Sources and methods

Budgetary resources, obligations incurred, unobligated balance and gross outlays at TAS grain, as submitted by reporting agencies. The backbone of the execution chain. Every Department account is also kept with its period of availability, as the resources the program-year view rates obligations against.

Source system

USASpending Account Balances (GTAS submission chain)

Grain

Treasury Account Symbol x fiscal year

Path

data/usaspending/warehouse/accounts/file_a

Refresh

daily

Limitations. Carries five agency identifier codes; 011 (Executive Office of the President) is not the Department and is excluded from every DOW scope. The current fiscal year is a period-to-date submission, not a closed year. File A carries no direct/reimbursable attribute: its obligations and resources include reimbursable work, so the direct split shown beside them comes from File B.

Every procurement (P-1), reserve-equipment (P-1R) and RDT&E (R-1) budget line in the eight President's Budget books held here, at the grain the exhibits themselves publish: a book year and a fiscal year. Because each book restates three fiscal years -- prior-year actuals, current-year enacted, budget-year request -- a single fiscal year appears in three successive books with three different numbers, and all three are kept. The exhibit account symbol is resolved to a Treasury account so a budget line can be carried into the execution files, and the Department's own Program Acquisition Cost by Weapon System book is carried alongside as its answer to which of these lines is a major program.

Source system

DoD Comptroller President's Budget justification books, PB2020-PB2027

Grain

Budget line item x President's Budget year x fiscal year

Path

knowledge-bank/DOD-FM-Knowledge-Bank/11-Budget-Justification/_Archive

Refresh

annual, on publication of each President's Budget

Limitations. The P-1R exhibit is National Guard and Reserve equipment already counted inside the P-1 lines, and R-1 rows marked Include-in-TOA = N sit outside total obligation authority; both are kept, flagged is_memo, and excluded from every total. Amounts are in thousands of dollars as published, are total obligation authority rather than outlays, and are not comparable to the execution files without ageing prior-year balances. The link from a weapons-book system to a budget line is derived from shared type designators or shared significant words -- neither source carries the other's key -- and every link records the evidence it rests on.

Obligations decomposed into USSGL undelivered orders (480100 series) and delivered orders (490100 series), plus OMB object class. Shows what stage each obligated dollar has reached.

Source system

USASpending Account Breakdown by Program Activity and Object Class

Grain

TAS x program activity x object class x fiscal year

Path

data/usaspending/warehouse/accounts/file_b

Refresh

daily

Limitations. Object class is reported by the component; unreported and mixed classes appear as their own buckets rather than being redistributed.

Federal action obligations for DoW contract actions. Two vintages are retained so that movement in closed fiscal years is measurable rather than assumed.

Source system

USASpending contract award archive, agency 097

Grain

Contract action (transaction) x fiscal year x vintage

Path

data/usaspending/warehouse/contracts

Refresh

monthly (source publishes monthly full files)

Limitations. Federal action obligation is an FPDS reporting concept and is not the same measure as an account-linked obligation; see the reconciliation. The most recent fiscal year is incomplete.

Federal action obligations for DoW financial-assistance awards — cooperative agreements, project and formula grants, and direct payments. A separate warehouse tree and reporting chain from contracts and File C; not account-linked and not reconciled against anything.

Source system

USASpending financial assistance archive, Department of War / Department of Defense awarding agency

Grain

Assistance transaction x fiscal year x vintage

Path

data/usaspending/warehouse/assistance

Refresh

as vintaged (two vintages currently retained)

Limitations. Rows are filtered to awarding-agency name ‘Department of Defense’ or ‘Department of War’ rather than an agency-identifier code, because this warehouse cut carries no code column. The assistance-type codes present are cooperative agreements, grants and direct payments — no loan or loan-guarantee type codes appear in this cut, so it is not a credit-programs picture and must not be described as one. Only two vintages are retained so far; vintage drift is measurable but thin.

Contract obligations that components have linked to a Treasury account. Compared against the award files to measure linkage completeness.

