[aeronauty-flexcompute]

The native drag decomposition skipped low-Rtheta stations, clamped signed shear and used absolute x distance, so it disagreed with XFOIL’s actual CDf/CD−CDf report even though total CD matched. Replace those heuristics with CDCALC’s signed freestream projection, preserve surface y coordinates in the faithful force bridge, and gate both component outputs.

XFOIL’s polar-file CDp is a different surface-pressure integral; this PR explicitly compares the native cd_pressure = CD−CDf field with the matching final iteration report. It does not redefine that API or add a new pressure-integral output.

Measured evidence:

• All 48 old native cases fail the matching component checks (maximum discrepancy about 0.00424).
• Corrected 48/48 native cases pass; maximum CDf difference 4.91e-6 and CD−CDf 4.99e-6, both within 1e-5. Existing total-CD/CL/CM/transition/BL tolerances are unchanged.
• An instrumented pinned-source control matches all 158 surface intervals: x/y exactly, projected distance within 3.47e-18, wall-shear differences ≤3.62e-7. The instrumented Rust/Fortran drivers and logs are retained; production contains no probe code.
• 574 local Rust tests pass (23 inherited ignored); final-head Linux Rust CI also passes after inheriting the two additional flap regressions, 155 Python tests pass, four gate-negative/definition tests pass, 48/48 rebuilt shipping-WASM/reference/native cases pass, and docs/API build passes.

PR impact

Base/merge base: #39 at ad4a361. Head: 5a056e4. Twelve incremental files; inherited gate/cache/docs/flap slices are separate reviews.

Source-confirmed route: faithful update_force_state → canonical_force_state carries physical U, density-weighted Cf and actual surface y → shared compute_forces_from_canonical_state → signed compute_row_friction_drag → native/CLI CDf and CD−CDf output. The shared canonical integrator can also reach existing legacy callers; legacy station-based integration and general legacy-engine validity remain outside the reference claim. No new public Rust signature is introduced by this slice.

Gate route: final converged XFOIL iteration CDf/CD−CDf → native serial and batch comparisons. A negative test distinguishes that report from polar-file CDp and rejects missing component output. The WASM bridge does not expose these component fields, but its coefficients/Cp still pass the complete existing binding gate. Tests do not establish arbitrary geometry/Mach, experimental physical accuracy, global derivatives or interacting multi-element flow. The advisory reference scan is capped.

Changelog, Versions & Docs

• CHANGELOG covers the changed native component behavior
• UI package/top entry and both lock metadata versions agree (1.1.16)
• Python distribution bumped (1.1.16)
• Force/gate/API docs updated with equations, actual extracted source, interactive projection and measured limits
• Docs built with no broken links or anchors

Final-head Linux CI passed: run38557641634, rust/python and Bugbot green. Base/merge base and impact were refreshed after the flap follow-up.