Frequently asked questions

Two expert corpora of Canadian federal law — customs and trade on one side, tax on the other — connected to Claude and ChatGPT. Each corpus carries the trade’s method, serves the official texts in the version in force at the date that matters, in both languages, and cites its source to the section. It is not a search engine: it is reasoning that holds its conclusion.

Two things no document base gives you: the method, and the know-how that applies it. The connector does not stack texts — it knows how to work them: walking a classification in the order of the rules, reading a section in its dated version, computing at the date that governs, drafting a filing to the tribunal’s standards. That craft, learned from practice, is built into the connector: documentation is the raw material; the value lives in the method.

No. Carter is a research aid for professionals. It reads and cites the official record, and it drafts for review; it creates no lawyer-, accountant- or broker-client relationship. Returns, notices of objection and other filings ship as drafts, for review by a qualified professional, and reasonable care remains with the filer. The judgment, and the signature, stay with you. The full frame is in the Terms of Use.

The official sources, kept with their dates:

• The Customs Tariff at its current revision: 10,992 lines, the full hierarchy with Section and Chapter notes; the duty is decided at the eight-digit item.
• Tariff vintages over the prescription window: a line dropped from the current tariff is still served at its date, description and rate included.
• 16 federal customs Acts — the Customs Act, the Customs Tariff, SIMA, excise — and the third-party laws enforced at the border.
• 1,197 advance classification rulings, 63 SIMA measures, and 574 D-memoranda and customs notices served as administrative position.
• The single window and its participating agencies, and the case law of the Canadian International Trade Tribunal.

The full contents are on the Carter Customs page.

• The Income Tax Act, section by section, read in the version in force in the year that governs.
• 13 federal tax Acts — including the Excise Tax Act — the provincial layer and Québec; the abatement and interprovincial allocation named, never guessed.
• 63 tax treaties by partner, article by article; the multilateral instrument that modifies them is flagged.
• 50 CRA income tax folios, served as administrative position, never cited as the law.
• The index of Tax Court of Canada case law: reference, subject and a link to the decision (see below).

The full contents are on the Carter Tax & Law page.

Yes. The tax record carries fourteen jurisdictions: the federal level, the provinces and the territories, