The Need – Unknown – Alternatives – Evidence screen
Does this project deserve an early R&D credit conversation?
Answer five questions about one specific project, not the company as a whole.
Your answers never leave this page. Nothing is submitted or stored.
1 · Need
Did the project pursue a measurably new or improved function, level of performance, reliability, or quality?
Example: raising extraction accuracy from 65% to a target of 95%, or meeting a strict response-time limit.
2 · Unknown
At the outset, was it genuinely uncertain whether the result was achievable, which method would work, or what the right design was?
If the team simply implemented a known solution or configured vendor options, the answer is no.
3 · Alternatives
Did the team identify and evaluate alternative approaches through testing, modeling, simulation, or systematic trial and error?
Comparing models, architectures, matching methods, or prompt structures against measured results counts. Casual tweaking until it "seems better" does not.
4 · Evidence
Did results, including failures, change the technical design, and do records of that process exist or could they be kept going forward?
Evaluation datasets, test results, abandoned approaches, tickets, and architecture decisions all count.
5 · Substance
Is the development performed in the United States, with meaningful spending on employees, contractors, or compute?
Federal qualified research must occur in the US, Puerto Rico, or another US possession. A small side project rarely supports the economics.
Book a free 20-minute Fit Screen
Prefer to review on paper first? Download the one-page AI R&D Project Spotter.
This screen is general educational information, not tax advice or a determination that any project qualifies for a credit. A "yes" answer reflects your own characterization of the facts; eligibility depends on the taxpayer's specific activities, contracts, expenditures, locations, and applicable law. The Fit Screen is a free conversation and does not create a client relationship.