BOXES -- OPTIONAL SEARCH-ADVISOR HANDOFF
Revision: 2026-10-10 (named boxes with symmetry pruning)
Status: Specification complete for the agreed scope; ready for optional search-advisor work.
Preparing this handoff does not invoke the search advisor or authorize experiments.

RECORDS
Requirements: problem-analysis.txt
Current implementation and evidence: spec-analysis.txt
Specification: ../../../probs/problem-boxes-advisor.lisp
Earlier counts/occupancy revision and its checks: git history (commit ec7b9ab
and earlier).

DELIVERED MODEL
Path B specification, no include-tech. Named boxes: (loc object $area),
(holding agent $box), bijective (on $box $plate). Gate openness derived from
plate occupancy. Three unit actions (PICKUP, PLACE, CROSS-GATE); one
initialization action builds per-area plate/gate lists. MIN-STEPS-REMAINING?
is a placeholder bound: 0, 1 or 2 crossings by adjacency to area4, plus the
placement and pickup needed when every gate into area4 is closed.

OBJECTIVE
One globally shortest unit-action path; goal (loc agent1 area4). The no-plate-
stepping, interchangeable-box and no-repeated-arrangement reductions are agreed
in problem-analysis.txt section 4.

SETTINGS AND VALIDATION
File settings: planning, MIN-LENGTH, GRAPH, symmetry pruning on, depth cutoff 0,
16 threads. User-reported: stages and returns the ten-action optimum. No timing
or work counts recorded for this revision.
Reproduction: (stage boxes-advisor) then (solve).

OPEN PERFORMANCE QUESTIONS (search advisor; each needs its own approved plan)
- Does symmetry pruning save more work than it costs as box counts grow?
- Design a useful MIN-STEPS-REMAINING? bound (for example all-open crossing
  distance to the goal, plus unavoidable pickups and placements) and measure
  whether its pruning outweighs its cost as topology and box counts grow.
- Graph memory, 16-thread benefit, and behavior of cyclic or disconnected layouts.
Vary box counts and connectivity separately before studying their interaction.
No size limit or speedup has been established. Fidelity defects return to the
spec review rather than being treated as tuning.
