feat(engineer): simulate the real firmware and check the whole system fits

[devin-ai-integration]

Summary

Three gaps closed, all versions of the same problem — the stages produced artifacts that were never checked against each other.

1. The simulator runs the firmware, not a parallel sketch. lib/sim/arduino.ts translates a supported Arduino C++ subset into the existing JS runtime, so src/main.cpp is the simulation source and there is nothing to keep in sync by hand. It is a subset on purpose: unsupported constructs are refused rather than approximated, because a sim that quietly ignores Servo.h is worse than one that says it cannot run.

sketchSourceFor("src/main.cpp", src) // -> { source, warnings, errors }
// delay(500)            -> yield delay(500)      (keeps the virtual clock deterministic)
// helperWithDelay()     -> yield* helperWithDelay()
// Serial.println(x)     -> print(x)
// classes/templates/... -> errors[], nothing runs

2. Schematic parts can carry internals. PartSpec (data, never model-authored code) compiles to the PartModel the engine already consumes — drives, shorts, indicator, analog, interactivity — and is stored on CircuitDoc.models, validated on the way in. A part with neither a built-in model nor a spec is inert and named: "Hall (UNMODELLED)", surfaced in the canvas and the fit check, instead of silently doing nothing. save_circuit/import_wokwi_diagram now return partsWithoutSimulationModel, and a new define_part_models tool fills them in.

3. A fit check that actually runs the thing. evaluateFit(input) compares schematic ↔ PCB ↔ firmware ↔ BOM ↔ CAD and then executes the firmware against the schematic. Pin usage is recorded at the simulator boundary rather than parsed out of the source, so a pin reached through a variable or a helper still counts:

class RecordingSimulator extends Simulator {
  override setMcuDrive(id, pin, drive) { this.touched.add(pin); super.setMcuDrive(id, pin, drive); }
}
// firmware blinks 13, schematic wires the LED to 9 ->
//   error: "Firmware uses U1 pin 13, which nothing is wired to."
// LED wired but never driven ->
//   warning: "LED1 never activated in 3s of simulated run."

Exposed as verify.fitCheck (read-only query), the copilot tool check_integration, and a Verify panel labelled SIMULATED — behavioural, not compiled firmware on hardware.

Prompt. CAD and schematic were serialised behind each other for no reason; a Zoo generation takes ~5 minutes. They now go in the same step, with PCB/firmware after the schematic and check_integration as a mandatory final step that must be re-run after any cross-stage change.

Testing

test/part-spec.test.ts (10) and test/fit-check.test.ts (9) added, alongside the existing test/arduino-translate.test.ts. Full suite 466 passing; prettier, eslint and tsc --noEmit clean. Not exercised in a browser — the Verify panel and canvas changes are untested end-to-end.

Link to Devin session: https://app.devin.ai/sessions/27a2860693d444f8a9a37442a83762d7
Requested by: @alanshen27

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring