Control over reporting

Asserts. Obligations incurred plus unobligated balance equals total budgetary resources, within 0.1%.

Why. The status section of the SBR must equal the resources section. A break means the extract has dropped or double-counted a TAS.

Authority · OMB Circular A-136; DoD FMR Vol 6A

Asserts. Appropriated authority, unobligated balance brought forward, adjustments, and other budgetary resources sum to total budgetary resources, within 0.5%.

Why. Proves the resources waterfall shown on the execution page is complete. Note that the source's other-budgetary-resources column already rolls up borrowing authority, contract authority and spending authority from offsetting collections; adding those separately double-counts them.

Authority · OMB Circular A-136

Asserts. Gross outlays are less than or equal to total budgetary resources.

Why. An outlay above available resources indicates a scope or sign error, not an Antideficiency event in this dataset.

Authority · 31 U.S.C. 1341 context; DoD FMR Vol 3

Asserts. Obligations incurred per File B are within 0.5% of File A for the same scope, fiscal year and submission period.

Why. File A and File B are separate submissions of the same execution at different grain. A variance is a cross-system reconciliation finding to be published and explained, not a reason to withhold the data - so this control reports rather than blocks.

Authority · Treasury USSGL; OMB Circular A-136

Asserts. No File B row repeats an obligation figure already published against another program activity for the same Treasury account, object class, funding source and emergency fund code.

Why. File B's activity identifier changed from program_activity_code to the Program Activity Reporting Key in FY2026 and the account's object-class figure is repeated against every PARK rather than split across them. Added up as published it counts the same money once per key. The extract counts a replicated group once; this control reports how many rows that was and what it was worth, so the repair is published rather than assumed.

Authority · OMB Circular A-11 program activity reporting; Treasury USAspending File B specification

Asserts. No DOW-scope figure includes agency identifier code 011.

Why. File A carries five agency codes. Including the Executive Office of the President in a Department total overstates it materially.

Authority · Treasury Account Symbol structure; OMB agency codes

Asserts. File C account-linked contract obligations do not exceed award-file obligations for the same fiscal year.

Why. Linkage above 100% would mean the subset is larger than the set, indicating a scope mismatch between the two chains.

Authority · USASpending DATA Act submission model

Asserts. Every current dm_load row has a non-null vintage and extraction timestamp.

Why. A figure without a vintage cannot be defended, reproduced, or compared against a later extract.

Authority · Internal reporting standard

Asserts. Any fiscal year whose submission period is not P12 is marked as partial.

Why. An in-progress year shown beside closed years reads as collapsed execution unless it is labelled.

Authority · Internal reporting standard

Asserts. Every published definition names at least one authoritative source.

Why. A definitions registry without citations is an opinion, not a standard.

Authority · Internal reporting standard

Asserts. For each dimension, the sum of the retained top-N buckets does not exceed that fiscal year's award total.

Why. Guards the double-counting failure mode where a record increments more than one bucket.

Authority · Internal reporting standard

Asserts. For each dimension, the sum of the retained top-N buckets does not exceed that fiscal year's assistance total.

Why. Same double-counting guard as AWD-01 (contract awards), applied to the assistance dimension cuts.

Authority · Internal reporting standard

Asserts. Every row in dm_program_fy carries a non-empty program_code that exists in dm_program_dim, and hangs off a dm_load row with a vintage.

Why. A program-level figure with no program identity or no vintage cannot be cited, corrected, or reproduced. This is the same provenance rule the rest of the site enforces by shape, asserted here because the program tables are the first cut keyed to a budget line rather than an account.

Authority · Internal reporting standard

Asserts. For every fiscal year, the sum of dm_program_account.obligation for a program does not exceed that program's fiscal-year obligation.

Why. federal_accounts_funding_this_award lists every account funding an award and states no split. Summing a program's obligations by account would fabricate an apportionment the source does not contain -- the specific error this table exists to make impossible.

Authority · Internal reporting standard

Asserts. Every dm_program_fy row carries traceable_obligation and untraceable_obligation summing to its obligation, with traceable_pct consistent with them.

Why. A program obligation published without the share of it that names a funding account reads as reconciled when it is not. In FY2025 only 7.6% of F-35 obligations name an account, so the total alone is the misleading figure and the pair is the honest one.

Authority · Internal reporting standard

Asserts. Where an account outside agency codes 097/021/017/057 is named on a program's obligations, dm_program_account records it in out_of_scope_accounts rather than omitting the row.

Why. SCOPE-01 keeps agency 011 out of Department budgetary totals, and that is right for budget. On the execution side, Foreign Military Sales trust dollars (011-8242) sit on the same F-35 contract actions as Department appropriations. Dropping those rows would hide why a program's obligations are not comparable to its appropriation.

