You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Problem.path_strategy="stock_front" is documented as a development strategy: it can stream a productive prefix but raises if coverage or steering is unresolved. engagement_front now does what it was meant to become (plans from the stock, engagement proven, #167/#189). Keeping both is a third adaptive planner (src/adaptive_front.rs, about 4,400 lines) with its own tests and fixtures.
Acceptance.
Anything stock_front proves that engagement_front does not (for example the joint material-side proof, check_front) is either carried over, with a test, or shown unnecessary in this issue.
stock_front, its fixtures and tests are removed for Roughing and AdaptivePocket; docs updated.
Benchmarks unchanged.
Do after the AdaptivePocket issue above (DECISIONS #11: no legacy before 1.0).
I checked the material-side proof in #363. The deleted check_front was a private admission test for proposed stock_front orbits. It asked two questions on the same motion: does the total contact stay below the limit, and is the forbidden side clear throughout? The shared interval traversal was an optimization for ranking those orbit candidates.
Both whole-motion queries survive as EngagementScene::check and check_material_side. The retained finite_line_witness_proves_contact_and_side_without_scanning_older_history test checks both on the same motion in both cutting directions. It also rejects a tighter contact limit despite a passing side check, and rejects the wrong side despite a passing angle check. I ran it on Linux. The stock-driven planner uses its own leading-contact admission; the orbit-ranking traversal has no remaining caller. #363 removes the old strategy and its fixtures. Its benchmark check on the submitted head passed without changed expectations.
Activity
Roadmap: place stevecad#188 and stevecad#190-stevecad#194; drop close…
Roughing: adaptive clearing follows the stock, with its engagement pr…
10-X-eng commentedon Nov 21, 2026
I checked the material-side proof in #363. The deleted
check_frontwas a private admission test for proposedstock_frontorbits. It asked two questions on the same motion: does the total contact stay below the limit, and is the forbidden side clear throughout? The shared interval traversal was an optimization for ranking those orbit candidates.Both whole-motion queries survive as
EngagementScene::checkandcheck_material_side. The retainedfinite_line_witness_proves_contact_and_side_without_scanning_older_historytest checks both on the same motion in both cutting directions. It also rejects a tighter contact limit despite a passing side check, and rejects the wrong side despite a passing angle check. I ran it on Linux. The stock-driven planner uses its own leading-contact admission; the orbit-ranking traversal has no remaining caller. #363 removes the old strategy and its fixtures. Its benchmark check on the submitted head passed without changed expectations.