Codesign SEAM 1: the MockSoc rehearsal behind the hardware stage

[aarontrowbridge]

Important

Problem — The amicode hardware stage is an honest stub: it records intent (a device_session entity) but runs nothing, so the researcher's first contact with the closed loop is a promise, not a preview. The landed stack can already do better: Strumento.jl's MockSoc runs the entire transport path (translate → envelopes → execute! → synthetic IQ → Measurement → one strategy step) in pure Julia, no Python, no board — the identical seam real hardware uses.
Approach — the solved pulse runs a full sim rehearsal through the ACTUAL Strumento.jl path, recorded into the device session, honestly labeled sim. The stage stays an honest stub until a rehearsal PASSES — a failed rehearsal does not satisfy it (failures surface distinctly).
Scope — in: the rehearsal path (however the extension best invokes it — a CLI-launched script following the established run-launch pattern is the expected shape), the outcome-gated artifact, the stage wiring · out: any bespoke simulation (F1: if the real path cannot be used, ship NOTHING and the stage stays a stub), any hardware I/O, any change to Strumento.jl/Intonato surface.

Acceptance Criteria

• mocksoc_rehearsal_artifact_recorded == 1 AND rehearsal_outcome == success per hardware-stage invocation; a failed rehearsal is surfaced distinctly and does NOT satisfy the stage
• The artifact carries the pulse content-hash, the mock system's mismatch declaration, and the strategy-step outcome
• rehearsal_imports_strumento_path_only == 1 — an import-surface check: the rehearsal entry point imports no sim module outside the Strumento.jl path (no bespoke sim, mechanically checked)
• The rehearsal is honestly labeled sim in the device-session record and the UI surface

Testing Decisions

The import-surface check is a test (grep/import-graph over the rehearsal entry point + its module graph); the outcome-gating is a unit test over the stage's recording path (a failing rehearsal leaves the stub state honest); an end-to-end test drives one solved-pulse fixture through the rehearsal in the test env. Follow the repo's vitest conventions; the rehearsal script itself is exercised through whatever invocation pattern the extension uses for Julia work (the established run-launch seam).

Key Decisions

• The rehearsal runs the Strumento.jl MockSoc verbatim — the entire point is that the preview exercises the identical transport seam real hardware uses.
• MockSoc is pure Julia: the test/rehearsal env resolves Strumento + Intonato + Piccolo from the registry — no PythonCall, no qick extra, no board.
• F1 verbatim: if the slice cannot use the Strumento.jl path, it ships nothing rather than a declared-sim lookalike.

Constraints & Invariants

• The honesty invariant set (LIVE/STALE/DEAD, untrusted-until-verified, propose-only-on-hardware) applies; the sim label is part of the trust chain, not a disclaimer bolted on.
• No landed contract or API changes — the stage's entity schema may grow additively (wiring, not re-shaping).

Prior Art

The hardware-loop skill's MockSoc idiom (the canonical pattern); Strumento.jl's MockSoc + integration test (the whole QILC→board loop tested with no Python and no board); the amico CLI's run-launch pattern (how the extension already drives Julia); the device_session entity surface.

Source

Part of #679 · design-of-record: specs/spec-20260831-120000-amicode-outside-lab-codesign.md (SEAM 1) · no blockers.

Notes

First slice of the codesign decomposition — picked because it is unblocked, amicode-only, and makes the journey's stage-5 story real.

[aarontrowbridge]

The EXTENSION half landed in #692 (rehearsal + outcome-gated stage + the device-session record). The UI-pill half is the fork-flow follow-up: #693 — the overlay is fork-generated, so the pill rides the fork-first flow (the drift gate caught the hand-edit exactly as designed; the extractor is the only door).