Rayforge-to-JCZ semantic equivalence spike

Date: 2026-10-18
Rayforge commit: c28918c87f8273dd1d244f5a057addc192807fa5
MeerK40t commit: d511944e08788e88077727f682957759884ff5ca
Scope: deterministic semantic encoding with no transport

Result

A real Rayforge raygeo.ops.Ops stream was constructed with job boundaries, 50 percent power, 250 mm/s feed, 40 kHz frequency, 200 ns pulse width, a synthetic square, a one millisecond dwell and job end. The external addon encoded it into a deterministic semantic trace without discovery or transport.

The encoded values matched the pinned BalorMK mock golden trace for all six checked semantics:

Concept	Rayforge input	JCZ/Balor semantic value
Power	`0.5` fraction	`0x0800`
Mark speed	`15000` mm/min	`0x0094` for synthetic 110 mm field
Frequency	`40000` Hz	Q-switch period `0x01f4`
Pulse width	`200` ns	`0x00c8`
Geometry	four marks	matching `0x7000`/`0x7100` coordinates
Dwell	`1` ms	`0x0064` ten-microsecond units

Two runs produced identical results with SHA-256 6507c887b624a28820a79cf92a49b12b518a47aca7612a17af1f4ba5302b1e8d. The checked result is tests/fixtures/rayforge-jcz-equivalence.json.

Unit finding

Rayforge's laser-step UI contract labels pulse width as nanoseconds, matching BalorMK's MOPA UI. However, the current Ruida RPA encoder docstring calls the same Ops.pulse_width value microseconds. The spike follows the laser-step contract and records this disagreement as an upstream unit-clarity gap. A production adapter must not infer the unit from a driver-specific comment.

What this proves

Rayforge Ops carries the required frequency and pulse-width state into an external binary encoder.
The selected values can map deterministically to the same controller-level semantics emitted by the pinned BalorMK mock implementation.
Coordinate, speed and dwell conversions can remain inside a small encoder boundary with no USB dependency.

What this does not prove

The semantic trace is now accompanied by separately documented 12-byte command records; neither artifact is a complete production command list.
The 110 mm field is synthetic and applies no lens correction.
No machine/source/lens profile validation is implemented.
Unsupported Ops, cancellation, status, list buffering and error recovery still need explicit behavior and negative tests.
No USB enumeration, controller command, motion or emission was attempted.

Decision impact

Spike C passes at the semantic boundary. The result supports a Rayforge JCZ addon architecture. A follow-up spike now covers transport-free 12-byte command records. At this checkpoint, list behavior, full Windows packaging and the remaining stock MeerK40t workflow assessment were still open. Later offline progress is summarized by the M1 no-device acceptance checkpoint; ADR 0002 remains proposed because its external gates are still open.

2.9 KiB Raw Blame History