The Scorecard — Seven Observable Milestones That Will Tell Us Which Trajectory USDA Is On

The most expensive moment in a federal IT contract is not the day it’s signed. It’s roughly eighteen months in — when the agency’s workflows have been rebuilt around the vendor’s platform, switching costs have hardened into real budget line items, and oversight bodies are still drafting their first information requests. That’s when leverage shifts, and it shifts permanently.

USDA is walking into that window now, not at the end of the contract.

The Clock Is Running on the Palantir BPA

The oversight window that matters is pre-audit, and USDA is inside it — with $129.4M already committed against a $300M ceiling before the fiscal year has even closed.

According to federal spending data from USASpending.gov, USDA has obligated $129.4M to Palantir in a partial FY2026 — roughly 43% of the $300M Blanket Purchase Agreement ceiling announced on April 22, 2026. That is not a warm-up pace. It is a spend curve on track to exhaust the ceiling well before the standard audit cycle would notice.

Fiscal Year	USDA Total Obligations ($)	Transaction Count	Palantir Obligations ($)	Palantir Share of Total (%)
2,020	1.5B	6,480	~0 (near-zero per public record)	<0.1%
2,021	1.6B	4,970	~0 (near-zero per public record)	<0.1%
2,022	1.9B	4,982	~0 (near-zero per public record)	<0.1%
2,023	2B	4,984	~0 (near-zero per public record)	<0.1%
2,024	2.1B	5,213	~0 (near-zero per public record)	<0.1%
2,025	1.7B	4,166	Not separately broken out in digest	N/A
2026 (partial)	1.1B	1,706	129.4M	~11.3%

Palantir at roughly 11.3% of every USDA dollar obligated in a partial fiscal year is not a vendor relationship. It’s a structural reweighting of how the department buys technology, compressed into a single year — after five years of near-zero footprint. Add the shift in per-transaction size — the FY2026 average obligation per USDA transaction has climbed to about $668K, more than 1.5x FY2025’s roughly $406K average — and the picture becomes uglier: bigger checks, fewer of them, harder to inspect at the line-item level.

The procurement record is already public. Here’s how to read it. Below is a seven-point scorecard drawn entirely from public procurement records — the kind of framework a congressional staffer or Inspector General analyst can run without waiting for a formal review to open.

What the Vendor Roster Tells Us Before Any Milestone Is Checked

The rest of USDA’s FY2026 IT roster reveals which vendors are auditable at the transaction level and which are hidden behind reseller opacity.

Recipient	FY2026 Obligations ($)	Transaction Count	Avg. per Transaction ($)
PALANTIR TECHNOLOGIES INC.	129.4M	9	14.4M
CARAHSOFT TECHNOLOGY CORP	91.9M	30	3.1M
SYNERGY BUSINESS INNOVATION & SOLUTIONS INC.	49M	6	8.2M
LUMEN TECHNOLOGIES GOVERNMENT SOLUTIONS, INC.	45.7M	37	1.2M
DYNAMO TECHNOLOGIES, LLC	35.4M	19	1.9M
GUIDEHOUSE DIGITAL LLC	27.6M	3	9.2M
ACCENTURE FEDERAL SERVICES LLC	26.9M	41	656.3K
MICROSOFT CORPORATION	26M	5	5.2M
COLOSSAL CONTRACTING LLC	23.9M	57	419.9K
KONIAG TECHNOLOGY SOLUTIONS INC	23M	9	2.6M

Palantir’s nine orders average $14.4M — but the average is not the story; the concentration is. If a single task order accounts for the majority of that $129.4M, USDA isn’t buying a platform in pieces. It bought it once and is papering the rest. Compare either scenario to Accenture’s $656K per transaction or Lumen’s $1.2M: Accenture is buying labor hours, Lumen is buying circuits, and both are auditable in small pieces. Palantir is being paid in enterprise-scale chunks — the kind of award where the deliverable lives inside a platform license, not a line item. That is not disqualifying. It’s a warning about what kind of evidence you’ll need to see delivery.

Then there’s Carahsoft — $91.9M in USDA FY2026 obligations as prime recipient, and a well-known Palantir reseller. If the majority of Carahsoft’s $91.9M is Palantir Foundry, USDA’s real ceiling penetration isn’t 43% — it’s north of 70%, and that figure is currently unverifiable from the outside. That’s not a transparency gap. That’s a blind spot at the scale of a mid-size federal program. A second signal: USDA’s FSRS filings show zero reported Palantir subawards. That may be a reporting lag — or it may mean the subcontracting commitments aren’t being made. At BPA scale, that distinction matters, and the clock for distinguishing them is already running. As we argued in the vendor-lock analysis, the reseller channel is where the real dollar map goes dark.

