Provenance
The parameter ledger
Every number in the simulator, with where it came from and how much to trust it.
A model you can't audit is an opinion with sliders — this page is the audit.
Machine-readable copy: /gibill/data/ledger.json.
Grades
| A | Stated directly in RFI 36C10D26Q0163 (VA/VBA, July 16, 2026) — the primary source. Quotation given verbatim. |
| D | Modeling assumption, published as such: chosen for calibration or normalization, sourced to no document, and open to challenge. |
There are deliberately no B or C grades yet: version 0 uses only the RFI plus declared assumptions. As VBA workload reporting and GAO/OIG material is folded in, entries will land between those poles with their own citations.
Stated in the RFI (grade A)
| Parameter | Value | Basis (quoted from the RFI) |
|---|---|---|
| Baseline end-to-end automation rate | ≈60% | “an overall automation rate of approximately 60% across all benefit chapters” |
| Automated claim resolution time | ≈1 day | “Automated claims are resolved in approximately one day across all chapters” |
| Manual adjudication time | 9.5–18.7 days | “compared to 9.5–18.7 days for manually adjudicated claims, depending on chapter” |
| Automation quality rate | 95%+ | “overall automation quality rate of 95+%” — modeled as 5% of automated claims offramping to the manual queue |
| Annual benefits processed | $12B+ | “processes more than $12 billion in education benefits each year” (context; not a model input) |
| Benefit chapters | 30, 33, 35, 1606 | Title 38 chapters named in the background; the model is chapter-agnostic because the chapter mix is not published |
| Platform inventory | 700+ repos, 50+ DBs, 76 EC2, 590+ artifacts, 30+ S3, 90+ licenses | Current-systems configuration (context for the migration questions; not a model input) |
Declared assumptions (grade D)
| Parameter | Value | Why this value |
|---|---|---|
| Baseline arrivals | 100 claims/week | Normalization unit. The RFI publishes no absolute volumes, so the model publishes none; every backlog figure is in these units. |
| Manual hands-on/in-process time | lognormal, median 13.0d, σ 0.45 | Calibrated so the baseline scenario's manual median (14.1d) sits inside the RFI's stated 9.5–18.7d band; queue wait supplies the remainder. |
| Baseline examiner utilization | 90% | Staffing assumption that defines capacity = 1.0; the capacity slider scales it 60–160%. |
| Enrollment-surge window | 9 weeks | Fall-enrollment seasonality is directionally well known in education claims; its magnitude is a slider (×1.0–×3.0), not an estimate. |
| Stochastic forms | Poisson / binomial / lognormal | Standard queueing choices; 200 Monte Carlo runs per scenario, fixed seed 36101263 for reproducibility. |
Known limits
- The model treats examiners as a single pooled capacity; the real system routes by chapter and competency (the RFI's Workload Manager section is precisely about this routing).
- Original vs. supplemental claims are not distinguished; the RFI defines both but publishes no split.
- Automation-rate changes are applied instantaneously; in reality they arrive as releases over a contract's period of performance.