Authority · Internal reporting standard

Asserts. A program fiscal year is flagged partial exactly when it is the newest fiscal year in the current contract vintage.

Why. The newest year in the warehouse is always period-to-date. An unflagged partial year reads as a collapse in program spending rather than a year still being reported.

Authority · Internal reporting standard

Asserts. dm_program_coverage carries one row per fiscal year in which attributed and unattributed obligations sum to the fiscal-year total, and attributed_pct is consistent with them.

Why. FPDS records 'no acquisition program' as the explicit code 000 with description NONE rather than as a null, so a null test finds almost nothing and makes program tagging look complete. It is not: roughly three quarters of DoD contract dollars carry no program at all. Publishing the coverage denominator stops the program pages from implying they cover the file.

Authority · Internal reporting standard

Asserts. For every fiscal year in dm_exhibit_program_fy, the number of distinct fy_role values equals the number of President's Budget books held that cover that year (a year is covered by the books for pb = fy, fy+1 and fy+2).

Why. Each -1 book restates three fiscal years in three roles: prior-year actuals, current-year enacted, budget-year request. A fiscal year therefore appears in three successive books with three different numbers, and the difference between them is the restatement history -- the one thing these exhibits carry that exists in no other source here. Any de-duplication that leaves a year with fewer roles than the books we hold would silently destroy it, so the load is blocked rather than published.

Authority · DoD Financial Management Regulation Volume 2A/2B, exhibit formats

Asserts. For every row in dm_exhibit_line and dm_exhibit_program_fy, fy_role equals prior_actual when fiscal_year = pb_year - 2, enacted when fiscal_year = pb_year - 1, request when fiscal_year = pb_year, and other otherwise.

Why. The role is what turns a number into a claim: the same $6.9 billion is a request in one book, an enactment in the next and an actual in the third. If the role were inferred from a column heading it would drift with the heading, so it is derived arithmetically from the book year and this control asserts nothing overrode it.

Authority · Internal reporting standard

Asserts. No row of dm_exhibit_line with is_memo = false belongs to the P-1R exhibit or carries a cost type naming a memo non-add.

Why. The P-1R exhibit is National Guard and Reserve equipment already counted inside the P-1 lines, and R-1 rows outside total obligation authority are the same kind of restatement. Both are kept rather than dropped, because a total that silently omits an exhibit cannot be explained on the page; that only works if the flag is right, which is what this asserts.

Authority · DoD Financial Management Regulation Volume 2A, P-1R exhibit

Asserts. For every budget line, book year and fiscal year, dm_exhibit_program_fy.amount_k equals the sum of dm_exhibit_line.amount_k for the same key, within a thousandth of a percent.

Why. The rollup is what the page reads; the line rows are what the exhibit published. A rollup that does not foot to its own detail is a fabricated figure with a citation attached, which is worse than no figure, so the load is blocked.

Authority · Internal reporting standard

Asserts. Every row of dm_exhibit_line carries a total_basis of total_column, sole_column or sum_of_components and a non-empty total_column label.

Why. The exhibits change shape between eras -- Base and OCO, Less Supplementals, Discretionary and Reconciliation -- and the broadest figure for a year is not always in the same place. Two rules pick it, and which one fired changes what the number means, so the basis is recorded per row and asserted here rather than assumed.

Authority · Internal reporting standard

Asserts. Every non-null treasury_account in dm_exhibit_program is three digits, a hyphen and four digits, and the share of budget lines that resolved to one is reported.

Why. The exhibit account symbol (1506N) and the Treasury account symbol (017-1506) differ only by the organisation letter, and that crosswalk is the whole bridge from a budget line to the execution files. A malformed symbol would join to nothing and read as an execution gap rather than as a parsing failure, so the shape is asserted and the coverage published beside it.

Authority · Treasury Financial Manual, agency and account symbols

Asserts. Every row of dm_exhibit_weapon_link names a match_method and the evidence it rests on, and the share of the newest book's weapon systems that reached a budget line is reported.

Why. Neither the weapons book nor the -1 exhibits carry the other's key, so every link here is derived. Publishing derived links without the evidence would let a reader mistake a name-similarity match for a Department-published mapping. Systems that reached no budget line are counted rather than hidden, because that count is the honest measure of how far this crosswalk goes.

Authority · Program Acquisition Cost by Weapon System, DoD Comptroller

Asserts. For every President's Budget year whose Program Acquisition Cost by Weapon System book states an investment breakdown, the non-memo P-1 request lines total the procurement figure it publishes and the non-memo R-1 request lines total the RDT&E figure, each within half of the last printed digit.

