What qualifies as SR&ED

Scientific Research and Experimental Development is defined in the Income Tax Act, and CRA reduces the definition to a pair of tests: why the work was done, and how. Work qualifies when it was done to advance scientific knowledge or achieve a technological advancement — and when it was done by systematic investigation rather than trial and error or routine engineering.

The three questions

Every project on a claim is judged on the same three questions, and the T661 asks you to answer them in writing:

1. Was there a scientific or technological uncertainty? Something you needed to achieve, that could not be worked out from generally available knowledge or from the experience of a competent professional in the field. It is not enough that you did not know; the point is that the knowledge did not exist to be looked up.
2. Did you investigate it systematically? Hypotheses formed, experiments or analysis designed to test them, results recorded, conclusions drawn, the next step chosen from those conclusions. The pattern matters more than the outcome — a failed experiment done properly is SR&ED; a success arrived at by guessing is not.
3. Was the goal an advancement? New knowledge, or a technological capability that did not exist before, even if it did not succeed. A new product is not an advancement by itself; the advancement is in what you had to learn to build it.

What counts, and what does not

Three kinds of work qualify: basic research, applied research, and — by far the most common — experimental development, which is work done to create new or improve existing materials, devices, products or processes. Support work counts too when it is directly in aid of qualifying work: engineering, design, data collection, testing, and so on.

Work that is expressly excluded, however well done: market research and sales promotion; quality control and routine testing; research in the social sciences or humanities; prospecting for minerals; commercial production of a new product; style changes; and routine data collection.

Software

Software projects qualify on the same tests, and fail on the same ones. Building something new with known techniques is development, not SR&ED, however difficult it was. Building it where the known techniques did not work — where performance, scale, integration or a novel algorithm forced you to experiment — is where the claim lives. The uncertainty has to be technological, not commercial: not knowing whether customers will want it is not SR&ED.

Signals reviewers look for

SREDDY screens each project the way CRA's own pre-claim tools do, by reading what actually happened against the tests above. Things that support a claim:

• No publicly available solution, or the available knowledge was insufficient for the objective.
• Hypotheses formulated and validated by experiments, with data analysed and fed back into the next attempt.
• Iterative prototypes driven by test results.
• Methods modified on the basis of a new understanding, not a known practice.

Things that count against it, or that a reviewer will probe:

• Applying known industry or company practice.
• Hiring someone who already knew how.
• Testing within the normal parameter range of the equipment.
• No data collected — a systematic investigation produces records, and a project with none has an evidence problem even where the work was real.

None of this is a determination. CRA decides eligibility on the facts of each claim; a screen tells you where the questions will fall.

The three write-ups

For each project, the T661 asks three questions, with word limits:

Line 242 should say what was known at the start and why it was not enough. Line 244 is a chronology — what you tried, in order, and what happened — and must identify any work a contractor did. Line 246 is what you learned, whether or not the product shipped.

What SREDDY does with this

Your AI reads your records — specs, tickets, commit history, test logs, meeting notes — and drafts each project's answers in your words; SREDDY checks the drafts for the three questions, the chronology, the word limits and the contractor disclosure, and flags the signals above where the record is thin. What goes on the form is your account of your own work. It is never templated and never borrowed from another claim.