Prediction calibration is running
The corrected prediction method is running its fresh calibration campaign on four numerical threads. Independent validation will follow only if the complete calibration supplies a finite, independently verified error threshold.
| Phase | Independent simulated families | Maximum fit attempts | Expected time | Cooperative ceiling |
|---|---|---|---|---|
| Calibration — running | 299 | 7,176 | About six days | Ten days |
| Validation — conditional, not started | 181 | 4,344 | Another three to four days | Six days |
These examples have known simulated answers. Calibration establishes a prediction-error limit; separate validation checks its coverage on new examples. This work adds no aircraft measurements and does not yet establish confidence for the original 53 aircraft comparisons.
What changed before this run
The preceding calibration stopped at a numerical verification failure. Its results and unused allowance remain closed. The solver correction now passes the complete 48-case qualification: all cases, 24 starting-point pairs and four original-input replays. The new release passes 98 software checks and freezes 253 source files and both phase policies. The new cohorts exclude all earlier declared calibration and validation families.
Keep the original physical bounds, numerical limits and acceptance checks. Every required setting must supply a usable fit. Calibration stops cooperatively if both prescribed starts fail a required setting; validation stops at its fifth failed family. Interrupted starts and failures remain recorded without retries.
Storage and completion
Temporary case files are expected to use about 12–20 GiB during calibration and 7–12 GiB during validation. Publication can temporarily duplicate them. Verified archival and compaction should retain about 8.4 GiB of payload for both phases, plus metadata and archive overhead. Require 55 GiB free before calibration and 35 GiB before validation.
After calibration closes, independently audit and publish every returned and unexecuted outcome, verify its hashes, then archive and compact recoverable working copies. Prepare validation only from a complete finite handoff under the same sources and environment. A stopped or failed calibration cannot launch validation or supply a statistical claim.
Full finite plan · Preparation snapshot · Next analyses · Original aircraft findings.