Why. Every other control here asks whether the extract is consistent with itself. This one asks whether it is right: the weapons book totals the same request from the same submission, so a disagreement means the extract has gained or lost money, not that the sources differ. It is what caught the P-1 advance-procurement subtotal being counted alongside the by-year detail it summarises -- a $14.4 billion double count in PB2026 alone that no internal check could have seen.

Authority · Program Acquisition Cost by Weapon System, DoD Comptroller, Introduction

Asserts. Every row of dm_exhibit_program_link carries a match_method of designator or exact_name, non-empty evidence, and a budget line that exists in dm_exhibit_program.

Why. This crosswalk is the step that lets an appropriation be followed into a contract, which makes a wrong link worse than no link: it would put one program's contracts under another program's budget line. Nothing weaker than a shared type designator is accepted, an ambiguous designator produces no row, and every accepted row states what it rests on so a reader can reject one without distrusting the rest.

Authority · Internal reporting standard

Asserts. For every weapon system and fiscal year, the appropriation blocks extracted from the page sum to the total the page prints, within the greater of $0.6M and 0.2%. At least 97% of system-years must foot.

Why. The Program Acquisition Cost by Weapon System book prints a total beneath the appropriation blocks it is made of, so a page that does not foot has been read wrongly. The book changes shape between eras -- Base/OCO in PB2020-21, Discretionary/Mandatory in PB2026 -- and a column assigned to the wrong year publishes an OCO or reconciliation add as a whole year. A small number of pages are typeset so tightly that a figure cannot be assigned to a column at all; those are named rather than absorbed, which is why this is a threshold rather than an absolute.

Authority · Program Acquisition Cost by Weapon System, FY2020-FY2026 President's Budget

Asserts. For each weapon system and book year, the non-memo request on every budget line tied to that system is compared with the system total the weapons book publishes. Neither direction is repaired; both are published on the system.

Why. This is the per-system form of EXH-08 and the only check on the weapon crosswalk that reaches outside the extract. A shortfall means the crosswalk has not found some of the system's spares, modification or support lines -- their titles carry no designator to tie back -- which is why no programme total on this site is presented as complete. An excess means a budget line is broader than the programme: the Next Generation Squad Weapon ammunition line funds ammunition the weapons-book page excludes, and one R-1 program element can fund several systems. The exhibits publish no split inside a line, so nothing here apportions one. The control exists to state the gap on the page beside the figures; it must never be satisfied by dropping links until the two numbers agree, which would be fitting the crosswalk to the answer.

Authority · Program Acquisition Cost by Weapon System; P-1 and R-1 exhibits, FY2020-FY2027

Asserts. Every row in the alias table names a system that exists in the weapons book, quotes the phrase the expansion was read from, and was made by one of the two recorded methods.

Why. The alias table is what lets a search for 'JSF' find the F-35. Its risk is not arithmetic -- no figure is ever aggregated through it -- but that an abbreviation quietly starts meaning something the Department never wrote. So an alias exists only where the book expands it into the system's own name, sharing two or more significant words with it, and the sentence is carried on the row so a reader can reject one alias without distrusting the rest. The same evidence test the weapon crosswalk uses.

Authority · Program Acquisition Cost by Weapon System, FY2020-FY2026

Asserts. Every fiscal year with a published File C linkage figure carries its full submission-period series, and the published period appears in that series flagged as chosen.

Why. The published linkage figure is the product of a choice: the warehouse holds four File C snapshots per fiscal year and the extract publishes the one with the most rows, on the reasoning that it is the most complete copy held. That is a defensible rule and it is still a rule. This control is a shape rather than a check, in the sense PROG-03 is: it makes it structurally impossible for a page to render the headline without the rejected alternatives beside it, because a reader who cannot see the alternatives cannot know the headline was chosen.

Authority · DATA Act File C (award financial), USASpending account-level submissions

Asserts. For each fiscal year, the linkage percentage is computed from every substantive File C snapshot held and the spread between the highest and lowest is published. A spread wider than a factor of two is reported as a finding.

Why. This control measures the size of the choice FILEC-01 exposes. Reading FY2022 at period 3 gives 1.4% linkage and at period 12 gives 11.6% — the same warehouse, the same year, an eight-fold difference. Where the spread is that wide, comparing one year's figure with another's is comparing two arbitrary picks, and a falling or rising line drawn through them measures the picks rather than the Department. The control does not fail because the data is wrong; it fails to stop the site asserting a trend the files cannot carry. It must never be satisfied by narrowing the series until the spread closes.

Authority · DATA Act File C (award financial); FPDS-NG contract actions

Asserts. For every exhibit sheet and fiscal year, the rows counted into total obligation authority plus the rows set aside as memo equal the 'Total of Displayed Rows' figure the workbook itself states for that column, to within one thousand dollars.

