, absent, incompatible or unverified, with function/hash references. Older-tree savings often disappear on newer trees; re-price the actual composition.
- Rank measured unreviewed gains first, then source-supported cross-family transfers and one-qubit cliffs. For a cliff, name every binding phase and use the distinct Pareto and product budgets. Hand off a five-line proposal under the shared contract.
Example, from this skill directory:
python3 scripts/frontier_mine.py board.json --q 796 825 --notes ledger.md results.md --out mining.json
The script uses explicit local files only. Review all prospective gains against complete logs; it neither certifies exactness nor changes campaign state. Stop when a lead is already owned, dominated, unsupported by receipts or killed without a changed premise.