Design closed-loop start gating at the output layer

Summary

As of 2026-04-11, stimulus.start is implemented for open-loop only. Closed-loop still begins classifier evaluation immediately at launch.

We discussed extending the same scheduling concept to closed-loop, but decided not to implement it yet because the correct semantics depend on future non-laser outputs such as notifications.

Problem

A naive closed-loop implementation would delay the classifier itself until the scheduled start time. That is too blunt if we later want trigger-driven behaviors that should remain active before laser output is armed, for example:

• notifications
• event logging
• counters / monitoring

Desired Direction

When closed-loop scheduling is added, the gate should apply to the output/action layer, not necessarily to classifier evaluation.

This should let us support behavior such as:

• classifier continues evaluating before the start gate
• UI/status can still reflect trigger state before the gate opens
• laser output remains suppressed until the configured start time
• future output modes may choose whether they obey the same gate

Design Questions

• Should gating be configured per output mode?
• Should laser be gated while future notify actions remain active?
• How should pre-gate trigger events appear in the UI and logs?
• Should the trigger/stim status panels explicitly show that stimulation is suppressed by schedule?

Proposed Follow-up

Revisit this after we add multiple trigger output modes. At that point:

1. Define action-layer routing and output modes.
2. Add a closed-loop start gate that applies to laser output.
3. Decide whether other outputs (notify, logging-only actions, etc.) should ignore or honor the same gate.

Current Decision

Do not implement closed-loop stimulus.start scheduling yet. Keep the open-loop implementation. Revisit after output modes are expanded.

Start by reviewing the existing open-loop stimulus.start implementation and the current trigger/output flow. Revisit this issue only after multiple trigger output modes exist; the intended outcome is action-layer routing with laser gating while classifier evaluation and any explicitly ungated outputs remain active before the scheduled start.