Wing Optimizer 2.6.0

The 2.6 guide explains the CG-selection fix, integrated drag checks and role-aware descriptions following a research/three-run audit. Close an old window and reopen Open V2.bat to load 2.6.0. See current validation. Saved requirements and scoring remain intact; no solver starts automatically.

The 2.5 guide explains bounded objective-elite training, nonlinear utility uncertainty, consistent fine-promotion comparisons, clearer model diagnostics and Compare saved settings…. Existing scoring and physical requirements remain intact. See 2.5 validation.

The 2.4 update guide explains the revised late-search scheduling, selective fine checks, measured completion summaries and separate debug workbook. Close an older app window and reopen Open V2.bat to load the updated app. Saved requirements, scoring policies and results are preserved.

The 2.3 update guide describes the new soft pitch policy, prediction-guided search before the first passing wing, fine promotion and prediction diagnostics. Select Balanced efficiency and trim for a new run; older saved settings keep their original policy on resume. Upgrading still improves the proposal algorithm used by resumed surrogate searches.

The Excel repair guide explains lower-memory exports, safe replacement and recovery when a workbook is open in Excel.

The 2.2 update guide explains neutral cruise assessment, optional angle/margin preferences, bounded retries, continuing reserve challenges, allowance reuse and corrected polar exports. Start a new run to change the objective; resume keeps the original run's aircraft and scoring settings.

V2 keeps the Settings, Airfoil library, Checked design and 3D, and Results tabs. Action descriptions remain hover help on buttons. Original DAT files and V1 histories remain unchanged; V2 has independent polar copies, settings and dated run folders.

Normal workflow

1. Open V2/Open Wing Optimizer.bat. Review mass, slow/cruise/fast speeds, extra drag, geometry and achievable CG bounds, packaging limits and aircraft role.
2. Choose automatic foil screening or your manual root/tip pools. Automatic selection uses compatible real section polars for rough screening only. The library tab can preview coverage and manually fill missing selected Re cases. Pause/resume remains user controlled.
3. Preliminary search is optional. It provides rough geometry seeds and an inexpensive comparison; it is not the final 3D geometry search.
4. For a flying wing, choose Balanced efficiency and trim, enable Neutral cruise trim preference, and normally leave Require supported 3D cruise crossing off. In Checked design and 3D, choose surrogate_trust_region and your runtime, evaluation limit, independent starts and coarse/fine meshes. Use Start Flow5 geometry search when you want to start real solver work.
5. Review saved required-lift scores, raw polars, evidence status and numerical disagreement. Pause keeps a checkpoint. Resume verifies the original aircraft, foil and solver identities; it does not silently change the objective.
6. Export Excel and native Flow5 projects from Results. Each dated run contains its workbook, exported projects and ZIP; repeated exports receive a suffix. Exporting loaded V1 results places the new artifacts under V2's results/imported_history and preserves the original evidence.
7. Manually check actual CG, elevon trim, control authority and trimmed drag in Flow5 before CAD. Optional sensitivity and identical-setup repeats remain explicit actions. Native checked physical trim still needs a verified adapter.

What the search changes

The scheduler learns shared geometry trends across categorical root/tip foil choices and local trends within individual pairs. Several independent search regions compete. Complete valid physical rejects can supply provisional centres before a wing passes; they retain their original rejection status. Predicted promising trials are interleaved with global exploration, alternative foil challenges and direct-search polls, so early leaders do not consume all the trials. Insufficient data or an unusable model falls back to direct search. The legacy option allows comparison with the preserved scheduling method.

Model suggestions have no aerodynamic score until fresh Flow5 polars are available. Incomplete solver runs are unknown, not fabricated zero-performance training samples. A solved physical rejection retains its coefficients and continuous constraint margins. Matched coarse/fine evidence informs model caution; coarse predictions cannot certify a fine-mesh result.

