Catch the SF-425 error before it blocks your drawdowns
Every number on the Federal Financial Report is derived from three sources maintained by different people — the general ledger, the payment system's drawdown history and the award terms — and the form has to cross-foot. FFRProof re-derives the four line identities, reconciles 10a to the drawdown record, tests the match, checks the budget variance against the award's own prior-approval threshold, and catches the period-vs-cumulative and double-counted dollars — each finding with its form line and source page. A field the award terms do not state blocks the check that depends on it — never a guess.
The schematic's figures are the worked sample; the report computes yours from the worksheet, drawdown statement, award terms and GL export you supply. Nothing is filed or certified on the recipient's behalf.
The declared composite below is the worked sample: a $100,000 award whose form cross-foots cleanly, but 10a drifts from the drawdown statement by $0.04 and the match is $5,000 short. Press Run the pre-flight to re-derive everything in your browser — no model, no upload. Nothing is filed or certified for the recipient; the output is the audited worksheet the human signs.
Watch the launch video
24s · nothing estimated · no signupFFRProof — the SF-425 pre-flight that catches the four cents Watch on YouTube
Five checks, one deterministic chain
- Four line identities10c = 10a − 10b, 10g = 10e + 10f, 10h = 10d − 10g, 10k = 10i − 10j — each cross-footed with zero tolerance. A break returns both sides of the equation and the difference, never a bare "invalid".
- Cash reconciliationLine 10a/10b against the payment system's per-subaccount receipts and disbursements, drift reported by subaccount and to the cent — the named FFR error of not reconciling the drawdowns to the report.
- Match & final rulesRequired recipient share (10i) vs provided (10j) vs remaining (10k). A final report with 10k > 0 is a hard finding, and 10f/10h must be zero on a final.
- Budget & periodActual vs approved per cost category against the prior-approval threshold the award's own terms state (unstated when they do not), plus the period-vs-cumulative double-count check.
The award terms are parsed per filing, never stored as a shared rule table: the form's arithmetic is fixed and published, but every agency-specific field, threshold and match percentage is read from your award's own terms. An award term the reader cannot resolve returns unstated and blocks that finding rather than defaulting — a stale Uniform Guidance can never produce a wrong number, at worst "the award terms do not state this".
Agent Surface & WebMCP Integration
How an agent invokes the federal financial report pre-flight. The extraction layer reads declared fields; the audit itself is this deterministic engine.
// 1. In any browser with WebMCP support (auto-registered on page load):const audit = await navigator.modelContext.executeTool("audit_federal_financial_report", { declared: compositeDeclaredJsonOrText}); console.log("Verdict:", audit.report.verdict);console.log("Findings:", audit.report.findings.length);console.log("Cross-foot:", audit.report.crossFoot.lines);JSON Schema Parameters Contract
declaredRequiredThe full composite as declared fields: report + awardTerms + drawdown + glCsv + budgetCsv (money in integer cents, percentages in basis points). The GL export is parsed deterministically from CSV, never by a model.
Pricing
Use it free in the browser. Pay only for exports, the suite, or agent calls.
Free
$9.00 one-off
$29.00/month
| Agent tool | Metered rate |
|---|---|
audit_federal_financial_report | $1.50 per successful call |
Agent calls are metered per successful call and billed monthly to your Stripe customer. Billing shows your usage, history and spend cap.
Manage Billing & API Keys (/billing) →Agent surface: audit_federal_financial_report is registered as a WebMCP tool and advertised in /.well-known/mcp.json at $1.50 per successful call. The extraction layer it needs must be configured in the deploy environment first — until then the tool answers with an explicit unconfigured error naming the missing variable, never an estimated worksheet.