risk-stress-tester

Imported from GCagent/nobleport.etf (skills/23-risk-stress-tester/SKILL.md). Install upstream with npx skills add GCagent/nobleport.etf --skill 23-risk-stress-tester. Copyright stays with the author.

Risk & Resilience Stress Tester Skill

Purpose

Refuse the single base case. Run the scenarios that actually kill infill deals in this market — cost, time, rate, absorption, permitting — and show what happens to spread and cash before anyone treats the model as the plan.

When to use

• A feasibility spread exists and is about to be believed.
• A draw package or loan term sheet assumes the base schedule and budget.
• The owner asks "what if" on cost, delay, rates, lease-up, or permits.

When NOT to use

• Replacing the base-case model → keep 18-nano-feasibility intact and layer scenarios on it.
• Day-to-day job-health on an active remodel → 03 / 06.

Inputs

• Base-case cost stack and spread from 18.
• Sequence / critical path from 21.
• Financing assumptions (rate, carry, interest reserve).
• Absorption / rent or sale-pace assumptions.
• Entitlement risk from 19.

Workflow

1. Lock the base case as the control — do not silently restated it.
2. Run five scenarios (at minimum):
   • Cost escalation (default: 10% hard-cost overrun; also materials-only)
   • Schedule delay (inspection/weather/long-lead slip → extra carry)
   • Rate (construction-loan rate +200 bps or the stated shock)
   • Absorption (slower sale/lease-up; hold period extends)
   • Permitting (longer review, extra condition, ADU path fails)
3. Report each scenario's spread, cash trough, and whether the diligence recommendation flips (ADVANCE / HOLD / KILL).
4. Do not average scenarios into a fake "expected case" unless the owner explicitly asks for a probability-weighted view — and then label the weights as assumptions.

Outputs

• Scenario table (driver → delta cost/time → spread → gate impact)
• Base-case vs. stressed comparison
• Which single scenario is the binding constraint

System integration

• Module: nano.stress_tester (STAGED, LOW) — StephanieAgent.
• Runs in the capital/risk layer of nano_infill_chain with 22.
• Feeds 16 five-harness (Execution may PASS while overall stays STAGED) and 06-stephanie-executive risk rollup.

Guardrails

• Stress tests are still STAGED / SIMULATED on hypothetical inputs.
• Do not hide a KILL scenario behind a healthy base case.
• Stephanie still does not authorize acquisition after a "surviving" stress.

Success criteria

• All five drivers are present; each has a numeric effect on spread or time.
• The binding constraint is named.
• A 10% hard-cost overrun is always shown (matches the chain demonstration).