Integrate the squared jerk in closed form rather than sampling it

[Eelis03]

The jerk cost was only ever a Simpson quadrature of a sampled trace, and the design notes recorded that as a known limitation with its own remedy: integrate the coefficient table segment by segment. On one segment the jerk of one joint is a polynomial, so its square is a polynomial too and the integral over that segment is a quadratic form in the jerk coefficients, which makes the objective exact for every trajectory this package produces. The S-curve of the README costs 240.0 analytically; the quadrature reported 240.3600 at 4001 samples and the integral reports 239.99999999999994 at any sample count, because it reads none.

• PiecewisePolynomialTrajectory.jerk_cost sums the quadratic form over every segment and joint, and IntegrableTrajectory is the protocol that names the capability, alongside CertifiedTrajectory.
• certify_jerk_cost is the analysis-layer entry point, next to certify_limits, and refuses with a TypeError that points at compute_metrics when the trajectory cannot be integrated.
• TimeScaledTrajectory forwards by dividing by the fifth power of its scale and SynchronisedTrajectory by adding its components, both raising TypeError when what they wrap cannot answer.
• tests/test_jerk_cost.py pins the value against a hand-computed integral, against 720 d^2 / T^5 and 400 * 4 * 0.15, against the quadrature at 4001, 40001 and 400001 samples, and against the claim that the natural quintic spline is the minimum-jerk interpolant of its via points. Opaque moves to conftest.py so both suites share it.
• The design notes move the limitation into "Closed limitations" with what it cost and what remains; the README gains the exact figure beside the quadrature it explains, and its test count and statement count are updated.

Local gates: 218 tests pass, coverage 97.15 percent of 1230 statements against the 95 percent gate, ruff check . and mypy clean.

🤖 Generated with Claude Code