Project · executable model stress test
FraudTrust Sample, Synthetic Temporal Stress Study
SYNTHETIC FIXTUREExecutable model · logistic manifest
This is a persistent, seeded sample project built entirely from committed synthetic data, the same generator behind the seeded demo. It is not a real institution, customer, or loss figure, and it is safe to re-open at any time.
reference: synthetic_reference_window.csv (300 rows, 300 labelled) · comparison: synthetic_comparison_window.csv (300 rows, 300 labelled)
- reference · synthetic_reference_window.csv
- comparison · synthetic_comparison_window.csv
Next recommended step: apply a deterministic scenario to the reference window to see how the three evidence channels move. You can also search for a blind spot or view existing evidence cases.
01 · locked
Evidence ingested
600 validated rows · hashes recorded
02 · locked
Execution resolved
Raw features + logistic contract
Evidence channels
Three independent channels, reference window vs. comparison window. Each is computed and reported separately, no combined score.
0.667 → 0.664
0.037 → 0.057
1.000 → 0.996
View as table
| Channel | Metric | Baseline | Stressed | Note |
|---|---|---|---|---|
| Discrimination | PR-AUC | 0.667 | 0.664 | Computed from labelled rows in each synthetic fixture window. |
| Calibration | ECE | 0.037 | 0.057 | Computed from labelled rows in each synthetic fixture window. |
| Explanation stability | ESI (proxy) | 1.000 | 0.996 | Cosine-similarity proxy between the two windows' attribution profiles. This is this app's own simplified formulation, not the dissertation's validated ESI, see /method. |
Execution capability
Validated raw numeric features and a data-only logistic manifest are stored for this project. Supported raw-input operators rerun that exact model.
| Capability | Available | Reason |
|---|---|---|
| PR-AUC and ECE from supplied windows | AVAILABLE | Computed from labels and scores in the validated rows. |
| Attribution-profile stability proxy | AVAILABLE | Exact model contributions or supplied attribution columns are present. |
| Raw-feature mutation and model rerun | AVAILABLE | Validated logistic scoring contract and raw features are attached. |
| Evidence-layer attribution sensitivity | AVAILABLE | Runs directly over supplied or computed attribution evidence; scores stay fixed. |
Attribution profile
Recent scenario runs
View all →Provenance integrity
- reference input hash:
- dac1eae5c70f16623b1139166bdf274d5167c3065e1d4ff3b733ba53d028a9dd
- comparison input hash:
- 0ccbf54d3a6b073ff2f6dd478dc8df0c7759e788041dace0acd2001592c8dee3
- computation version:
- chaos-lab-replay-0.1.0
- output hash:
- e50d44b1d08fe2d5cabcd99374cbd41df932b99cffcdf9bcd56fabdd504c0da1
- evaluated at:
- 2026-11-19T09:00:42.628Z
This is a Mode B executable-model computation over the two uploaded CSVs, see the method page for what discrimination, calibration and ESI-proxy do and do not mean, and for this build's current limitations.