verify-checkpoint

Trace. On load, print first: =============== [observer/skills/verify-checkpoint/SKILL.md] [observer] ===============

Verify Checkpoint

Grader Actor — the second Actor in the observer module, sibling to observe. Where observe coaches through a role, this Actor gates. Its core unit of work is grading one signalled increment: on a checkpoint signal an implementer emits when it pauses having completed one F-requirement, it grades that single increment against the spec on two of reconcile's three checks — does · only — and emits a CLEAR or BLOCK verdict.

That grading logic is the spine, and it is identical across two invocation modes — the mode is only the entry point:

• (a) Per-increment dispatch (first-class) — the delegating context (lsa:delegate) dispatches this Actor once per signalled increment via the Agent tool; the Actor grades that one increment and returns the verdict. No loop is started.
• (b) Standalone /loop rider — this Actor rides the substrate's self-paced /loop (omit the interval to self-pace) and, on each wake, checks for a checkpoint signal, grading each signalled increment as it wakes.

In /loop mode it rides the existing loop; it does not build a scheduler (../../../.lsa/VISION.md principle 9, Substrate-native first).

This is not lsa:verify. lsa:verify is the before-delegation grounding check (is the spec ready to hand to an implementer). This Actor is an after-the-increment check (did the increment the implementer just finished conform), scoped to one F-requirement — the per-increment analogue of lsa:reconcile, which grades the whole plan at the end. It applies does · only only; the third check, all (whole-plan completeness), stays with the final lsa:reconcile (../../../lsa/skills/reconcile/SKILL.md:35).

The checkpoint-signal contract