Documenting R&D Activities the IRS Will Accept

Credits are not lost at the point of calculation. They are lost eighteen months later, when an examiner asks which project a wage allocation belongs to and nobody can answer.

What examiners actually ask for

The standard is contemporaneous records that tie specific people and specific money to specific qualified activities. In practice, that means four things:

• A business component list — the products, processes or software the claim is built on, each one a discrete unit rather than 'the platform'.
• Technical narratives mapping each component to the four-part test, in the engineers' own language, naming the uncertainty and the alternatives evaluated.
• Payroll allocations showing what proportion of each person's time went to which component, and how that proportion was arrived at.
• Evidence of the experimentation itself — commit history, test plans, design reviews, prototype logs, failed branches.

What does not survive

A spreadsheet of departmental totals assembled after year end. A flat percentage applied across an engineering team because 'they all work on the product'. A narrative written by a consultant who never spoke to an engineer. These get disallowed not because the underlying work didn't qualify, but because nothing connects the money to the work.

The uncomfortable version: if a claim's substantiation was created after the return was filed, its weakest evidence is the study itself. Documentation written to defend a number reads differently from documentation written while the work was happening.

Section G, from TY2024

Form 6765 now carries a Section G requiring business-component-level detail on the face of the return: the component, its information-sought and uncertainty, and the officer and wage breakdown behind it. Detail that used to sit in a study you produced on request is now something you file.

The practical effect is that a claim which was never really documented is now visibly undocumented at the moment of filing, rather than at examination. For firms who had been relying on a light-touch provider, this is the change that makes the difference obvious.

What good looks like

Every number traces to a source: a wage row in payroll, a transaction in the general ledger, a contractor invoice. Interviews happen while the year is still fresh. The narrative names the projects that were considered and excluded, and why — a study with no exclusions is a study nobody stress-tested.

Done that way, the documentation is not a defence prepared for an examination. It is simply the record of how the number was built, and it reads that way to anyone who asks.