# Load-Bearing Icing

> A red-team check for a gingerbread house plan. Enter the front width, side depth, wall height, roof peak, dough thickness, glue (royal icing, melted sugar or soft frosting), how long the walls set before the roof goes on, the roof candy (none, light, heavy), the display days, the room (dry, ordinary, steamy) and a short note. The page draws the front to scale on a cake board, packs every panel onto 38 x 28 cm trays, rates 6 load-bearing assumptions low, medium or high by fixed rules, ranks them by impact x likelihood x cheapness to test, and gives a free verdict. One metered run makes the honest case for each assumption at medium or high risk, attacks it, and gives the cheapest test before baking, a give-up line and the fix.

URL: https://load-bearing-icing.skillsafe.ai/
API: https://load-bearing-icing.skillsafe.ai/api.html

## The risks and the verdict rule (free, in the browser)
- Walls stand: wall height over dough thickness; over 22 medium, over 30 high.
- Roof stays put: roof angle from the peak over half the depth; 50 degrees or more high; 40 or more, or a roof panel longer than 40 times the dough thickness, medium.
- Glue holds: soft frosting high; royal icing with under 2 hours before the roof medium.
- Roof takes the candy: heavy candy high on dough under 6 mm, medium otherwise.
- Lasts the display: steamy room medium (high past 3 days); ordinary room medium past 7 days, high past 14; dry room medium past 28.
- Panels fit the tray: a panel bigger than 38 x 28 cm high; 3 or more trays medium (packing estimate, 1 cm roof overhang each side).
Score = impact (walls, roof, glue 3; candy, display 2; tray 1) x likelihood (low 1, medium 2, high 3) x cheap to test (glue, tray 3; walls, roof, candy 2; display 1); ranked highest first, ties in the order above.
Verdict, checked in order, first match wins:
1. Redesign - a high risk on walls, roof or glue.
2. Test first - any assumption at medium or high risk.
3. Build it - otherwise.

## The metered run (task "redteam")
Input strings: task, plan (the page's lines: fixed numbers, the house line with its verdict, 6 ranked assumption lines), notes. Reply: one JSON object with the verdict copied from the page, a headline, attacks (1 per assumption at medium or high risk, in rank order: assumption, risk, steelman, attack, cheapest_test, give_up_if, fix) and a closing. The page re-checks the verdict, the order and risk of every attack, every number, the word limits and the glue named.

## Sources
- Built on the SkillSafe skill @phuryn/strategy-red-team: https://skillsafe.ai/skill/@phuryn/strategy-red-team

For fun and home baking, not professional or food-safety advice. Melted sugar and hot trays burn; keep children and pets away from them.