The Seven Milestones Any Analyst Can Track

Each milestone below maps to a specific public data source, a specific failure mode, and a specific signal that separates a healthy trajectory from a repeat of USDA’s own MIDAS.

1. Task-order concentration ratio. Does the BPA spend fan out across distinct project areas, or does it collapse into one or two license-equivalent awards? Data source: USASpending task-order records. The failure mode is well-established: vendors who capture budget through license consolidation rather than discrete deliverables have no incentive to close out milestones, because completion ends the revenue stream. Watch for a distribution where a single order accounts for the majority of obligated dollars — that’s the pattern that precedes stalled implementations.
2. FPAC “One Farmer, One File” adoption metrics. Deployment announcements are not utilization data. “Live in field offices” and “used daily by field staff” are two different claims with two very different evidentiary bars. The structural failure pattern is familiar: press releases announced deployment; county-office staff had never seen the system. Look for county-level adoption rates, transaction volumes flowing through Foundry, and, critically, whether legacy FSA systems are actually being decommissioned or merely running in parallel. The best-case scenario we sketched in part 1 depends entirely on this milestone.
3. Carahsoft flow-through transparency. The positive signal is task-order-level disclosure that lets outside analysts attribute reseller dollars to specific Palantir products and use cases. The negative signal is continued opacity at the prime-recipient level, which makes the true Palantir footprint at USDA impossible to verify. Given how large the Carahsoft position already is, an oversight-friendly agency should publish the mapping voluntarily. A refusal to do so is itself a data point.
4. FSRS subaward and subcontracting compliance. Federal Subaward Reporting System filings should show flow-down to small businesses and specialty integrators within roughly two quarters of major task-order awards. The zero-subaward reading noted above is a monitoring flag, not a finding — but if it doesn’t populate by mid-FY2027, it graduates from lag to gap. Small-business subcontracting commitments in a BPA of this size are not optional; verifying they’re being honored is squarely inside the oversight remit.
5. IG and GAO engagement timing. Early engagement is the system working. Late engagement — after switching costs have hardened — is the failure mode. Auditors who arrive after the money is committed can diagnose the failure. They cannot recover the funds. The VA EHR modernization is the current-day version of the same problem: as of December 2024, VA had deployed the new EHR system to just six locations with more than 160 sites remaining — and GAO keeps flagging it in the high-risk series. USDA IG staff and GAO’s IT acquisitions team should be issuing information requests now, at the 43%-obligated mark, not after the ceiling is exhausted.
6. Recompete posture at ceiling or term. Watch what USDA does when the BPA approaches its cap. A competitive solicitation with open-data portability standards baked into the requirements is a healthy signal. A sole-source extension citing “transition risk” or “data integration complexity” is the standard procurement language of vendor lock-in. Those phrases are distinguishable from legitimate technical constraints only by one test: whether data portability was designed into the original architecture. If it wasn’t, the “risk” is the lock, not a defense against it.
7. Spend-curve inflection versus published deliverable log. An obligation pace that moderates in FY2027 as implementation phases complete is a signal of genuine delivery. Continued acceleration toward the ceiling — without a public, dated milestone log tied to obligations — is the pattern that has preceded every failed federal platform consolidation in the modern record. This is where the scorecard becomes predictive rather than diagnostic.

Why the Scorecard Is Composite, Not Sequential

No single milestone is dispositive. Task-order consolidation on its own could reflect legitimate architecture decisions. Zero subawards on its own could be a reporting artifact. What forces the accountability conversation is a cluster — two or three milestones flashing red together, on the same quarterly review. That’s the pattern signal, and it’s the one that oversight bodies should be scanning for.

The Precedent That Makes This Urgency Non-Negotiable

USDA has failed at this exact type of consolidation before, and the federal record shows the failure mode has a repeatable structure.

Start with the closest analog: MIDAS — Modernize and Innovate the Delivery of Agricultural Systems — was USDA’s own attempt to consolidate Farm Service Agency programs onto a single platform. By March 2015, MIDAS had overrun its baseline cost estimate by $93 million, and USDA’s Farm Service Agency spent $423 million on the program before Secretary Vilsack halted it. Same agency. Same mission area. Same mechanism: obligation pace outran delivery evidence, and by the time GAO’s postmortem on MIDAS landed, the money was gone.

Widen the aperture and the pattern gets worse. The US Secretary of Defense cancelled the Air Force’s Expeditionary Combat Support System in 2012 after spending over $1 billion without fielding any usable capability. The VA’s EHR modernization remains on GAO’s high-risk list; as of December 2024, VA had deployed the new EHR system to just six locations with over 160 remaining, with four more planned for 2026. FBI Virtual Case File. DoD DIMHRS. Healthcare.gov’s launch. The catalog is long enough that the pattern is not vibes — it’s structure.

