Canadian NI 31-103 for solo crypto-quant devs: what you can and can't build without registration

Not legal advice. I'm a builder, not a lawyer. What follows is the regulatory map I built for myself after reading National Instrument 31-103 end-to-end, talking to three Canadian securities lawyers (one paid, two acquaintance coffees), and shipping a signal product in public. If any of this sounds load-bearing for your business, hire a lawyer in your province. OSC, AMF, and BCSC all have distinct interpretations.

Canada has a structural advantage for solo quant devs that most US-centric writing ignores: you can ship a signal service without registering, as long as you stay on the correct side of four hard lines. You also have access to SR&ED — a refundable tax credit that can return 35–65% of your R&D spend while you build. The combination is genuinely excellent.

This is the map I follow. Full tactical chapter (worked examples, SR&ED timesheet templates, sole-prop-to-CCPC cutover with actual CRA forms) is at cipher-x402 for $0.25 USDC. The free version of the playbook and the 150-page solo dev guide is at cipher-starter. Prior articles on the playbook itself add context.

The four hard lines

NI 31-103 (harmonized across provinces via CSA) triggers registration when you cross any of these lines for another person. "Another person" here includes subscribers, Discord members, and anyone who pays you any amount for anything.

Line 1: Recommend a specific security

Telling a subscriber "buy SOL today" is advice. Telling them "here is a momentum score for SOL; the score is currently 0.8 out of 1" is information. The distinction matters enormously.

You cross the line when your content answers "should I buy this?" Stay on the safe side when your content answers "is there a statistically interesting signal in this instrument?" The second framing is indistinguishable from what Bloomberg, TradingView, and any finance newsletter does. The first is what registered portfolio managers and investment advisers do.

Concrete rules I follow:

• Never use the word "recommend."
• Never answer "what should I do in my portfolio?"
• If a subscriber asks "should I buy X," the response is either a link to the methodology doc or "that's a PM question, here are links to CIRO-registered PMs."
• Publish the scoring rubric. If your signal is reproducible from public data, it's research, not advice.

Line 2: Personalize the advice

Even a non-recommendation becomes registration-triggering if it's personalized to a specific client. "Here's a momentum score for SOL" is fine. "Given your account is 60% SOL, here's a momentum score for SOL that factors in your concentration" is investment advice.

The subtle trap: a chat bot that accepts a wallet address and returns tailored "hygiene flags" (e.g. dust, stale stakes, low-liquidity positions) is personalized — but it's analyzing a wallet, not advising an investor. As long as the output is descriptive ("your wallet holds 12 low-liquidity tokens with < $1k daily volume") and not prescriptive ("you should sell these"), it's on the safe side.

Rules I follow:

• Any per-user output must describe, not direct.
• No "given your risk tolerance..." branches anywhere in the product.
• Onboarding questionnaires must be clearly scoped to product calibration (which symbols to watch), not portfolio fit.

Line 3: Custody

Touching a client's funds — holding a key, signing a transaction, co-managing a multisig, operating a pooled account — is the category with the highest registration bar in Canada. It is Portfolio Manager (PM) or Investment Fund Manager (IFM) territory, and it requires capital, a CCO, a compliance program, O&E insurance, and audited financials. The bar is high for good reason.

Rules I follow:

• Never hold a user's keys.
• Never co-sign on a user's multisig.
• Never accept user deposits, period.
• Webhooks trigger actions in the user's wallet, initiated by them, on their own infrastructure.
• The signal engine and any execution engine must be legally separable. Execution = subscriber's problem.

Line 4: Pooled investing

The moment you aggregate subscriber capital into a common pool for a common strategy, you are operating an investment fund. That is the Investment Fund Manager regime plus, typically, a prospectus exemption (offering memorandum, accredited-investor-only, etc.). It's registration on top of registration.

Rules I follow:

• No pooled strategies.
• No "I'll trade with $X of your money and you keep Y%" arrangements.
• No proprietary fund the subscribers are loosely exposed to.
• If you run a public bot that trades your own capital, it's your capital — that's not a fund.

The simple test

If a subscriber reads your output, has to make their own decision, executes their own trade on their own exchange with their own keys, and bears their own loss, you are a publisher. If any of those "own"s becomes "shared with me," you are building a regulated entity.