Source system

USASpending Account Breakdown by Award

Grain

Award x TAS x fiscal year

Path

data/usaspending/warehouse/accounts/file_c_contracts

Refresh

daily

Limitations. A low linkage percentage measures reporting completeness between two chains. It is not an estimate of unsupported or missing obligations, and must never be presented as one.

Contract obligations cut by FPDS dod_acquisition_program_code -- the only field in this warehouse keyed to a budget line rather than to an account -- with the share of each total whose funding Treasury account is actually named, the accounts named on those actions, contract-level concentration, and the File C tie-out for the same contracts.

Source system

USASpending contract award archive; USASpending File C

Grain

Acquisition program code x fiscal year

Path

data/usaspending/warehouse/contracts + accounts/file_c_contracts

Refresh

daily

Limitations. FPDS records 'no acquisition program' as code 000 / description NONE, and roughly three quarters of DoD contract dollars carry that code, so this covers only the program-attributable quarter of the file -- dm_program_coverage states how much per year. federal_accounts_funding_this_award lists every account funding an award without apportioning the obligation across them, so no figure here is split by account. A program obligation total is not comparable to that program's appropriation: it draws on prior-year balances this cut cannot age, and Foreign Military Sales trust dollars (011-8242) sit on the same contract actions as Department appropriations.

The join that lets a President's Budget line be followed into the contract file. The -1 exhibits name budget lines; FPDS names acquisition program codes; neither carries the other's key, so each row here records the evidence the link rests on -- a shared type designator, or identical names.

Source system

Derived from the President's Budget -1 exhibits and the FPDS contract archive

Grain

Budget line item x FPDS acquisition program code

Path

derived: budget_exhibits x program_execution

Refresh

with each exhibit or contract refresh

Limitations. Derived, not published by the Department. Only a shared type designator or an identical name is accepted, and a designator naming more than one program code produces no row rather than a guess -- so this covers the lines whose identity is unambiguous in both vocabularies and no others. A link means the two names refer to the same system, not that the budget line's appropriation funded the contracts under that code.

Definitions with authorities, document inventory by collection, FY2027 justification exhibits, and the congressional hearing record.

Source system

Local knowledge bank; curated DoD-FM wiki pages

Grain

Document / wiki page

Path

knowledge-bank/

Refresh

daily

Limitations. Hearing dates are the acquisition date embedded in the GPO filename, not the date the hearing was held; they are labelled as ingest dates and never as hearing dates.

Audit opinion, material weakness counts and scope limitations, read from the Agency Financial Report rather than inferred from file counts.

Source system

DoD FY2025 Agency Financial Report; DoD OIG

Grain

Reported metric

Path

knowledge-bank/DOD-FM-Knowledge-Bank/07-Financial-Statements-and-Audit-Reports

Refresh

annual

Limitations. Read via the curated wiki summary of the AFR, which is a secondary read of a primary document. Each row records which it is.

Notices of Findings and Recommendations issued, reissued and closed by fiscal year, the per-entity tables the reports publish, and the Agency-Wide material weakness roster for each year a roster is printed. Around each FY2025 weakness, a ten-element audit-risk object: account, assertion, risk, control, failure, root cause, exposure, evidence, remediation, outcome.

Source system

DoD OIG annual audit-result reports FY2018-FY2025; GAO

Grain

Fiscal year; reporting entity; material weakness

Path

database/seed_nfr.json, built by scripts/build_nfr_seed.py

Refresh

annual

Limitations. Individual NFRs are not public documents. Nothing here is an extracted notice: the OIG publishes counts and material weakness narratives, so the finest grain available is the material weakness. Every element records whether it is reported in the cited document or read out of it, and the outcome element is computed from the rosters rather than asserted. FY2018 is published at year grain only, because its per-entity table does not foot to its own published total. FY2023 publishes no roster.

Every column each source carries, how often it is populated, three real values, and whether this site reads it - plus the records that demonstrate each broken join.

