Two cosmetic smells the cycle-0065 close audit flagged (#132), neither a
correctness issue:
1. Broker label. run_stage1_r reused sim_optimal_manifest, recording
broker: "sim-optimal(pip_size=...)" for a harness that also runs a
RiskExecutor branch. It now overrides the label to
"sim-optimal+risk-executor(pip_size=...)" so the dual-tap harness reads
honestly. A new assertion in run_stage1_r_synthetic_folds_an_r_block pins it.
2. col[i] leak. stage1_r_blueprint did format!("col[{i}]").leak() per
PositionManagement field (14 strings) to satisfy the &'static str port-name
API -- harmless for a CLI one-shot, but a leak-per-build smell once the
harness is reused in a sweep / Monte-Carlo loop (the #133 direction). Replaced
with a process-wide LazyLock<Vec<String>> (COL_PORTS), interned once: any
number of builds now reuse the same 14 strings. (#132 deferred this to #133,
but the intern is robust to any reuse pattern, so it lands here cleanly.)
Behaviour-preserving (the port names and the recorded R-record are unchanged);
the broker label is the only observable change, and it is pinned.
Verified: clippy --all-targets -D warnings clean; full suite 514 passed / 0
failed; live `aura run --harness stage1-r` reports
"broker":"sim-optimal+risk-executor(pip_size=0.0001)".
closes#132
Brummel
3d3baaeb24