The three registrations, from cheapest to most painful

If you do cross a line, know which category you're in. These are the typical Canadian categories:

Portfolio Manager (PM)

• Triggered by: discretionary management of client accounts, personalized advice.
• Capital requirement: $100k working capital minimum (NI 31-103, Part 12).
• Must have a CCO and a UDP (Ultimate Designated Person).
• Must maintain O&E insurance, fidelity bond, and audited books.
• Proficiency: CFA Charter + 12 months investment management experience, or CIM + 48 months. There are other paths, but these are the common ones.

Investment Fund Manager (IFM)

• Triggered by: operating an investment fund.
• Capital requirement: $100k working capital or 0.1% of AUM (whichever is greater).
• Most people stack PM + IFM because you usually need both.
• Proficiency: CCO / UDP requirements similar to PM.
• Required insurance + audited financials.

Money Services Business (MSB, FINTRAC)

• Triggered by: dealing in virtual currencies (exchange, transfer, payment processing) for clients.
• Not a securities registration; it's an AML/CTF registration under FINTRAC.
• Capital requirement: none in the registration itself, but you must maintain a compliance program, report transactions over $10k CAD, and update KYC.
• This is the one that catches crypto-native builders by surprise — running an automated swap router for users is almost certainly MSB territory.

For a signal-only SaaS, you need none of these. Stay in "publisher" mode.

The exemption path: signal service as research

CSA staff notices have repeatedly treated bona fide research differently from advice. The rough tests applied:

1. Does the content describe market conditions or a security in general terms? Research.
2. Is the content available to all subscribers on equal terms (no per-client customization)? Research.
3. Does the content answer "what should I do"? Advice.
4. Is the content delivered as part of a regulated financial relationship (account management, discretion)? Advice.

The signal service I run — momentum, relative strength, sentiment layer scores on public crypto and stocks, daily — scores all four tests on the research side. The subscriber pays for access to the research feed. That model maps cleanly to financial newsletter publishing, which has been an unregistered activity in Canada for decades.

Two further practical safeguards I put in place and recommend:

• Terms of service explicit: "This is a research publication. Nothing in the product constitutes investment advice. The publisher is not registered under NI 31-103 and is not acting as an investment adviser or portfolio manager."
• No per-subscriber branching in the signal engine. One score, one rubric, one feed. Each subscriber gets the same output.

SR&ED: 35–65% of your R&D back

The tax side of the Canadian advantage is SR&ED. It refunds a big chunk of your R&D spend, including your own labor. This is the single largest reason to build from Canada.

The federal credit, as of Bill C-15 (March 2026):

• Refundable 35% on the first $6M of qualifying Scientific Research and Experimental Development expenditures for Canadian-Controlled Private Corporations (CCPCs). The cap was raised from $3M.
• Non-refundable 15% above the cap or for non-CCPCs.
• Covers salaries, contractor fees (at 80% of cost), materials consumed, and a prescribed overhead proxy (usually 55% of eligible salaries).

Provincial top-ups (refundable):

Quebec is the highest-combined jurisdiction. For many solo quant builders, incorporating in Quebec pushes the effective R&D subsidy into the 55–65% range.

What counts as SR&ED for a quant dev:

• Building novel signal models that advance understanding in a measurable way. "We tried X technique on crypto data and validated it against Y benchmark" is claimable.
• The experimental iteration on hyperparameters, when documented with a hypothesis, an experiment, and a result.
• The infrastructure engineering directly supporting the experiments (data pipelines, backtest harness).

What does NOT count:

• Routine production work.
• UI polish.
• Re-implementing a published algorithm without improvement.
• Documentation and marketing.

The ask is the paperwork. SR&ED claims live or die on contemporaneous documentation. I keep:

• A dated experiment log (one markdown file per experiment, with hypothesis / method / result / next-step).
• A git history that maps commits to log entries.
• A weekly timesheet allocating labor hours across experiments.
• Copies of all contractor invoices with R&D line items flagged.