Source system

USASpending account and award files as held in the local warehouse

Grain

source file x column

Path

accounts/ and contracts/

Refresh

with the warehouse

Limitations. populated_pct, distinct counts and sample values are measured on the first batch of rows of one partition rather than the whole file, and rows_scanned states how many. A parquet row group in these files is the whole file, so a complete profile costs more memory than the extract host has.

Every justification book in the archive, held at (book, President's Budget year) and never collapsed: the sections each edition prints, in the order it prints them, with the word band each runs to, how much of it carries dated or scheduled language, and the format drift between editions. Plus the exhibit grain underneath it — R-2 and R-2A exhibits from the current Defense-Wide RDT&E books, with their bodies.

Source system

DoD Component budget justification books (comptroller.defense.gov), FY1998–FY2027

Grain

book × President’s Budget year × section

Path

knowledge-bank/DOD-FM-Knowledge-Bank/11-Budget-Justification

Refresh

annual, on publication of each President's Budget

Limitations. Read from the published PDFs by layout extraction, so a table section keeps its source spacing rather than being re-tabulated. The archive is the Defense-Wide and Department-level material: a service's own book appears only where it was collected (a few FY2003–FY2005 O&M volumes), so the picker shows what is held rather than what the Department publishes. Structure and measurements come from every edition; example TEXT is kept only for a book's two most recent editions, because the archive is several gigabytes of prose. A book with no text layer (an older scan) contributes page counts and nothing else, and says so.

All seven "-1" display tables -- C-1, M-1, O-1, P-1, P-1R, R-1, RF-1 -- for the latest two President's Budget books, keeping the hierarchy the exhibit is printed in so the request can be drilled into rather than listed flat. Each row carries the fiscal year's total, its mandatory half, the column the figure came from, and whether the row is money, a negative offset the published total includes, or a memo restatement that must never be summed.

Source system

DoD Comptroller budget materials, FY2026 and FY2027 President's Budget

Grain

book x exhibit x appropriation x budget activity x sub-activity x budget line item x fiscal year

Path

knowledge-bank/DOD-FM-Knowledge-Bank/11-Budget-Justification/_Archive

Refresh

annual, on publication of each President's Budget

Limitations. A display table is what is being asked for, not what was executed or enacted; the FY(pb-1) column is an enacted position as of the book's own cut-off and is restated in the following book. The seven exhibits do not add to a Departmental topline: P-1R restates equipment already inside P-1, and RF-1 covers revolving funds outside the appropriation titles. Discretionary is derived as the year's total minus its mandatory columns, because several exhibits publish alternative measures of the same money side by side rather than components of a sum.

File B published at the grain it is reported at rather than as one row per fiscal year: obligations, undelivered and delivered orders unpaid, prepaid and paid outlays, deobligations, and the upward and downward adjustments to prior-year orders, for every Treasury account in Department scope. Each account carries the period of availability read as a fund life, because annual authority expires on 30 September and multi-year and no-year authority does not. Each account also carries its program year -- the beginning of its period of availability -- so a fiscal year's execution can be split into the current year's money and the prior years' money still executing in it.

Source system

USASpending account breakdown by program activity and object class (File B)

Grain

fiscal year x Treasury account x object class x program activity x direct or reimbursable x emergency fund code

Path

data/usaspending/warehouse/accounts/file_b

Refresh

monthly, as the Department's submissions are published

Limitations. One submission per fiscal year is held, so there is no within-year series in the account chain at all -- this file gives the Department's position, never a curve, and within-year timing can only be read from contract action dates, which cover a third of obligations. The current fiscal year is period-to-date at its latest submission and is not comparable with a closed year without saying so. From the FY2026 submission the program activity name is null on every row, so an activity can be identified by its reporting key but not read. Rows that differ only by reporting key and repeat one figure are counted once, which is a repair of a source defect rather than a property of the file. Detail is carried for the three most recent fiscal years; earlier years are held as account and object-class rollups. Direct and reimbursable obligations are carried separately and every execution figure defaults to direct: reimbursable work is paid back by a customer, and when the customer is another Department account the same work is already that customer's direct obligation.

