Context
Piccolo 2 (with bending-energy defaults, harmoniqs/Piccolo.jl#312) and DirectTrajOpt 0.10 are landed upstream, and Intonatissimo has migrated with bit-for-bit loop parity (harmoniqs/Intonatissimo.jl#75). The shipped amico env (`~/.amico/julia`) still pins Piccolo 1.19.0 / DirectTrajOpt 0.9.7 / NamedTrajectories 0.9.1 — all pre-DTO-#122. Today's vetted-template behavior is therefore unchanged; nothing is broken now.
The forward exposure (this issue is the gate)
DTO #122 (v0.9.9) changed `QuadraticRegularizer` weighting from Δt² to single-Δt. On a uniform grid the old value is reproduced by `R_new = R_old · Δt`. The vetted template (`scores/pulse-designer/templates/solve.jl`) uses `SmoothPulseProblem(qtraj, N; Q = 100.0, R = 1e-2)` with ZeroOrderPulse — at default T=10 ns, N=50 (Δt = 0.2), keeping `R = 1e-2` on DTO ≥ 0.9.9 makes the regularizer effectively 5× stronger than the value the template's convergence defaults (F > 0.999) were tuned against.
Measured evidence for scale: in the Intonatissimo production reference (linear spline, R = 1e-2), the incumbent QuadraticRegularizer contribution is ~0.02 vs the fidelity term's ~0.018 — same order, i.e. a 5× shift is not negligible for one-shot solves. (The QP-ILC loop itself was proven insensitive: per-iteration renormalization cancels the change — all five regression problems |ΔF| = 0.0.)
The gate
When the shipped env bumps to Piccolo 2 / DTO 0.10:
- Vetted template: pass `R = 1e-2 · Δt` (or re-tune and re-validate) so the shipped default matches the semantics the F > 0.999 claim was earned on.
- Re-run the default-regime convergence check (T = 10 ns, N = 50, max_iter = 60, X gate) and record F before shipping the bump.
- Sweep the other score templates (`pasqal-mis`) for the same pattern.
Demos carry the same exposure but are dev-facing; the amico env bump is the one tracked moment that covers shipped artifacts.
Context
Piccolo 2 (with bending-energy defaults, harmoniqs/Piccolo.jl#312) and DirectTrajOpt 0.10 are landed upstream, and Intonatissimo has migrated with bit-for-bit loop parity (harmoniqs/Intonatissimo.jl#75). The shipped amico env (`~/.amico/julia`) still pins Piccolo 1.19.0 / DirectTrajOpt 0.9.7 / NamedTrajectories 0.9.1 — all pre-DTO-#122. Today's vetted-template behavior is therefore unchanged; nothing is broken now.
The forward exposure (this issue is the gate)
DTO #122 (v0.9.9) changed `QuadraticRegularizer` weighting from Δt² to single-Δt. On a uniform grid the old value is reproduced by `R_new = R_old · Δt`. The vetted template (`scores/pulse-designer/templates/solve.jl`) uses `SmoothPulseProblem(qtraj, N; Q = 100.0, R = 1e-2)` with ZeroOrderPulse — at default T=10 ns, N=50 (Δt = 0.2), keeping `R = 1e-2` on DTO ≥ 0.9.9 makes the regularizer effectively 5× stronger than the value the template's convergence defaults (F > 0.999) were tuned against.
Measured evidence for scale: in the Intonatissimo production reference (linear spline, R = 1e-2), the incumbent QuadraticRegularizer contribution is ~0.02 vs the fidelity term's ~0.018 — same order, i.e. a 5× shift is not negligible for one-shot solves. (The QP-ILC loop itself was proven insensitive: per-iteration renormalization cancels the change — all five regression problems |ΔF| = 0.0.)
The gate
When the shipped env bumps to Piccolo 2 / DTO 0.10:
Demos carry the same exposure but are dev-facing; the amico env bump is the one tracked moment that covers shipped artifacts.