Here is a minimal CSV template I use. It's the format my accountant wants for the T661:

week_start,experiment_id,hours,task,hypothesis,result
2026-01-06,EXP-001,18,Kronos on SPL momentum,Kronos outperforms Prophet on 1-day SPL momentum,MAE 0.043 vs 0.058; adopted
2026-01-06,EXP-002,12,Oracle-gate bps threshold,Fixed 50bps vs dynamic age-based,Dynamic reduces false-trades 22%; adopted
2026-01-13,EXP-003,22,Confirmation matrix IC weighting,IC-weighted beats static 1/n,Sharpe 1.8 vs 1.4 paper; adopted

And a Python script to generate a quarterly SR&ED-ready summary:

"""sr_ed_summary.py - aggregate a weekly timesheet for SR&ED filings."""
import csv
import sys
from collections import defaultdict

def summarize(path: str) -> None:
    by_experiment = defaultdict(lambda: {"hours": 0.0, "weeks": set(), "task": ""})
    with open(path, newline="", encoding="utf-8") as f:
        for row in csv.DictReader(f):
            k = row["experiment_id"]
            by_experiment[k]["hours"] += float(row["hours"])
            by_experiment[k]["weeks"].add(row["week_start"])
            if not by_experiment[k]["task"]:
                by_experiment[k]["task"] = row["task"]

    total = sum(e["hours"] for e in by_experiment.values())
    print(f"Total hours: {total:.1f}")
    print(f"{'Exp':<10}{'Hours':>8}{'Weeks':>8}  Task")
    for k, v in sorted(by_experiment.items()):
        print(f"{k:<10}{v['hours']:>8.1f}{len(v['weeks']):>8}  {v['task']}")

if __name__ == "__main__":
    summarize(sys.argv[1])

Run it before your quarterly CRA check-in. Your accountant will thank you.

Sole-prop to CCPC: when to flip

Refundable SR&ED at the 35% federal rate requires CCPC status. A sole proprietorship gets only the 15% non-refundable rate. This is a ~20 percentage-point difference on every dollar of R&D spend — enormous.

The case for staying a sole prop early:

• Zero incorporation cost.
• No corporate tax filing.
• Losses offset other personal income.

The case for flipping to CCPC:

• The 35% refundable rate is only for CCPCs. If you're spending > ~$20k/year on R&D (including your own labor at a reasonable rate), the refund alone pays back incorporation costs in year one.
• Limited liability matters more once subscribers pay.
• Stock-option-like compensation for future hires.

My rough threshold: flip when your monthly product revenue or R&D spend crosses $2k. At that rate you're producing > $24k/year of SR&ED-claimable expenditure, and the federal + provincial credit differential pays for the accountant.

The mechanics in Ontario (sole prop → Ontario CCPC):

1. Incorporate federally or provincially ($300–$600).
2. Open a corporate bank account; most banks want $100–$500 minimums.
3. Transfer business assets via Section 85 rollover (defers tax on latent gains; requires a T2057 filing and an accountant).
4. Register for HST (required above $30k/year revenue).
5. Get a CRA business number and payroll account (required if you'll pay yourself a T4 salary).
6. File year-end T2 + T661 (SR&ED).

Expect $2k–$4k/year in accounting fees for a small CCPC filing SR&ED. Expect $10k–$40k/year back in refunds.

Putting it together for a solo crypto-quant dev

The operating pattern that keeps you on the safe side of NI 31-103 and inside the SR&ED envelope:

1. Ship a signal service as a publisher. Generic scores, no personalization, no custody, no pooling. Terms of service explicit.
2. Keep user execution in the user's wallet. The product emits signals; users trade on their own infrastructure. No account integrations that hold funds.
3. Document every experiment from day one. Weekly timesheet + per-experiment markdown. This isn't overhead — it's the paperwork that converts a 65 cent R&D dollar into a 100-cent R&D dollar.
4. Flip to a CCPC when R&D crosses $20k/yr. Quebec if you can, Ontario if you can't, and always audited-friendly bookkeeping.
5. Stay in your lane. When subscribers ask "should I buy?", the answer is "I can't tell you that — here's the scoring rubric and three CIRO-registered PMs in your area."

The deeper chapter

Full worked examples — the SR&ED filing I used, my TOS template, the sole-prop-to-CCPC section-85 rollover walkthrough, and the exemption-path research I cited to my lawyer — are at cipher-x402.vercel.app/premium/canadian-compliance behind a $0.25 USDC paywall. The 150-page free playbook is at cipher-starter on GitHub. The earlier writeup on what building the playbook taught me adds the meta-context.

If you're a Canadian crypto-quant dev and you've been dithering on the compliance question, the rough answer is: it's more workable than you think, if you stay on the correct side of the four hard lines and document your R&D. SR&ED will reimburse a meaningful chunk of your runway.

Again — not legal advice. Please hire someone in your jurisdiction before you take this piece as gospel.