Why. The memo rules on these seven exhibits are not one rule -- P-1R is a whole memo exhibit, P-1 hides memo cost types and an advance-procurement subtotal inside Add rows, R-1 and O-1 put rows outside total obligation authority, C-1 publishes a reconciliation breakout of projects already in the year sheet, and M-1's identically-flagged rows are negative reimbursable offsets the published total includes. Argued one at a time they are assertions. Checked against the footer the Department prints on each sheet they are verifiable: a rule that drops a row it should not, or keeps one it should not, stops matching immediately. Twenty-eight sheet-years across the FY2026 and FY2027 books tie exactly, and the single failure found while building this control was a real row -- $24K on account 5286A with every line-item column blank -- that a 'skip rows with no line item' guard was silently discarding.

Authority · DoD Comptroller "-1" display workbooks, 'Total of Displayed Rows' line

Asserts. For every exhibit and fiscal year, the sum of the discretionary and mandatory figures equals the sum of the totals, to within one thousand dollars.

Why. Discretionary is derived by subtracting the mandatory columns from the year's total rather than by adding the columns that look discretionary. Several exhibits publish alternative measures of one figure side by side -- C-1 prints Authorization, Authorization of Appropriation, Appropriation and Total Obligation Authority for the same project -- and adding those as components put every construction project's discretionary figure at three times its own total. This control is what stops that class of error returning: subtraction cannot produce a split that does not foot, so a failure here means the total column itself was picked wrongly.

Authority · DoD Comptroller "-1" display workbooks

Asserts. No P-1R row is counted into total obligation authority in any book or fiscal year.

Why. The P-1R exhibit is National Guard and Reserve equipment already carried inside the P-1 lines. It is retained rather than dropped so its absence from a total can be explained on the page, but a single unflagged row would add its money to procurement twice. The FY2027 book puts $6.6B in this exhibit.

Authority · DoD Comptroller P-1R exhibit, cost types marked (MEMO NON ADD)

Asserts. Every project on a C-1 Mandatory Reconciliation sheet appears in that fiscal year's C-1 sheet, and the year sheet's amount is at least the reconciliation amount.

Why. The reconciliation sheets look like additional money and are not: on the FY2027 book all 25 projects appear in the FY 2027 sheet, 19 at the identical amount and 6 as part of a larger project total. Adding the sheet would have counted $2.68B twice in FY2027 and $4.90B in FY2026. This control is the evidence for treating those sheets as a breakout, and it fails the moment a future book stops behaving that way -- which is exactly when the rule would need revisiting.

Authority · DoD Comptroller C-1 display workbook, year sheets and Mandatory Reconciliation sheets

Asserts. Every row counted into total obligation authority carries an exhibit account symbol that resolves to a Treasury agency and account.

Why. The account symbol is the only key these display tables share with the execution files, and the organisation letter is the whole of the difference between '2020A' and '021-2020'. A row that does not resolve is a row the request can be read for but never traced to what was obligated against it, so the count is published rather than left as a silent gap in every crosswalk downstream.

Authority · DoD Comptroller "-1" display workbooks; Treasury Account Symbol structure

Asserts. For every fiscal year, the sum of obligations across the account-level File B extract equals the Department obligation total already published on the execution page, to within 0.01%.

Why. A detail table that does not foot to its own total is worse than no detail table: every drill-down built on it would quietly disagree with the figure printed above it, and a reader has no way to tell which of the two is wrong. The detail and the total are produced by the same grain and replication rules, so any difference here means those rules were applied twice differently.

Authority · USASpending File B; DoD FY2025 Agency Financial Report

Asserts. For each fiscal year carrying detail, the sum of obligations across detail rows equals that year's obligation total, to within 0.01%.

Why. From the FY2026 submission File B identifies a program activity by reporting key rather than by code, and where an account holds several keys it repeats the account's object-class figure against each one instead of splitting it. At detail grain that defect is worse than in a total: a drill-down would show the same money under four different keys and a reader would have no way to know. Rows that differ only by reporting key and repeat one figure become one row, and this control is what proves the collapse neither lost nor duplicated a dollar.

Authority · USASpending File B, FY2026 P09 submission

Asserts. For each fiscal year, obligations are split by period of availability and the share whose availability could not be resolved is published; more than 1% unresolved is reported.

Why. Annual appropriations expire on 30 September and cannot be obligated afterwards; multi-year and no-year authority can. It is the single most decision-relevant attribute in the file in September and it is not published as a field -- it has to be read off the beginning and ending periods of availability. An obligation rate that mixes the two answers no question at all, so the split is measured and the unresolved remainder is stated rather than folded into either side.

