Documents: what to give SREDDY

SREDDY works from your documents, not from a questionnaire. You never fill a template and you are never asked a tax question — give it whatever exists in whatever form it is in, and it works out what the documents support, decides the rest, and tells you what it decided. Every booked figure records which file it came from.

Start with the work, not the accounting

The R&D record is what decides whether you have a claim at all, and it names your projects. Costs only quantify a claim that already exists — so they come second, and they attach to projects SREDDY has already found. It is also the material you already have on your machine, where payroll usually means asking a bookkeeper.

1. The technical record. Specs, design notes, test logs, tickets, commit history, post-mortems, architecture docs, meeting notes — especially from when it was not working. These name your projects, decide whether the work qualifies, and become your write-ups. This is the evidence CRA cares about most.
2. Who did the work and what they cost. A payroll or salary export — CSV is ideal, Excel and PDF are fine. Bonuses shown separately helps, because some pay counts differently.
3. How time was spent. Timesheets if you have them; if not, a Jira or GitHub export, a project tracker, even a calendar. Time is the single biggest lever on the number, so records beat memory. No time records at all? SREDDY decides a split and tells you what it used.
4. Invoices and funding. Contractor and supplier invoices or an A/P export. Any grant or funding letters (IRAP, Alberta Innovates, and the like) — funding reduces the claim, so it has to be in; leaving it out is the kind of surprise you do not want later.
5. Last year's T661, if you have claimed before. That is the SR&ED form filed with your corporate return — ask your accountant for it by that name and it takes them a moment; ask for "last year's claim" and they will ask you which schedule. For the Alberta side it is the AT1. The Alberta grant pays a higher rate only on growth over the two previous years, so SREDDY needs those two figures and that filing has them. Never claimed before? Say so once and it never comes up again.

Things SREDDY asks for by name, rather than asking you

Some figures live in documents you already have, so SREDDY asks for the file rather than expecting you to know the answer — who owns what, the company's size on last year's balance sheet, your business number off any CRA letter. It names the document the way your accountant files it, because a form number is the fastest way for either of you to find the right piece of paper, and it says what the document is worth so you can decide whether it is worth digging out. What it will never do is explain the form's internals or ask you a question the form answers. If the document does not exist, say so: SREDDY takes a stated position, shows what that position costs, and moves on.

Getting files in

One folder is the tidy way (name it anything — SRED-2026 works). If your AI cannot read a file type from disk — some setups struggle with Excel or PDF — upload the file directly into the chat instead; that always works. Either way, if your AI can read the file at all, SREDDY can take it: the file name does not matter, only whether the content comes through. Send everything and let SREDDY triage: it says what it booked, what it skipped, and why. Do not pre-filter the folder yourself.

What SREDDY does with them

Each file is read once, remembered, and skipped if unchanged next time. Every person and dollar booked carries its source file. If a file looks like a template or sample rather than real data, SREDDY flags it instead of quietly booking it. Your files are never modified; SREDDY's own outputs go in a SREDDY/ subfolder.