DataVisuals™ The decision governance company Request a demo

01 / The scale of the graph

The graph is bigger than the records.

A federal dollar crosses five tiers and a dozen disjoint data surfaces before it reaches a person. Each surface holds a fragment. None holds the path. That gap — between the money that moved and the records that describe it — is where improper payments live.

Zoom out — one award to the whole field • widening
10⁰One award · one decision, one path down1 case
10²One program · thousands of awards, each branching~64 measured
10⁵One agency · awards × subawards × periods15 agencies
10⁷The federal field · every grant, every entitlement, every tier$3T since FY03
— · —Below first-tier subaward · the public trail endsno path on record

02 / The denominator nobody can size

The measured part is the small part.

The improper-payment figure everyone cites covers a fraction of federal programs. The rest of the graph is not clean — it is simply unmeasured. Scale isn’t the headline number. It’s the part below the line that no surface reports.

$186B

Estimated improper payments, FY2025 — up $24B over the prior year.

GAO-26-108694 · Apr 2026

64

Programs that produced that estimate, across 15 agencies — a subset of the whole.

GAO FY2025 reporting

~$3T

Cumulative estimate since FY2003. GAO notes the real figure may be higher.

GAO · since FY2003

Tier 1

Depth the public award trail reaches: prime award plus first-tier subaward. Below that — dark.

USAspending / Treasury

03 / What the graph actually is

Five tiers. Two property sets. Nine ways it breaks.

Model the money as a graph: party nodes connected by FLOW edges. Every edge carries a ledger entry — always. A decision wrapper attaches only when one is promoted. The whole structure is a projection of an append-only ledger, rebuildable from records.

Exhibit · Fund-flow graph model with signals tokens only — no PII
Fund-flow graph model with signals A vertical chain of party tiers connected by FLOW edges, with edge and node property callouts, the ledger-projection note, and a band of nine flow-graph signals that produce alerts. Agencyfunding source State / passthroughallocates downstream Recipientprime award holder Subrecipientsub-award holder Beneficiarypayment endpoint FLOW edge — money movement archetype · which decision produced it amount · flow_date · period ledger_entry_id — always present decision_id — null unless promoted Party node tier — role dimension, not a label identity_cluster_id · conflict = signal exclusion flags · tokens only, no PII Graph = projection of the append-only ledger · rebuildable from records Flow-graph signals → alerts 1 · Conservation breachout exceeds in 2 · Cyclemoney round-trips 3 · Skip-tiera tier is bypassed 4 · Orphan flowno path to source 5 · Identity collisionsame id, two tiers 6 · Excluded partyflagged in path 7 · Temporal inversionpaid before funded 8 · Fan-in convergencemany feed one 9 · Duplicate flowrepeated payment

The edge is the unit of truth

Each FLOW carries an amount, a date, a period, and a ledger_entry_id that is always present. The decision_id is null unless that flow was promoted to a governed decision. Recording is universal; promotion is conditional. That separation is what makes the trail reconstructable instead of selectively logged.

Identity conflict is a signal, not a merge

A party node carries a tier, an identity_cluster_id, exclusion flags — and tokens, never raw PII. When the same identity resolves across two tiers, the instinct to deduplicate is the wrong one. The collision is the finding. The graph keeps the conflict and raises it.

04 / Why scale is the problem

The data that should track the dollar is scattered across surfaces that don’t reconcile.

No single system holds the path from agency to beneficiary. It’s assembled from federal surfaces built for other jobs — each partial, each with its own gaps. Stitching them is where the trail breaks, and where the breaks hide.

USAspending.gov
Prime + first-tier subaward only. Below tier one, no path is recorded. Subaward records duplicate and sometimes exceed the prime amount.
SAM.gov
Registrations and exclusions — but a registration is an entity snapshot, not a flow. It can’t tell you where the money went.
Treasury Do Not Pay
A screen at the moment of payment. No durable link back to the award decision that authorized it.
HHS LEIE · SSA DMF
Exclusion and death matches as point-in-time lookups. Timing drifts — a clean check today says nothing about the flow last quarter.
Across all of them
Reporting delays, unlinked awards, double-counting between prime and sub. The graph is real; the record of it is full of holes.

05 / What a whole graph surfaces

Nine signals you can only see end to end.

These are graph-shape anomalies, not record-level flags. None of them are visible from one surface or one tier — they appear only when the full path is present and the records reconcile. That’s the payoff of holding the whole thing.

01

Conservation breach

Out exceeds in.

02

Cycle

Money round-trips.

03

Skip-tier

A tier is bypassed.

04

Orphan flow

No path to source.

05

Identity collision

Same id, two tiers.

06

Excluded party

Flagged in path.

07

Temporal inversion

Paid before funded.

08

Fan-in convergence

Many feed one.

09

Duplicate flow

Repeated payment.

06 / The resolution

Scale stops being the problem when the graph is a projection, not a pile.

Decision Pipelines™ holds one append-only ledger. The fund-flow graph is rendered from it, so every flow carries its own entry and the whole path rebuilds from records — at any tier, for OIG, GAO, or a court. The dollar doesn’t outrun its own trail.

Accountability follows the dollar. Every decision. Every level. On the record.

Request a briefing See FraudGuard™