Authority · 31 U.S.C. 1502; DoD FMR Volume 3

Asserts. For every fiscal year, the sum of the daily contract obligation curve equals the contract obligation total published for that year, to within 0.01%.

Why. The daily curve is the spine of every pace comparison, year-end share and September projection on the execution page. If it does not foot to the total the site already publishes, every figure drawn from it is wrong by the same amount and none of them would look wrong on their own.

Authority · USASpending contract award files (FPDS-NG)

Asserts. Every execution signal is computed against at least three of its own category's fiscal years, and every one carries the evidence it was computed from and the method that produced it.

Why. A deviation measured against one or two prior years is an anecdote with a z-score printed on it. Three is the minimum at which a median means anything, and a signal that cannot name the years behind it cannot be argued with. The deviation is also capped: a category whose prior years are nearly identical has a scale near zero, and dividing by it produced a z of 416 -- a number that says nothing except that the denominator was small. A capped value means far outside the category's own history, and the page says so rather than printing it as a measurement.

Authority · Robust statistics: median and scaled median absolute deviation

Asserts. Every fiscal year records whether it is complete, the last action date observed, and how many whole fiscal months it covers, and the completeness flag agrees with the observed day count.

Why. The live fiscal year is shorter than the years it is compared with. Reading its total beside a closed year's shows a collapse that is only the calendar, and it is the easiest mistake on the site to make silently. Every pace comparison here runs over whole months only, and this control is what guarantees the page has the information it needs to say so.

Authority · USASpending contract award files; submission period conventions

Asserts. Every fiscal year publishes a reporting frontier no later than its last action date, the count of whole months observed agrees with whether the year is complete, and fewer than 20% of the year's actions fall after the frontier.

Why. The FY2026 extract runs to 2026-08-04 and is substantially complete only to the end of April: October through April carry 280,000 to 400,000 actions a month, May carries 75,000, and June, July and August carry 180 between them. Taking the maximum action date as the extent of the file was wrong in three ways at once — the cumulative curve ran flat for three months, the year reported ten whole months observed where it had seven, and every pace comparison and September projection measured seven real months of the live year against ten of every prior year. That put each sub-agency's pace at 68–99% of its own norm when the true figures are 104–127%: not a cosmetic error but the opposite conclusion, on a page built to support a decision. The frontier is the end of the last fiscal month, counting consecutively from October, that carries at least half the median month's action count. Consecutively, because a gap in the middle of a year is a hole in the data rather than the end of it, and treating it as the end would hide the hole. The 20% bound on the excluded tail is what distinguishes cutting off the end of a file from cutting into it.

Authority · USASpending contract award files (FPDS-NG); warehouse vintage extraction dates

Asserts. The submission period held by the account files is the newest monthly period revealed on the public submission calendar. Where it is not, the gap in periods is published.

Why. This is not a check on the extract but on its age, and it is the one the rest of the execution page depends on: a page about the year being executed that silently trails the published record by a month is presenting last month as this month. Measured on 2026-09-10 the warehouse held FY2026 P09, covering June and revealed 2026-07-31, while FY2026 P10 covering July had been revealed on 2026-09-01 and was not in it -- because the warehouse snapshot was taken on 21 August. Nothing on the site could say so. The control reports rather than blocks: a stale load is still the best available answer and refusing it would leave the site with an older one.

Authority · USASpending submission periods; OMB Circular A-11 reporting windows

Asserts. For every File A and File B account in Department scope, the beginning and ending periods of availability carried by the extract equal the two years written into the Treasury account symbol, and every account without a program year is a no-year account.

Why. The program-year view splits a fiscal year's execution by the year the money was appropriated for: FY2026's File B carries FY2025 procurement still being obligated, expired FY2025 operation and maintenance taking its adjustments, and FY2026 money, and an execution rate that mixes them answers no question a program office asks. The year is taken from the published period-of-availability columns. The account symbol writes the same two years independently, so this control re-derives them from the symbol rather than reading the extract's own answer back: where the two disagree, every figure on that account is filed under the wrong program year.

Authority · Treasury Financial Manual, Volume I, Part 2, Chapter 1500 (Treasury Account Symbol components); 31 U.S.C. 1502(a) (a fixed-period appropriation is available only for obligations incurred within its period of availability)

Asserts. For every fiscal year, File A's account-level resources, obligations and unobligated balance sum to the Department Statement of Budgetary Resources to within 0.01%, File B's accounts sum to the year's File B obligations, and no File B account lacks the dimension row that carries its program year.