The procurement structure itself — enterprise licenses, a reseller channel, an aggressive obligation curve, no published milestones — matches the preconditions of every large federal IT failure in the modern record. The evidence supports that claim. The scorecard is how outside observers distinguish between trajectories before the money settles the question. As the worst-case analysis in part 2 laid out, remediation costs after month eighteen are almost always higher than prevention costs before it.

How to Actually Run This Scorecard

Every milestone above resolves against data that is already public — no FOIA required.

The stack is straightforward: USASpending.gov for obligation records and task-order flow, FSRS for subaward filings, agency procurement portals for solicitation activity, IG semiannual reports, and GAO’s correspondence and high-risk updates. What the raw data doesn’t do easily is track velocity and detect the cluster signals that matter — that’s where a purpose-built monitoring layer like a federal contract watch service earns its keep, or where analysts can build their own using structured award data feeds.

Cadence matters. Given the obligation pace, quarterly review is the right frequency; annual reviews will miss the inflection points that decide whether this contract lands on the success side of the federal IT record or the other one. And read the scorecard as a composite — the story lives in the pattern, not in any single milestone.

How can I track Palantir’s USDA task orders on USASpending.gov?

Filter USASpending.gov by recipient name (Palantir Technologies Inc.), awarding agency (Department of Agriculture), and fiscal year, then drill into individual award records for parent-child task-order relationships. Cross-reference with Carahsoft prime-recipient records to catch reseller-channel obligations that don’t show up under Palantir’s name.

What are the early warning signs of federal vendor lock-in?

Three signals cluster before every lock-in outcome: rapid early obligation pace against a ceiling, task-order consolidation into a small number of large license-equivalent awards, and the absence of open data portability standards in the underlying architecture. If a recompete is later justified by “transition risk,” the lock was designed in from the start.

When should the Inspector General review a federal IT contract?

Before switching costs harden, not after. In practice, that means engagement should begin when obligations cross roughly one-third of the ceiling — which USDA passed for the Palantir BPA in a partial FY2026. Waiting for the formal audit trigger is exactly the pattern that produced MIDAS, ECSS, and the VA EHR problem.

What lessons does USDA MIDAS offer for the current Palantir contract?

MIDAS proved that the same agency can obligate hundreds of millions against a modernization vision while field-level delivery stalls, and that GAO’s diagnosis arrives too late to recover the money. The specific lesson is not “big IT fails” — it’s that adoption evidence at the county-office level, not deployment announcements, is the metric that separates real modernization from expensive theater.

The Bet Worth Making

Here’s the position: this contract will not be decided by whether Palantir Foundry is good software. It will be decided by whether USDA leadership publishes deliverable milestones tied to obligations, whether Carahsoft’s flow-through gets disclosed, and whether GAO and the USDA IG open information requests in the next two quarters rather than the next two years.

The evidence to run that scorecard is already sitting in public databases. What has been missing is a structured way to read it. That’s what this framework is for — and if the seven milestones start flashing red in combination, the appropriate response is not another op-ed. It’s a formal review, opened while the leverage still exists to change the trajectory. After month eighteen, the question stops being “is this working?” and becomes “how expensive is it to stop?” USDA answered that question once before, at $423 million. The scorecard exists so someone doesn’t have to answer it again.

Frequently Asked Questions

What is the USDA Palantir BPA and how large is it?

It is a Blanket Purchase Agreement with a $300M ceiling announced April 22, 2026. USDA has already obligated $129.4M — roughly 43% — within a single partial fiscal year, a pace that could exhaust the ceiling before a standard audit cycle responds.

Why does Carahsoft's $91.9M USDA position create a transparency problem?

Carahsoft is a major Palantir reseller, so if most of its obligations fund Palantir Foundry, actual BPA ceiling penetration could exceed 70% — a figure that is currently unverifiable from public spending records alone.

What was MIDAS, and why is it directly relevant to this contract?

MIDAS was a USDA farm-records system scrapped after roughly $423M in spending. It illustrates the failure mode the scorecard is designed to detect: large enterprise IT bets without measurable adoption milestones or deliverable logs.

When should IG and GAO staff engage with a contract at this scale?

The article argues oversight bodies should issue information requests now, at the 43%-obligated mark — not after the ceiling is exhausted, when switching costs have hardened and the leverage to correct course is effectively gone.

How can an analyst tell the difference between a healthy recompete and vendor lock-in?

A competitive solicitation with open-data portability requirements built into the specifications is the healthy signal. A sole-source extension citing 'transition risk' — without portability designed into the original architecture — is the lock-in signal.