Drift
How Drift is built
Methodology & limitations — how every figure is sourced and reconciled, the rules that keep it honest, what is deliberately left out, and where the numbers still need a final check.
What Drift measures
Drift tracks cost growth in Major Defense Acquisition Programs (MDAPs) — the gap between what a program was first promised to cost and what it costs now. Each program contributes two anchor figures: a baseline (the original total acquisition cost, set at Milestone B / the first Acquisition Program Baseline) and a current estimate (the latest total). The percentage change between them is the program’s cost growth.
Where the numbers come from
Every figure is reconciled to a public primary source. No classified or CUI data is used.
- GAO — Weapon Systems Annual Assessments and program-specific reports
- DoD Selected Acquisition Reports (SARs) — the quarterly cost filings, read directly where accessible (DTIC mirrors, the Wayback Machine) or via GAO/CRS where the whs.mil originals block automated access
- CRS program backgrounders · CBO independent cost estimates · DoD Comptroller budget exhibits (for the counter-UAS appropriations panel)
Each program in the dataset carries its own citation, browsable in the Provenance ledger and the in-app Table; document numbers link straight to GAO, CRS and DTIC.
The three rules that keep it honest
1. Never mix then-year and base-year dollars. SARs report costs two ways — then-year (TY), which carries inflation, and base-year (BY), which strips it to a fixed year. Comparing a BY baseline to a TY current manufactures growth out of inflation alone. Every program records its cost basis, and growth is only quoted where both endpoints share one. The Real · FY24$ toggle goes further, deflating every figure to constant 2024 dollars so you can see how much of a headline number is real — F-35’s +108% becomes roughly +27%; Sentinel’s +81%, already in constant dollars, holds.
2. Separate quantity growth from cost overrun. A program’s total can rise simply because it bought more units. The Waterfall and Quadrant views split every program into a quantity effect and a unit-cost effect — the difference between the F-22 (bought a quarter of the planned fleet, unit cost tripled) and GMLRS (bought 2–3× more rockets at a flat unit price). One is a cautionary tale; the other is disciplined scaling. A big percentage is not, by itself, evidence of mismanagement.
3. Exclude rather than fabricate. Where no honest baseline-vs-current pair exists on a single stated basis, the program is left out — and the reason is documented (below). Nothing on Drift is our estimate.
Data quality, right now
These counts are generated from the live dataset, not typed by hand.
Still awaiting a final SAR check (4). These rows rest on GAO / CRS / CBO reporting; their primary figure sits in a whs.mil-hosted SAR that blocks automated access, so they are flagged in the app and should be confirmed against the SAR itself before being cited to the dollar. They are shown, clearly labelled — not hidden:
LCS · CH-47F · Comanche · B-52 CERP
Deliberately excluded
These are left out because no honest cost-growth pair exists for them — not because they lack interest:
- B-21 Raider — acquisition and procurement costs are classified; two public anchor points sit in different bases, which is not a cost-growth curve.
- FARA — cancelled in development, zero delivered, no current total.
- V-22, F/A-18E/F — basis mismatch / no clean single current total.
- Virginia SSN, DDG-51 Flight III — no clean single baseline (Virginia is also an acquisition success, not growth).
- NSSL / EELV (space launch) — no documented inception baseline on a comparable basis; the only public pair is a single-period SAR delta, not lifetime growth.
- SM-3, Next-Gen OPIR, GPS IIIF, PAC-3 MSE, AMPV, and others — SAR-exempt, R&D-window only, or a quantity-scope artifact that looks like growth but isn’t.
The full excluded list, with reasons, is in the data notes.
Real dollars — the honest limitation
The Real · FY24$ toggle re-expresses every figure in constant FY2024 dollars using the OMB GDP (chained) price index, so cost growth reads as real growth with inflation stripped out. This is a point-deflation: a then-year program total is a sum of dollars spent across many years and already embeds projected inflation, so deflating it by a single year’s index is an approximation, not a rigorous year-by-year reconstruction. Read real-mode growth as directional — how much of a nominal number is inflation — not as a precise statistic. The true real figure for a long program sits between its nominal and point-deflated values.
The counter-UAS programs are not in the cost-growth core
LIDS, FS-LIDS, M-LIDS, Coyote and KuRFS are an emerging frontier, but they are not MDAPs — no SAR, no Acquisition Program Baseline, and not individually assessed in GAO’s annual review. There is no certified baseline-vs-current pair to plot, and Drift does not invent one. The counter-UAS panel charts the real President’s-Budget appropriation lines instead, labelled explicitly as budget appropriations, not acquisition cost growth.
What Drift is not
- Not a blame index. Cost growth has many causes — quantity, scope, requirements churn, inflation. Drift surfaces the mechanism; it does not assign fault.
- Not contractor-level. SAR data is program-level, not blame-the-prime.
- Not the final word. The flagged rows above should be confirmed against the SAR before being cited to the dollar.
Reproducibility
The dataset is a typed, open-source module; the whole app is static. Download the current data as CSV or JSON from the toolbar, or read every row’s source in the Table and Provenance views. Corrections are welcome — the point is to be checkable.
Drift — an independent project by Aous Abdo, built from public GAO, CRS and DoD sources. Cost growth has many causes; a larger number is not by itself evidence of mismanagement. How this is built · Explore the data →