The interpretation of the research and development (R&D) tax credit under Section 41 of the Internal Revenue Code has changed through legislation, Treasury regulations, and judicial decisions. Eustace v. Commissioner, 312 F.3d 905 (7th Cir. 2002), affirmed T.C. Memo. 2001-66 and illustrates the restrictive standards applied to software research at that time. The Tax Court also found the taxpayer’s reconstruction of qualified expenses inadequate. The case did not categorically prohibit retrospective estimates, impose a universal time-tracking system, or permanently change the burden of proof. Its historical discovery standard must be distinguished from the later regulations. Recent changes to Form 6765 make careful identification of business components and substantiation of expenses especially useful, but their requirements arise from the applicable filing instructions rather than from Eustace itself.

The Statutory Architecture and Historical Context of Section 41

Congress introduced the research credit in 1981 as a temporary incentive for business research. Section 41 now requires an activity-based analysis, commonly described as a four-part test. First, the research expenditures must satisfy the applicable research-or-experimental expenditure requirement. This was historically described as the Section 174 test; for amounts paid or incurred in tax years beginning after December 31, 2024, Section 41(d)(1)(A) refers to domestic research or experimental expenditures under Section 174A. Second, the research must seek technological information. Third, that information must be intended for use in developing or improving a business component. Fourth, substantially all of the research activities must constitute elements of a process of experimentation for a qualified purpose: improved function, performance, reliability, or quality. The statutory exclusions must also be considered. Technical uncertainty may concern capability, method, or appropriate design; ordinary business uncertainty alone is insufficient.

The following table distinguishes the statutory requirements, their present application, and the historical issues illustrated by Eustace:

The Facts of Applied Systems and the Eustace Petition

The petitioners in Eustace v. Commissioner were the shareholders of Applied Systems, a Subchapter S corporation that developed and sold sophisticated software used by independent insurance agencies to manage their multifaceted business operations. During the early 1990s, Applied Systems embarked on a massive effort to enhance its existing software packages. These enhancements included the development of additional ratings computations to handle complex transaction modules, the implementation of concurrency management to allow multiple users to access files simultaneously without data corruption, and the creation of a memory-efficient proprietary text editor to replace a third-party word processing module.

Applied Systems did not claim research credits on its original returns for 1990 through 1992. It later hired a new tax manager who prepared claims on amended returns. According to the IRS Research Credit Claims Audit Techniques Guide’s discussion of the Tax Court proceedings, Applied Systems had five departments and approximately 450 employees. The tax manager interviewed some employees, not all 450, and prepared a salary worksheet covering 227 employees he considered eligible. Six employees testified at trial. The IRS disputed both qualification and substantiation, and the Tax Court denied the credits. The Seventh Circuit affirmed on the technological-information and experimentation issues.

The Judicial Doctrine of the “Discovery Test”

A central issue in Eustace was the meaning of discovering technological information under Section 41(d)(1)(B)(i). Applying Seventh Circuit precedent, including United Stationers, Inc. v. United States, the Tax Court required information beyond the existing state of knowledge in computer science. The court found that Applied Systems’ software improvements did not meet that historical discovery standard. This was the interpretation applied in the litigation, rather than a description of today’s regulatory test or an uncontested national standard throughout the early 1990s.

The Seventh Circuit affirmed, observing that Applied Systems had not shown a technological advance of the kind its precedent required or uncertainty about the technological ability to produce software of that kind. The opinion also acknowledged a different formulation in the Tenth Circuit’s Tax & Accounting Software decision and explained that future final regulations would affect the analysis. The following table describes the software features used by the appellate court to illustrate the overall character of the work. These are summaries of its historical reasoning, not separate modern-law exclusions for each feature.

The Seventh Circuit and the Scientific Method of Experimentation

Judge Easterbrook’s opinion also discussed the process-of-experimentation requirement. Under the court’s then-existing precedent, experimentation required formulating and testing hypotheses to resolve uncertainty about technological feasibility. The opinion contrasted Galileo’s inclined-plane experiments with selecting paint nozzles and adjusting software for faster execution or lower memory use. Those comparisons explain the historical reasoning; they should not be treated as categorical exclusions under current regulations. Systematic trial and error can qualify today when it evaluates alternatives to resolve the relevant technological uncertainty and the other Section 41 requirements are met.

The enduring distinction is between resolving an uncertainty through a qualifying evaluative process and merely carrying out development work. Eustace’s parking-space analogy illustrates that not every uncertain task is an experiment. Under Treasury Regulation Section 1.41-4(a)(5), however, uncertainty about appropriate design can qualify even when capability or method is already known. Software iterations and engineering design cycles therefore require factual analysis of the uncertainties, alternatives, and evaluation performed. Work directed only to appearance, customer taste, or other excluded purposes does not qualify merely because it involves repeated revisions.