When contract obligations moved: the day-by-day cumulative curve for each fiscal year, monthly totals by awarding sub-agency, office, product or service code and supply group, recipient, pricing type and extent of competition, the September and final-five-day concentration of each category, and a set of signals computed against each category's own prior years.

Source system

USASpending contract award files (FPDS-NG)

Grain

fiscal year x day, and fiscal year x month x category

Path

data/usaspending/warehouse/contracts

Refresh

monthly, with each warehouse vintage

Limitations. Contract obligations are a third of Department obligations -- 33.9% in FY2025 -- and are not the whole of execution. The largest block they do not cover is personnel compensation and benefits, 40.4% of that year, which is paid on a schedule. There is no Department-wide within-year series in these sources to compare against: File A and File B each publish one submission per fiscal year, so the account chain gives a position and never a curve. A signal is a question rather than a finding: none of them observes impropriety, and a large September action is very often a multiyear definitisation or an exercised option with nothing unusual in it beyond its size. Deviations are measured against a category's own history, so a category held in fewer than three complete years produces none. The live fiscal year is shorter than the years it is compared with, and every comparison is made to the reporting frontier over whole months only.

Every agency submission window with the date its data was revealed to the public record. It is the only thing here that says what is AVAILABLE, as against what this warehouse happens to hold, and it is what makes the currency of every account figure on this site measurable rather than assumed.

Source system

USASpending API, /api/v2/references/submission_periods/

Grain

fiscal year x fiscal period

Path

https://api.usaspending.gov/api/v2/references/submission_periods/

Refresh

monthly, with each submission window

Limitations. Fetched over the network and therefore the one extract here that can fail without the load failing: where it cannot be reached nothing is written, the previous calendar stays with its own fetch date, and the page reports on that date rather than implying it is current. It describes the government-wide calendar, not any one agency's compliance with it -- a period being revealed does not guarantee that the Department submitted into it.

The files the extract reads, before anything is done to them. For every source a figure on this site comes from -- File A, File B, the three File C award-financial files, the FPDS contract transactions, the assistance transactions, each President's Budget exhibit workbook, and the submission calendar -- the first five records with every column the file carries, in file order and under the file's own names. Parquet samples come from the newest fiscal-year partition (or, for award files, the newest vintage and fiscal year); workbooks from their first sheet, with the rows printed above the header kept. Values are stored as the text the file holds.

Source system

USASpending account and award downloads (parquet warehouse); OUSD(C) President's Budget -1 exhibit workbooks; USASpending API

Grain

One row per source file (dm_raw_source) and one row per sample record (dm_raw_row): the first five records of each file, every column

Path

data/usaspending/warehouse/**; knowledge-bank/DOD-FM-Knowledge-Bank/11-Budget-Justification/_Archive/FY*/_Year-Level/*.xlsx; api.usaspending.gov/api/v2/references/submission_periods/

Refresh

With every extract (scripts/etl_analytics.py --step raw, included in --step all)

Limitations. A sample, not the file: five records show what the columns are and what the values look like, not what the file totals. The first records of a file are whatever order the file is stored in, not a representative draw. The warehouse is parquet converted from the USASpending CSV downloads with column names unchanged; the contract and assistance warehouse keeps a subset of the download's columns (64 and 30), so a column USASpending publishes and the warehouse did not keep does not appear. Workbook cells are shown as stored, so an amount the exhibit prints in thousands is shown as the thousands figure. Only the first sheet of each workbook is sampled; C-1, which the extract reads sheet by sheet, lists its other sheets by name.

Three populations kept apart. Dated: every contract action by fiscal month and by FUNDING sub-agency -- the fund holder, and the only place WHS, MDA and SOCOM are named in any source here; about a third of Department obligations. Annual: File A at account grain by programme year and colour of money, the whole population and no within-year timing at all. Modelled: the annual totals spread across the months using the dated layer's shape, marked as modelled on every row and never summed into an observed total. Beside all three, the appropriation calendar -- every continuing resolution, lapse and full-year enactment for FY2021-FY2026, each cited to its public law.

