[Sk1tch]

Problem

next, plan queue, and plan resolve can disagree about the next executable work. A QueueContext resolves the configured subjective target, but queue input resolution ignores that target and uses QueueBuildOptions.subjective_threshold (default 100) instead. Callers that repeat the target in both places work; callers that rely on the shared context synthesize additional assessment requests and can reject the real work shown by next.

A synthetic regression demonstrates a triaged review finding plus an unrelated stale assessment with a score above the default target. Before this change, next shows the finding, plan queue shows the assessment request, and resolving the displayed work cluster reports the assessment as the current next item. The fix leaves assessment data and configuration untouched.

Fix

• Make the shared context authoritative for both plan and subjective target.
• Preserve explicit threshold options when execution/backlog builders create their own context, including normalization through the existing input resolver.
• Add synthetic regressions for context precedence, explicit-option compatibility, agreement between next/plan queue, and normal resolution of the displayed work.

Validation

• Focused queue, planning, resolution, and command tests: 726 passed.
• Required general suite: 5815 passed, 5 skipped, 2 existing failures. Both failures are the duplicated TestCmdReviewPrepare.test_do_run_batches_dry_run_generates_packet_and_prompts cases. They assume an unsorted prompt glob selects the first batch; the same failures were independently reproduced on clean base 3a7735d5 before this change.
• Read-only verification against the originating local state confirms command queue agreement without assessment or plan changes. Public regressions contain only synthetic data.