The model chooses from a bounded set of RBF/Matérn kernels and length scales. Before a feasible wing exists, proposals seek reduced normalized constraint violation. Later proposals balance improvement, uncertainty and measured runtime with a bounded cost preference so a cheap weak foil pair cannot automatically exclude a slower alternative. Fine checks retain measured efficiency leaders and use conservative fine forecasts for challengers.

Results show both training observations and executed model-selected trials. The prediction diagnostics compare forecasts saved before a solver result with subsequent usable observations, including physical rejects. A trained model alone does not demonstrate effective guidance or predictive accuracy.

Defaults use 8 solved samples before a model is fit, at most 160 training rows per pair, 128 proposals per region and a 25% model-independent exploration allowance. Keep these defaults for initial tests. Finite budgets and finite foil shortlists can still miss a better design. Model uncertainty is a scheduling heuristic, not a physical confidence interval or proof of a global optimum.

Performance and pitch meaning

Mission scores use the required lifting CL at your selected speeds and one achievable CG. The score uses wing CD plus the extra drag coefficient exactly once. Excel's raw wing-only polars can therefore show higher L/D than the mission comparison; both are labeled. Peaks at unsupported lift or high alpha do not become mission performance.

Excel's separate Untrimmed mission sheet compares saved required-lift L/D, total drag and aerodynamic power against airspeed with its own wing and unit selectors. It does not add extra CD again or claim physical trim. Failed or uncovered points remain gaps. The existing Flight comparison keeps its physical-trim evidence gate, and the full raw alpha polars remain available.

Under Balanced efficiency and trim, flying-wing mode requires restoring local Cm slope and your minimum positive static margin. The upper margin is a soft target limit, with a penalty for excess at all three mission points. The neutral cruise preference favors small |Cm| at required cruise lift, a moderate margin near the configured target, and a supported crossing nearby. It does not reward reducing restoring stability toward zero. A crossing is optional unless Require supported 3D cruise crossing is explicitly enabled. Legacy scoring retains its original hard static-margin band. The optional absolute-Cm objective trades moment burden against efficiency; tailed-main-wing mode disables standalone aircraft pitch gates and Cm penalties. Neutral-control Cm is not a verified elevon deflection or trimmed drag estimate. An optional advanced search.neutral_cruise_cm_tolerance can require small neutral-control cruise Cm for flying wings; null leaves that extra gate disabled. It does not demand a physically impossible zero moment at every speed.

V2 rechecks saved polar files, interpolation support, lift, total drag, common CG and pitch evidence. Pinned native 7.54 readback can verify applied settings for supported saved projects. A completed process or attractive L/D alone is insufficient. Results remain untrimmed; numerical consistency, actual trim and flight readiness are separate states.

The full sweep to +20 degrees is retained as a diagnostic. Flow5's thin-surface methods and section-drag lookup do not validate separated 3D stall or all drag sources. Extra CD is a declared assumption, not a fitted accuracy guarantee.

Research basis and validation

This is a lightweight constrained surrogate scheduler inspired by TuRBO and SCBO, combined with the existing direct-search guards; it is not their reference implementation. Flow5 moments explains the CG and reference dimensions and the sensitivity of near-field moments. Flow5 stability guidance requires trim before aircraft stability-mode interpretation.

See REALISM_VALIDATION.md, EXPORT_FEATURE_PARITY.md and the local V2 validation record for software tests and measured-case scope. Mocked search benchmarks test scheduling behavior only. The 2.4 validation record reports the current software checks and equal-budget comparisons. The earlier 2.3 record retains the previous checks. Both fictional comparisons include mixed outcomes and do not demonstrate general superiority. No new real solver was launched for 2.3 or 2.4. The historical approved six-job fresh check completed with native settings and local viscous evidence; its unsuitable fixed wing was rejected. This verifies a selected solver round trip, not universal agreement with flight measurements. GitHub publication follows local build and verification, as requested.

Compatible Flow5 unit-solution reuse and batched speed cases were considered. They remain disabled pending an explicit solver round trip and native evidence checks. Separate repeatability evaluations must continue to request fresh work.