MetroVolt's register keeps its wounds: the δ-scan that refused to flatter deeper shaping, the robustness scan that named plasma current unforgiving, the breeding analysis that came back negative. They are deposited beside the favourable results, same format, same prominence.
S77 tested the reviewer-suggested deeper-triangularity lever and reported the adverse trend: safety factor falls with |δ|, so the frozen point already sits at the favourable end. S79's excursion matrix found B-field shortfalls recoverable but a 10% plasma-current shortfall not recoverable by density trim — 'least forgiving parameter' is our phrase, in our deposit. §3.5 computed prompt ³He breeding at ratio ≥1 as not achievable, forcing the staged supply strategy. The shield ledger carries its conductor-fluence line ~8% above a degradation onset, flagged rather than smoothed.
Even the near-thermal baseline's confinement requirement — the design's hardest single number — was moved to the top of the published band (≈2.2) when the systems code said so.
Selection bias is the quiet killer of technical credibility. A register that demonstrably retains its negative results converts every positive result into stronger evidence — reviewers know nothing was curated out. In diligence, our adverse findings are load-bearing assets.
| Adverse δ-scan | S77 — deeper shaping costs q95 |
| Unforgiving parameter | Ip −10% unrecoverable (S79) |
| Breeding ≥ 1 | computed negative (§3.5) |
| Flagged margin | conductor fluence ~8% above onset |
| Baseline H98 posture | ≈2.2 — top of band, stated |