E2: Planner Determinism & Performance (0.7.x)

Superseded 2026-10-08 by E4 Validation — 0.9.0. S2.1 → S4.1, S2.2 → S4.3/S4.4, S2.3 → backlog. Cancelled.

Goal

Eliminate the wall-clock dependence class in planner termination and the motion-check oversampling tax, exposed by the two pre-existing test failures root-caused during S1.4 verification (2026-09-22): PRM construction budgeted by wall-clock time with FFI-bound validity checkers fails or flakes depending on machine speed/load. Make planner behavior deterministic and ~10× cheaper at the boundary for oxmpl-py / oxmpl-js users.

Origin: defects found under S1.4 — Library hygiene remove println from library code completion verification; evidence in its completion record. Interim in-test mitigations already shipped with S1.4 (~4–5× headroom).

Stories

• S2.1 — Deterministic milestone-count termination for PRM construction (size-M, high priority) — root-cause fix; pull into sprint-001/E1 only if the release gate flakes before S1.5 closes (re-link to E1 if pulled)
• S2.2 — Motion-check oversampling: resolve the ×0.1 factor in all planners (size-S) — semantics change, needs deliberate review
• S2.3 — Goal-region re-sampling for RRT-Connect, OMPL parity (size-M) — found in blind review during S1.2 2026-09-22; single goal root weakens RRT-Connect on sampleable goal regions; extends the lazy goal-seed site created by S1.2

Live status in TaskNotes (query by tag epic/E2); scope to be confirmed at sprint-002 planning — then narrative in oxmpl - roadmap and oxmpl - sprint-002.