Source system

USASpending contract file (FPDS) + File A + File B, with a curated appropriation calendar

Grain

Fiscal month x fund holder; fiscal year x programme year x appropriation x agency

Path

contracts/, accounts/file_a, accounts/file_b, database/seed_approp_calendar.json

Refresh

With the warehouse; the calendar is curated and changes when a law is enacted

Limitations. A contract action names exactly one Treasury account on 8.5% to 23.0% of its dollars depending on the year, so the dated cut by colour of money and programme year is a SAMPLE whose size moves with the year: it carries shape, never level, and is never a denominator. File A and File B hold one submission per fiscal year and contain no within-year series at all. The model is withheld for a year in progress, because File B's figure is then period-to-date and the contract shape stops at the reporting frontier.

For the year's OWN appropriation: when authority arrived and at what rate, when execution actually started and on what type of spending, how long each observable link took, and what a continuing resolution costs in execution days measured from this Department's own six years. Four bases, never mixed: File A and File B measured and complete but carrying no within-year timing; the appropriation calendar measured and dated; the execution ramp measured but censored, an upper bound taken from the fraction of contract dollars that name their Treasury account; and the interior of the distribution chain assumed, because apportionment, allocation, allotment and sub-allotment appear in no published file.

Source system

File A + File B + FPDS action dates + the curated appropriation calendar

Grain

Fiscal year x component x colour of money x object class major group

Path

accounts/file_a, accounts/file_b, contracts/, database/seed_approp_calendar.json

Refresh

With the warehouse; the calendar changes when a law is enacted

Limitations. Apportionment, allocation, allotment and sub-allotment are published nowhere here, so only the endpoints of the chain are observed and the interior is an assumption with its authority printed beside it. The ramp is censored: only 1-6% of current-year dollars name their Treasury account, the attribution is award-level rather than action-level, and an earlier obligation that named no account is invisible. File A and File B hold one submission per fiscal year and contain no within-year series at all.

Conditions in appropriations law and the accompanying joint explanatory statements that gate obligation. Two kinds, kept apart: a BAR prevents obligation until the condition is met; a reporting direction burdens the Department without, on its own words, withholding the money. Each row says whether its requirement is the published text or a description of it, and the scope is measured from File A wherever a Treasury account carries the gate.

Source system

Department of Defense Appropriations Acts and their joint explanatory statements; 31 U.S.C.; OMB Circular A-11; DoD FMR

Grain

One condition standing between an appropriation and an obligation

Path

database/seed_legislative_gates.json

Refresh

Annually, when a new act and statement are enacted

Limitations. Curated by reading the acts and statements, not extracted mechanically, so it is a selection rather than a complete inventory of every congressional direction. Reprogramming thresholds by appropriation type are not published in a source this site could reach and are therefore not stated. Where a gate sits inside a component's accounts rather than in one of its own, File A cannot isolate its scope and none is given.

The argument published at /research. The prose is written once, by hand, and contains NO NUMBERS: every figure in it is a named placeholder resolved at extract time from the same staged extracts that feed the rest of the site. A finding carries the claim, what follows from it, and the observation that would falsify it. A recommendation names exactly one finding, and the load refuses to publish a recommendation whose finding does not exist or a sentence whose placeholder did not resolve.

Source system

This site's own extracts, argued over in database/seed_research.json

Grain

One finding, one recommendation, or one resolved figure

Path

database/seed_research.json + every staged extract of the same load

Refresh

Every load

Limitations. The figures move with the data even though the sentences do not: the prose is fixed and every number in it is resolved afresh at each load. The findings are this site's readings of its own published measurements, not the Department's, and they inherit every limitation of the sources beneath them. Recommendations are judgements rather than measurements and are labelled as such. The cost and horizon attached to each are estimates from the outside; nobody here has priced them from inside the Department.