Why. The program-year chart rates obligations against the resources of the same accounts. If File A's account table drops accounts, every rate is measured over too little money; if a File B account has no dimension row, its obligations fall out of every program year and the parts stop adding up to the total printed above them. Both failures are silent on the page, and a drill-down that does not foot to its own total is worse than none, so this blocks the load.

Authority · OMB Circular A-11, Section 130 (SF 133 / Statement of Budgetary Resources line structure); Treasury USSGL

Asserts. For every fiscal year, direct and reimbursable File B obligations agree between the stage totals and the account rollups to within 0.01%; no account rollup lacks the split or splits to more than its own total; and for every year carrying detail, each account's direct and reimbursable figures equal its detail rows summed by their own funding source.

Why. File B marks every row direct or reimbursable, and the site now reports execution as direct by default. A reimbursable obligation is work an account performs for a customer and is paid back for; when the customer is another Department account the same work is also that customer's direct obligation, so adding the two overstates the Department's own execution. A split that does not foot is worse than no split, because every direct figure on the site would then disagree with its own total. The control recomputes the split from tables it was not written into, and a missing split fails it rather than being read as zero.

Authority · OMB Circular A-11, Section 130 (direct and reimbursable obligations reported separately on the SF 133); DATA Act File B element direct_or_reimbursable_funding_source

Asserts. For every fiscal year the reimbursable share of File B obligations and the account carrying the most of it are published; more than 0.1% of obligations with no funding source is reported.

Why. This is a measurement, not a check on the extract: it is the size of the overstatement a direct-plus-reimbursable total would carry. FY2025: $212.9B of $1,451.1B (14.7%), $152.0B of it in the Defense Working Capital Fund. A handful of rows each year carry no funding source; they stay in the total and in neither side, and their amount is stated rather than assumed away.

Authority · OMB Circular A-11, Section 20 (reimbursable obligations and offsetting collections); DoD FMR Volume 11B (working capital funds)

Asserts. Every raw sample record carries exactly one value per column of its file and the records are numbered 1 to min(5, file rows); every file a current load's figures come from -- File A, File B, File C contracts, contract and assistance transactions, and each (PB year, exhibit) workbook with a line in the load -- has a sample; and the File A and File B sample records carry the submission period the published statements are labelled with, and the contract sample is the vintage the contract figures use.

Why. The raw-data page exists so a reader can check a figure against the record it started from and see what a file does and does not carry -- that File A has no direct/reimbursable flag and File B does, for instance. A sample that shifts a value one column over, stops one file short, or comes from a different vintage than the figures would make that check say the wrong thing with the authority of raw data. The control recomputes structure from the rows themselves, coverage from the loads and exhibit lines the sample did not produce, and currency from the published statements.

Authority · DATA Act Information Model Schema (Files A-C element names); OMB M-17-04 data quality; GAO-14-704G Green Book principle 13 (use quality information)

Asserts. Every book header's year span and edition count re-count in SQL from the edition rows, and every edition's distinct section titles and time-sensitive titles re-count from the section rows.

Why. A justification book identity spans up to twenty-nine President's Budgets and each edition restates its own format: DISA's OP-5 printed 'C. Reconciliation of Increases and Decreases' from PB2012 to PB2021 and has not since. That drift exists in no single book and is the reason this corpus is held at (book_key, pb_year). A GROUP BY that drops pb_year averages twenty-nine years into a book that was never published and reads like a tidy-up, so the header is re-counted against its own detail rather than trusted.

Authority · DoD FMR Volume 2A/2B (budget formulation and presentation); the published justification books themselves

Asserts. For every skeleton row, the editions it was seen in, the editions the book has, and the recency-weighted share all re-derive in SQL from the section rows and the edition weights, to within 0.15 percentage points.

Why. The skeleton is what a drafter is shown as the shape of their book, and it is weighted towards recent editions on purpose (half-life three years) so a section printed in every book since PB2020 outranks one dropped a decade ago. The published figure came from the extract's own observation lists; this control recomputes it from the rows that were actually written, which is the independent path. A skeleton measured over the wrong set of editions, or one carrying weights from before a new book landed, differs here.

Authority · Component editorial standard; measured from the published books

Asserts. Every exemplar passage comes from one of its own book's two most recent editions, belongs to a book marked current, and names the file and page it was read from.

Why. Bodies are held for current books only: the archive is several gigabytes of text and none of it belongs in a page database. The window is the BOOK's own last two editions, not the corpus's — a book published in PB2027 whose previous edition was PB2025 is current and both of its editions are usable material, while a book last published in PB2009 contributes structure and measurement and no text at all, which the page says rather than implies. What must never happen is an untraceable paragraph: an exemplar a drafter cannot follow back to a published page is an anonymous piece of prose, which is what this feature exists not to produce.

