DRAFT feat(template): verified Piccolissimo solve template (beta target) + integration contract

[Rchari1]

Stacked on #23 (base rchari/perf-smooth-solve). Do not merge.

Moves amicode's solve stack toward Piccolissimo (Harmoniqs' commercial stack — bundles Piccolo + adds the exact HermitianExponentialIntegrator) for the beta. Lands the verified template as a parallel artifact; the live raw-Piccolo path (solve_template.jl + AGENTS.md) stays the default until the Piccolissimo binary is consumable.

Verified

solve_template_piccolissimo.jl — transmon X gate via Piccolissimo's HermitianExponentialIntegrator, run end-to-end through amico-run against a local Piccolissimo checkout (Piccolo 1.10 + Piccolissimo + CairoMakie, julia 1.12.3):

• F = 0.99995, 60 iters, ~88 s cold
• full run-dir contract (manifest → 10 plot_pulse frames → FINISHED → result.toml)

What it took (vs raw Piccolo)

• Piccolissimo does not reexport Piccolo → load both.
• PiccoloOptions(timesteps_all_equal=true) is required — free Δt stalls at F≈0.976; pinned → 0.99995.
• Piccolissimo pins Piccolo 1.10 (template was 1.19) — all needed symbols exist in 1.10.

Integration contract (for Jack's in-flight binary) — see PICCOLISSIMO_INTEGRATION.md

• The artifact is a pkgimage, loaded via the Julia depot — not --sysimage; amico-run wiring differs.
• Only x86_64-linux-gnu is published; macOS labs need an aarch64-apple-darwin asset.
• Pinned to julia 1.12.6 — amicode must match.
• CairoMakie isn't in Piccolissimo's deps → plot_pulse stays ~35 s cold unless baked into the amicode image.
• PICCOLISSIMO_LICENSE provisioning once the (currently-disabled) license gate re-enables.
• Source install isn't distributable — Altissimo is an unregistered local-path dep, so the binary is the only delivery path.

Not in this PR (deliberately)

No change to AGENTS.md, the bundled julia project, or amico-run's load path — those flip when the binary lands (checklist in the doc), so the working demo stays intact.

🤖 Generated with Claude Code

[Rchari1]

@jack-champagne

[jack-champagne]

Heads up: a macOS Piccolissimo binary is now published on the Piccolissimo binary artifacts (see recent builds), built against julia 1.12.6. That closes the macOS gap your PICCOLISSIMO_INTEGRATION.md flagged — pull the recent build and you can validate the darwin path directly instead of being x86_64-linux only. Match julia 1.12.6 exactly (pkgimage, not sysimage).

[jack-champagne]

Retargeted from rchari/perf-smooth-solve to main.

The old base is now fully merged into main, so the PR was comparing against a dead branch. Against main the real remaining change is 2 files, +192 — that's what actually needs review here.

(Part of a sweep over the stacked b*/aaron/* chains: #119, #125 and #131 turned out to be entirely in main already and were closed; this one has real work left.)

[Rchari1]

Closing — superseded, and the parts that aren't superseded have gone stale.

The template is superseded by #219, which taught the live template (scores/pulse-designer/templates/solve.jl) to run Altissimo via a SOLVER = :altissimo flag, with per-iteration frames, AMICODE_ITER lines, a run.log the cloud poller can parse, and cooperative STOP on the Altissimo hook.

Worth recording why this one couldn't have shipped that: it lives at packages/extension/templates/, which no code references — the loader reads the scores/pulse-designer/templates/ path. It also solves via HermitianExponentialIntegrator (a Piccolissimo integrator) rather than AltissimoOptions (the solver backend), so it's a different axis from the HP tier, and it carries almost no telemetry.

PICCOLISSIMO_INTEGRATION.md's central claim is no longer true. It documents Altissimo as an unregistered local filesystem path dep (path = "/Users/raghavchari/Altissimo.jl") with the prebuilt image as the only delivery vehicle. Piccolissimo main now declares it as a private git dep tracking main:

[sources]
Altissimo = {rev = "main", url = "git@github.com:harmoniqs/Altissimo.jl.git"}

So porting the doc as-is would have preserved a constraint that has since been fixed. The real remaining constraint is narrower: an AMI freezes a resolved Altissimo commit at image-build time, and re-resolving needs credentials for the private repo — which is why a rebake, not a Pkg.update on the box, is the delivery route.

Nothing here is lost that we still need. Reopen if the integration contract wants rewriting against the current dependency shape — that'd be a fresh doc, not this diff.