Documenting R&D activities the IRS will accept

A research credit is only as good as the file behind it. The computation takes a week; the documentation is what decides whether the claim survives an examination three years later, when half the engineers who did the work have moved on.

Contemporaneous beats reconstructed

The word that matters is contemporaneous. Records created while the work happened carry weight that records created during a study do not. The good news is that most engineering organizations already generate excellent contemporaneous evidence and simply never think of it as tax documentation: tickets, commit history, design docs, test results, sprint boards, architecture decision records, failed branches.

A study that captures those artifacts and maps them to the four-part test is doing something fundamentally different from a study that writes narrative prose after the fact and hopes it reads well.

What a defensible file contains

• A project list — the unit of analysis. Qualification is argued per project, not per department.
• A technical narrative per project, written against the four parts: the uncertainty at the outset, why it was technological, the alternatives evaluated, and how they were tested.
• Wage allocations with a stated basis — who worked on what, at what percentage, and how that percentage was arrived at. Interview notes, time data, or role-based allocation with the reasoning written down.
• Evidence of the experimentation itself: iterations, prototypes, benchmark results, the approaches that were tried and abandoned.
• The linkage from each QRE row back to its source document. Every number should be traceable without a phone call.

The shortcuts that don’t survive

A flat percentage applied to the whole engineering department, with no explanation of where the percentage came from. A narrative that describes what the product does rather than what was uncertain about building it. Projects defined so broadly that the entire year is one project. Executives allocated at high percentages with nothing showing direct supervision of qualified research.

Each of these is defensible-sounding until someone asks the second question, and an examination is entirely second questions.

Form 6765 and Section G

The revised Form 6765 asks for business component detail directly on the return — descriptions and the split of qualified expenses by component in Section G. That shifts documentation from something a company might be asked for into something it reports up front, which raises the floor on how organized the underlying file has to be.

Firms that were already building component-level studies barely noticed. Firms that were computing a top-down number and writing a narrative around it had a hard year.

Retention

Keep the file for as long as the credit can be examined or carried forward, which is longer than most retention policies assume — a credit carried forward can be questioned in the year it’s used, not only the year it arose. We retain study files seven years, and we’d rather over-retain than explain to a client why the evidence for a claim they’re still using no longer exists.

General and educational, not tax advice, and not a substitute for review of a specific taxpayer’s records.