Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Property & casualty insurance loss reserving in Python. Chain ladder, Bornhuetter-Ferguson, Cape Cod, bootstrap simulation, and loss development pattern estimation. Actuarial triangle operations.
.claude/skills/mkurman-chainladder/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -15% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 4% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 21% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -26% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -41% | 0% |
ChainLadder implements actuarial reserve estimation methods for property & casualty insurance. Use it for loss reserving, claims triangles, and actuarial modeling in Python.
bashuv pip install chainladder
pythonimport chainladder as cl # Load sample auto liability triangle tri = cl.load_dataset("RAA") print(tri) # Select development pattern dev = cl.Development().fit_transform(tri) # Run chain ladder method model = cl.ChainLadder().fit(dev) print(model.reserve_) print(model.ldf_) # age-to-age factors
python# Estimate reserve variability mack = cl.MackChainLadder().fit(dev) print(mack.reserve_) print(f"CV: {mack.reserve_.std() / mack.reserve_.sum():.2%}") print(mack.conditional_standard_error_)
pythonbf = cl.BornhuetterFerguson().fit(dev) print(bf.reserve_) print(bf.expected_loss_) # a priori expected loss
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 13,254 | 8,082 | -39% | 1 | 1 | 0% | 2,251 | 1,471 | -35% | 0 | 0 | — |
case-03 | fail→fail | 12,605 | 8,261 | -34% | 1 | 1 | 0% | 2,149 | 1,603 | -25% | 0 | 0 | — |
case-04 | fail→fail | 13,650 | 14,352 | +5% | 1 | 1 | 0% | 2,319 | 2,331 | +1% | 0 | 0 | — |
case-05 | pass→pass | 20,696 | 13,162 | -36% | 1 | 1 | 0% | 3,040 | 3,068 | +1% | 0 | 0 | — |
case-01 | fail→pass | 14,530 | 10,814 | -26% | 1 | 1 | 0% | 2,333 | 1,979 | -15% | 0 | 0 | — |
case-06 | pass→pass | 10,673 | 11,672 | +9% | 1 | 1 | 0% | 2,223 | 2,776 | +25% | 0 | 0 | — |
case-07 | pass→pass | 4,644 | 1,874 | -60% | 1 | 1 | 0% | 667 | 560 | -16% | 0 | 0 | — |
case-08 | fail→pass | 7,307 | 1,948 | -73% | 1 | 1 | 0% | 550 | 573 | +4% | 0 | 0 | — |
case-09 | fail→pass | 3,948 | 2,462 | -38% | 1 | 1 | 0% | 549 | 663 | +21% | 0 | 0 | — |
case-10 | pass→pass | 7,408 | 4,639 | -37% | 1 | 1 | 0% | 1,065 | 672 | -37% | 0 | 0 | — |
case-11 | fail→pass | 5,393 | 2,706 | -50% | 1 | 1 | 0% | 846 | 626 | -26% | 0 | 0 | — |
case-12 | pass→pass | 4,499 | 4,396 | -2% | 1 | 1 | 0% | 793 | 1,104 | +39% | 0 | 0 | — |
case-13 | pass→pass | 3,028 | 2,549 | -16% | 1 | 1 | 0% | 404 | 657 | +63% | 0 | 0 | — |
case-14 | fail→fail | 8,827 | 7,386 | -16% | 1 | 1 | 0% | 1,505 | 1,657 | +10% | 0 | 0 | — |
case-15 | fail→fail | 9,440 | 5,784 | -39% | 1 | 1 | 0% | 1,624 | 1,302 | -20% | 0 | 0 | — |
case-16 | pass→pass | 5,781 | 1,745 | -70% | 1 | 1 | 0% | 497 | 548 | +10% | 0 | 0 | — |
case-17 | fail→pass | 7,166 | 2,692 | -62% | 1 | 1 | 0% | 1,153 | 678 | -41% | 0 | 0 | — |
case-18 | fail→pass | 17,066 | 8,847 | -48% | 1 | 1 | 0% | 3,501 | 1,864 | -47% | 0 | 0 | — |
case-19 | fail→pass | 3,146 | 2,942 | -6% | 1 | 1 | 0% | 496 | 791 | +59% | 0 | 0 | — |
case-20 | fail→pass | 11,564 | 4,680 | -60% | 1 | 1 | 0% | 1,834 | 1,045 | -43% | 0 | 0 | — |
case-21 | pass→pass | 2,886 | 1,698 | -41% | 1 | 1 | 0% | 339 | 500 | +47% | 0 | 0 | — |
case-22 | fail→pass | 3,791 | 1,869 | -51% | 1 | 1 | 0% | 557 | 519 | -7% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +41 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.