Authority · Site rule: every figure and every quoted passage names its source and vintage

Asserts. For every fiscal year published at entity grain, the entity NFR counts sum exactly to the total the same report states.

Why. The report prints both the table and its own total, so the total is a check the source supplies against the transcription. FY2018 fails it by 4 and is therefore withheld at entity grain.

Authority · DoD OIG annual audit-result reports

Asserts. Where a year states the split, new NFRs plus prior-year NFRs reissued equals the total issued.

Why. The two figures are printed in different places in each report. Requiring them to reconcile catches a year read from the wrong sentence.

Authority · DoD OIG annual audit-result reports

Asserts. For every year a roster is published, the number of roster rows equals the Agency-Wide material weakness count stated in the report.

Why. A roster short by one is how a renamed weakness turns into a closure. FY2023 states 28 and prints no roster, so it publishes a count and no rows.

Authority · DoD OIG annual audit-result reports

Asserts. Each material weakness object carries all ten elements, and every element carries a citation.

Why. An element missing from an object renders as a silent gap. An uncited element is an assertion, and nothing on this site reaches a page without its source.

Asserts. Recomputing each object from the roster rows gives the same last roster year, the same number of roster years present, and the same open or not-open state.

Why. The outcome is the one element the extract computes rather than transcribes. A control that read the computed value back would restate it; this one derives it again from dm_nfr_mw and compares.

Asserts. Every fiscal year in the calendar has either a full-year enactment or a full-year continuing resolution; no enactment is dated before its own fiscal year begins; every span ends on or after it starts; and no lapse in appropriations overlaps a day the full-year act was already in force.

Why. Every timing statement on /execution/timeline is read against these dates, so a wrong date does not produce a wrong date -- it produces a wrong finding about execution. The dates are curated and cited, which is exactly the kind of input that needs a machine check rather than a reviewer's eye. Day counts are derived in SQL from these dates and never stored, so the arithmetic cannot drift from them.

Authority · Public laws cited on each row

Asserts. For each fiscal year, the observed months of the Department-level timeline sum to the contract obligation the timing extract measured to the same reporting frontier.

Why. The two extracts read the same contract file through different code -- the timing step aggregates by day in pyarrow, the timeline step aggregates by month in Python -- so this compares two independent implementations rather than restating one. A control that read either extract's own answer back would be a restatement with a pass printed on it.

Asserts. Each fund holder's yearly contract obligation equals the sum of its own observed monthly rows, and the fund holders carried never exceed the Department total for the year.

Why. The fund holder is the only dimension on this site that names WHS, MDA and SOCOM, and it is carried for the twenty largest holders a year. A holder total that does not equal its own months means the ranking and the curve disagree, and the page draws both.

Asserts. Recomputing the index in SQL -- the average observed month before the full-year act, divided by the year's own average observed month -- reproduces the published value for every complete year.

Why. This index is the finding the page rests on, so it gets an implementation that does not share a line of code with the one that produced it. It is also the figure most easily got wrong: a raw share of the year is not comparable between years, because the years differ both in how long the continuing resolution ran and in how many months are observed at all.

Asserts. Each modelled group's twelve months sum to its stated annual obligation, and that annual obligation equals the current-year obligation File A reports for the same year, appropriation and agency.

Why. The model's only claim is that it redistributes a measured annual total across the months. The second half of this assertion is the one that matters: it ties the model back to the source it says it is distributing, so a model that has quietly drifted onto a different population fails rather than merely failing to add up.

Asserts. No modelled row exists for a fiscal year whose contract file is observed for fewer than twelve months.

Why. File B's figure for a year in progress is period-to-date at one submission and the contract shape stops at the reporting frontier. Spreading the first across the second produces a monthly figure that belongs to neither period, and it would look exactly like a measurement. The year in progress gets the observed layers and no model.

Asserts. Each fiscal year carries all five coverage measures, no numerator exceeds its denominator, and the current-year sample coverage is published for every year the sample is drawn on.

Why. The dated cut by colour of money and programme year is a sample whose size moves from 8.5% to 23.0% across these years. Published without its coverage it reads as a measurement of the Department, which it is not. This control is what stops the coverage silently going missing from a row the page then draws anyway.

Asserts. For every fiscal year and component, the obligations on the chain's analytic units equal File A's obligations on accounts whose programme year is that fiscal year.

Why. Every lag, ramp and recommendation on the page is read against these units. If the unit table has drifted onto a different population -- all programme years instead of the year's own money is the easy way to do it -- then every figure downstream is describing something other than what it is labelled.

