What documentation does the IRS accept for an R&D credit?

Records created while the research happened: project plans, design documents, test protocols and results, engineering notebooks, commit histories, meeting notes, and time or payroll records tying specific people to specific business components. Treas. Reg. §1.41-4(d) requires records substantiating the claimed activities; courts routinely reject credible testimony unsupported by contemporaneous records as insufficient on its own.

Are estimates ever acceptable when records are incomplete?

Sometimes. The Cohan doctrine lets courts estimate amounts where the taxpayer proves an expense occurred but not its precise size — but it requires a reasonable basis for estimation, is applied grudgingly, and is statutorily overridden where the Code demands strict substantiation, as Section 274(d) does for listed property. Cohan mitigates gaps; it does not replace a records system.

How long should specialty tax documentation be retained?

At least until the assessment statute closes for every year the item affects — which for carryforward items means the statute for the year the carryforward is used, not just generated. A research credit carried forward up to twenty years, or an NOL used a decade later, keeps its source-year documentation relevant for that long. Retain the study, the workpapers, and the underlying source records together.

Audit-ready documentation: what holds up and what does not

Every specialty tax claim is a factual assertion: this employee did qualified research, this wall is 5-year property, this building beat the energy baseline. On examination, the question is never whether the assertion sounds plausible — it is whether records made at the time support it. The burden of substantiation sits with the taxpayer under Section 6001 and Welch v. Helvering's long shadow, and the pattern across decided cases is monotonous: contemporaneous records win, after-the-fact reconstructions lose.

What holds up

The common trait of persuasive evidence is that it was created for a business reason other than tax, at or near the time of the events:

Time and payroll records tying named individuals to named projects — the backbone of any wage-based research credit claim, and now effectively demanded by Form 6765's Section G business-component reporting. Even quarterly project-level allocations beat annual guesses; ticket systems and commit logs beat both for technical staff.
Design and development artifacts: requirements documents, drawings, test plans and failure logs, sprint records, lab notebooks. These prove the process of experimentation Section 41(d)(1)(C) requires — uncertainty, alternatives, evaluation — better than any memo describing it later.
Invoices, contracts, and payment records establishing amounts, dates, and — critically for contract research — who bore the financial risk and retained rights, the funded-research battleground of cases like Fairchild Industries and its successors.
Third-party engineering detail: for cost segregation, the contractor's cost breakdowns, pay applications, and drawings the study relied on. A study is only as strong as the source documents behind its allocations, which is why the workpapers belong in the taxpayer's hands, not just the provider's.

What does not

After-the-fact narratives are the recurring failure. Interview-based "oral history" studies that generate qualification memos years later; time allocations assigned by managers estimating in hindsight; project descriptions written to track the four-part test's language rather than the work. Courts have been explicit — in Little Sandy Coal Co. v. Commissioner and a line of research credit cases before it, professionally prepared studies failed because the substantiation beneath them was estimation and recollection. The Cohan rule offers limited mercy for amounts, none for qualification, and none at all where strict substantiation statutes like Section 274(d) apply.

The same logic reaches deductions: a retroactive Section 174A amended-return claim, or a 481(a) catch-up from a method change, still requires the underlying cost detail for the years being trued up.

Documentation as a system

The practical program is unglamorous:

Capture in the ordinary course. Configure timekeeping and project systems so tax-relevant detail is a byproduct of operations, decided at the start of the year — not an annual archaeology project.
Map records to claims annually. When the return is prepared, index which documents support which business components, assets, or certifications. Gaps found in year one are fixable; gaps found on exam are not.
Retain to the live statute. Carryforward items keep source-year records relevant until the statute closes on the year of use — up to twenty years for research credits under Section 39, indefinitely for NOLs. Store the study, workpapers, and sources together; see the statute of limitations brief for the clocks.

Documentation is also the cheapest insurance in this field. The marginal cost of decent time tracking rounds to zero; the cost of a failed credit is repayment, interest, a possible Section 6662 penalty, and — in a transaction — a purchase-price haircut when diligence finds the file empty.