INTERSECT -- OPTIONAL SEARCH-ADVISOR HANDOFF
Revision: 2026-11-07 (revision 2: location8 added; revision 1 2026-11-05)
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 (rev 4)
Current implementation and evidence: spec-analysis.txt (checks 1-14)
Specification: ../../../probs/problem-intersect.lisp
Diagram: intersect-diagram.png

DELIVERED MODEL
Path B: talos tech (gate, plate, step, ladder, beam-relay, beam-crossing,
walkability, visibility) with coordinate-derived geometry, as in rumin-topo.
No problem-specific actions.  Stored state: agent location, held connector,
connector locations, links.  Colors, crossings, receivers, gates and plate are
derived after every action.  Reach is same-location only (reachability tech
omitted; no reach edge would be derived), so windows bar reach by construction.
Link cap: at most 3 links per connector, counting incoming and retained links.

OBJECTIVE
Any plan reaching (has-location agent1 location7); shortest preferred; every
action costs 1.  The gate order G1 -> G2 -> G3 -> G4 is forced (analysis).

ENHANCEMENTS
E1 connector symmetry pruning (*symmetry-pruning* t): safety check passes;
   0 duplicates within depth 7, so benefit unmeasured.
E2 min-steps-remaining? placeholder (0 at location7, else 1): sound; 0 prunes.

SETTINGS AND VALIDATION
File settings: planning, MIN-LENGTH, GRAPH, *max-connector-pairings* 3, symmetry
pruning on, *depth-cutoff* 7 (provisional, smoke limit).  *threads* 16 at the REPL.
Windows (user-run) and cloud SBCL 2.2.9 agree: stages without warnings; depth 7,
16 threads: 5,721 states, 3,534 repeats, 842 cutoff hits, 31 inconsistent states
abandoned (propagation oscillation, physically unstable configurations), no solution.
Reproduction: (stage intersect), (ww-set *threads* 16), (solve).

OPEN PERFORMANCE QUESTIONS (search advisor; each needs its own approved plan)
- Proposed first question: how the reachable space grows with depth beyond 7
  under the link cap, to set a feasible maximum depth for a full solve.
- A stronger min-steps-remaining? bound (e.g. closed gates left on the forced route).
- Whether omitting pickup-retaining-pairings is safe and worthwhile (D3).
- E1 benefit once connectors can be swapped; connect branching at the cap of 3.
No size limit or speedup has been established.  Fidelity defects return to the
spec review rather than being treated as tuning.

REMAINING
A full (test-talos) on Windows for the link-cap tech change (spec-analysis item 9).