Authority · OMB Circular A-136 (Statement of Budgetary Resources)

Asserts. Every unit's authority steps begin on the first day of the fiscal year, carry no authority through a lapse in appropriations, and where a full-year act exists end on File A's own budget authority appropriated for that unit.

Why. The authority curve is derived, not published, so both of its endpoints have to be pinned to File A figures and the rule only allowed to shape what runs between them. A curve carrying money through a lapse would assert the Department had authority it did not have.

Authority · 31 U.S.C. 1341; public laws cited on the calendar

Asserts. The assumed steps of each chain sum to that unit's own measured residual, the measured endpoints carry no day count, and a unit whose execution preceded its authority is assigned no assumed days at all.

Why. The interior of the distribution chain is published in no file, so the only discipline available is that the assumption divides a measured quantity and never invents one. Where execution ran ahead of the authority -- on the continuing resolution, or on carried-in balances -- there is no wait to divide, and filling the steps in anyway would manufacture a delay that did not happen.

Authority · DoD FMR Volume 3

Asserts. Every timing cell has d10 <= d50 <= d90, no quantile falls past the last day that cell was observed for, the same-point comparison window is no longer than any year's own observation, and every row is labelled as an upper bound.

Why. The ramp is measured on the fraction of contract dollars that name their Treasury account, so it is censored by construction and the label is not decoration. The window check is the one that matters: a live year's median is taken over fewer days, and comparing it against a closed year's full-year median reads as the live year running early when all it is doing is stopping sooner.

Asserts. Expanding each published slope's stored observations and taking the median of every pairwise slope in SQL reproduces the published days-per-30-continuing-resolution-days.

Why. This slope is the figure a senior decision rests on -- what another month of continuing resolution does to execution -- so it gets an implementation that shares no code with the one that produced it. It is also the figure most easily got wrong: with five or six observations a least-squares slope is set by whichever year is furthest out, and two of these six years are unusual ones.

Asserts. Each component total equals the sum of its own colours of money, and the Department total equals the sum of its components.

Why. The page now answers three different questions -- how is this account doing, how is this component doing, how is the Department doing -- from one set of rows. Every measure involved is additive across File A accounts, so a rollup that does not foot is a rollup built from the wrong population, and a reader comparing a component figure against the Department figure would be comparing two different things.

Asserts. Each chain's published minimum, likely and maximum equal the sums of its applicable steps; every step has min <= likely <= max; and the two measured endpoints carry no duration.

Why. The model's whole claim is that it is assembled from the documented process rather than fitted to the observation it is compared against. A total that does not equal its own steps would mean it had been adjusted, which is precisely the failure the earlier flat-share version of this page had. Keeping the measured endpoints at zero duration is what stops a measurement being absorbed into an estimate.

Authority · 31 U.S.C. 1513(b)(1); DoD FMR Volume 3, Chapter 2

Asserts. Each archetype's stored shape runs from at or above zero to one by the twelfth fiscal month, and every shape carries the comparison window it was cut to.

Why. The archetypes are discovered by clustering, so the only way to know the clustering saw what it was meant to see is to check the inputs. A shape that does not reach one is a truncated year that was not normalised, and clustering those together with complete years would sort by how long the year was observed rather than by when its money moved.

Asserts. Recomputing each baseline in SQL as the median of that unit's own earlier values reproduces the published baseline.

Why. A deviation is only meaningful against the right baseline, and the commonest way to get this wrong is to compare a category against a Department-wide average when these categories differ by three orders of magnitude. This recomputes the baseline from the values themselves rather than reading back the column the extract wrote.

Asserts. Every gate names a citation and an authority, a measured scope appears only where a Treasury account carries the gate, and a gate with no measured scope says why.

Why. These are quotations from public law and from congressional statements, and a page that characterises what Congress required had better be able to show where. The scope rule matters just as much: a gate whose money cannot be isolated in File A gets no number, because an estimated scope on a legislative requirement reads as a finding about Congress rather than about this site's data.

Authority · Annual Department of Defense Appropriations Act and its joint explanatory statement

Asserts. Every figure the paper's prose refers to resolves to an evidence row of this load, each finding declares exactly the figures its own sentences use, and every recommendation names a finding that exists.

Why. The paper argues from numbers, and an argument whose numbers have gone stale is worse than no argument, because it reads exactly like a current one. The prose is therefore stored with named placeholders rather than numbers and resolved against this load's own evidence at render time -- so the failure mode is not a wrong figure but an unresolvable one, and this control blocks the load rather than letting the page print a brace. The declared-set check exists because the page and this control both read the placeholders column, and a stale column would make the join agree with itself while disagreeing with the sentence. The recommendation check exists because a recommendation separated from the finding that justified it is an opinion wearing evidence's clothes.

Authority · This site's own publication standard