Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Synthetic Data Vault (SDV) — generate synthetic tabular data. Single-table, multi-table, and sequential data synthesis. CTGAN, TVAE, CopulaGAN, GaussianCopula. Privacy metrics and evaluation.
.claude/skills/mkurman-sdv/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -38% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -9% | 0% |
The Synthetic Data Vault (SDV) generates synthetic tabular data that preserves statistical properties while protecting privacy. Supports single-table, multi-table, and sequential data generation with CTGAN, TVAE, CopulaGAN, and GaussianCopula models.
bashuv pip install sdv
pythonfrom sdv.single_table import CTGANSynthesizer from sdv.datasets.demo import load_demo data, metadata = load_demo(dataset="census") synth = CTGANSynthesizer(metadata) synth.fit(data) synthetic = synth.sample(num_rows=500) print(synthetic.head()) print(f"Original columns: {data.shape}, Synthetic: {synthetic.shape}")
pythonfrom sdv.multi_table import HMA1Synthesizer synth = HMA1Synthesizer(multi_table_metadata) synth.fit(multi_table_data) synthetic = synth.sample(scale=0.5)
pythonfrom sdv.evaluation import evaluate # Statistical similarity report = evaluate(synthetic, data, metadata) print(f"Overall score: {report.get_score():.3f}") print(f"Column shapes: {report.get_property('Column Shapes'):.3f}") print(f"Column pairs: {report.get_property('Column Pair Trends'):.3f}")
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 13,881 | 7,586 | -45% | 1 | 1 | 0% | 2,528 | 1,703 | -33% | 0 | 0 | — |
case-02 | pass→pass | 9,754 | 6,279 | -36% | 1 | 1 | 0% | 1,748 | 1,473 | -16% | 0 | 0 | — |
case-03 | pass→pass | 45,194 | 12,989 | -71% | 1 | 1 | 0% | 2,654 | 2,870 | +8% | 0 | 0 | — |
case-04 | fail→pass | 12,361 | 7,341 | -41% | 1 | 1 | 0% | 2,492 | 1,718 | -31% | 0 | 0 | — |
case-05 | fail→fail | 10,858 | 7,033 | -35% | 1 | 1 | 0% | 2,273 | 1,775 | -22% | 0 | 0 | — |
case-06 | pass→pass | 3,137 | 2,094 | -33% | 1 | 1 | 0% | 401 | 653 | +63% | 0 | 0 | — |
case-07 | pass→pass | 6,249 | 2,717 | -57% | 1 | 1 | 0% | 1,089 | 815 | -25% | 0 | 0 | — |
case-08 | pass→pass | 4,694 | 2,217 | -53% | 1 | 1 | 0% | 747 | 744 | -0% | 0 | 0 | — |
case-09 | fail→pass | 8,525 | 3,601 | -58% | 1 | 1 | 0% | 1,594 | 993 | -38% | 0 | 0 | — |
case-10 | fail→pass | 8,310 | 6,452 | -22% | 1 | 1 | 0% | 1,413 | 1,542 | +9% | 0 | 0 | — |
case-11 | pass→pass | 11,970 | 4,240 | -65% | 1 | 1 | 0% | 1,796 | 968 | -46% | 0 | 0 | — |
case-12 | pass→pass | 8,565 | 3,154 | -63% | 1 | 1 | 0% | 1,508 | 897 | -41% | 0 | 0 | — |
case-13 | fail→pass | 5,235 | 3,212 | -39% | 1 | 1 | 0% | 799 | 854 | +7% | 0 | 0 | — |
case-14 | pass→pass | 4,980 | 2,338 | -53% | 1 | 1 | 0% | 758 | 687 | -9% | 0 | 0 | — |
case-15 | pass→pass | 6,254 | 2,990 | -52% | 1 | 1 | 0% | 1,128 | 844 | -25% | 0 | 0 | — |
case-16 | fail→pass | 14,509 | 2,733 | -81% | 1 | 1 | 0% | 942 | 853 | -9% | 0 | 0 | — |
case-17 | pass→pass | 11,589 | 5,121 | -56% | 1 | 1 | 0% | 2,118 | 1,453 | -31% | 0 | 0 | — |
case-18 | pass→pass | 8,120 | 3,296 | -59% | 1 | 1 | 0% | 1,407 | 936 | -33% | 0 | 0 | — |
case-19 | pass→pass | 6,768 | 3,798 | -44% | 1 | 1 | 0% | 1,088 | 1,024 | -6% | 0 | 0 | — |
case-20 | fail→pass | 11,450 | 2,029 | -82% | 1 | 1 | 0% | 1,935 | 593 | -69% | 0 | 0 | — |
case-21 | pass→fail | 7,913 | 4,605 | -42% | 1 | 1 | 0% | 1,227 | 1,177 | -4% | 0 | 0 | — |
case-22 | pass→pass | 5,000 | 2,023 | -60% | 1 | 1 | 0% | 833 | 695 | -17% | 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, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +23 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.