gurobi-harness-gatelisted

# Gurobi harness gate (process, not SKU) Pulse: real decisions need proof. The solver already lives in `tools/gurobi_fleet_lib.py`. This skill is the **process hook** so agents do not vibe-route spend or collide on ADB / `.git/index.lock`. ## Commands ```bash node tools/gurobi-harness-gate.js --json node tools/gurobi-harness-gate.js --session-start node tests/test-gurobi-harness-gate.js node tools/gurobi-acceptance-bench.js --json node tests/test-gurobi-acceptance-bench.js node tools/gurobi-solver-touchpoints.js --json node tests/test-gurobi-solver-touchpoints.js bin/agent-loop --health --json # observe.gurobi_proof ``` Wired from `tools/agent-session-start.js` (human `=== Gurobi proof ===` line) and `bin/agent-loop` observe. ## What it certifies | Check | Constraint | |---|---| | Token budget | allocated spend ≤ $10; `huge` ($12) excluded | | Exclusive ADB | at most one of `e2e-a` / `e2e-b` | | Receipt | `proof.certified_optimal` + `obj_bound` from gurobipy | Never mock an optimum. Skip (do not fail the whole session) if gurobipy is missing. ## Phase One Champion Kit (2026-08-19) Steal the **process**, not the sales kit. Do not copy Gurobi PDFs / Wiley book into git. | Kit piece | What we took | What we did not | |---|---|---| | LLMs and Optimization | Four certificates: OPTIMAL, MIP gap, IIS, UNBOUNDED. LLM explains; solver proves. | Intelligence Hub product | | Evaluating and Benchmarking | Frozen set + metrics + 3-seed mean/range + env log + heuristic vs solve