Documentation Failures and Unsupported Retrospective Reconstructions

The Tax Court’s substantiation findings in Eustace are discussed in the IRS’s May 2008 Research Credit Claims Audit Techniques Guide. The guide presents the case as an example of deficiencies that can occur in retrospective research-credit studies. The court rejected the particular expense reconstruction because its salary worksheet and testimony did not reliably establish qualified expenditures. The guide’s characterization of a typical prepackaged study should not be attributed as a quotation from the court. The guide is an examination resource, not binding law, and its historical discussion does not establish that every after-the-fact study is invalid.

The Requirement for Project-Level Nexus

As the IRS guide explains, the evidence failed to connect the claimed salaries to qualified activities at the relevant subcomponent level. A salary list and six employees’ testimony did not establish the amounts paid for qualifying research. The practical requirement is a supportable connection between the expenses claimed and the activities that satisfy Section 41. The regulation requires records sufficiently detailed and usable to substantiate eligibility; it does not prescribe one particular accounting method or require all taxpayers to use project timesheets. In a shrinking-back analysis, the evidence must support the qualifying subset rather than an unsupported allocation to the entire product.

The following table summarizes the documentation shortcomings in the Eustace case as identified by the IRS and the court:

The Limitation of the Cohan Rule

The Cohan doctrine allows reasonable estimation in appropriate circumstances when the evidence establishes deductible or creditable expenditure but does not establish its precise amount. It does not relieve a taxpayer of proving eligibility or providing a rational basis for allocation. In Eustace, the Tax Court declined to estimate qualified salary costs from the inadequate evidence presented. That outcome is a case-specific failure of proof, not a categorical rule against estimation. Reliable records and credible testimony can support an allocation, but an estimate cannot substitute for showing that qualifying research occurred and that the claimed costs relate to it.

The Regulatory Shift: T.D. 9104 and the Abandonment of the Discovery Test

T.D. 9104, published on January 2, 2004, finalized regulations that rejected a requirement to advance the common knowledge of skilled professionals. The general provisions apply to expenditures in taxable years ending on or after December 31, 2003, subject to their transition provisions and later amendments. Under Treasury Regulation Section 1.41-4(a)(3), the discovery requirement focuses on eliminating uncertainty concerning the development or improvement of a business component. The uncertainty may concern capability, method, or appropriate design. Existing science and engineering principles may be used. These rules do not require a new scientific principle, a patented invention, or successful completion of the project.

The regulatory change did not remove the separate requirement to evaluate alternatives through a process of experimentation. Eustace can still illustrate why evidence of ordinary development alone is insufficient, but its restrictive discovery and feasibility formulations should not be presented as the current test. Later decisions, including Phoenix Design Group, Inc. v. Commissioner, T.C. Memo. 2024-113, examine whether the taxpayer established the uncertainties and evaluative activities required by the regulations applicable to the years in dispute.

The following table compares the historical interpretation applied in Eustace with the uncertainty-focused approach in the later regulations:

Software for Sale vs. Internal Use Software (IUS)

Applied Systems developed software for sale to independent insurance agencies. Its argument that internal-use software cases should not govern its commercial software was unsuccessful: the Seventh Circuit explained that the general Section 41(d)(1) requirements apply separately from the internal-use exclusion. The current definition of internal-use software, including software developed for general and administrative functions such as financial management, human resources management, and support services, comes from later regulations. It should not be attributed to Eustace.

Under Treasury Regulation Section 1.41-4(c)(6), internal-use software generally must satisfy the ordinary qualification rules and a high-threshold-of-innovation test, unless a regulatory exception applies. That additional test addresses innovation, significant economic risk, and the absence of commercially available software that can meet the intended purpose without qualifying modifications. Software developed for sale, lease, or license, and qualifying software enabling third-party interaction, is outside the general-and-administrative internal-use definition. Dual-function software requires further analysis of any third-party subset and the regulatory safe harbor. Eustace’s narrower point was that commercial software remains subject to the general research-credit requirements.

The following table outlines the different hurdles for software qualification based on its intended use:

Implications for Future R&D Tax Credit Applications

Eustace remains useful as a historical software-credit decision and an example of inadequate substantiation. Its practical lesson is to evaluate the work under the law applicable to the claim year and connect eligible expenses to the qualifying activities. Contemporaneous project records can make that connection easier to demonstrate. They are strong evidence, but Eustace did not establish a mandatory project-costing system for every claimant.

The Rise of Project-Level Costing and the “Substantially All” Rule

The substantially-all rule applies separately to each business component. Under Treasury Regulation Section 1.41-4(a)(6), at least 80% of the research activities, measured on cost or another consistently applied reasonable basis, must constitute elements of a process of experimentation for a qualified purpose. This is not an 80% test of all company activity, the percentage of a product that is new, or the amount of the credit. Little Sandy Coal Co. v. Commissioner, 62 F.4th 287 (7th Cir. 2023), illustrates the need for a supported calculation instead of unsupported assumptions.

Little Sandy Coal affirmed the denial because the taxpayer did not substantiate the required proportion of experimental activities. The Seventh Circuit nevertheless rejected the Tax Court’s categorical exclusion of pilot-model production activities from the experimentation numerator: such activities can count when they satisfy the relevant requirements. Merely calling a vessel or product a prototype does not establish that result. If a business component fails the test, the shrinking-back rule requires applying it to the most signific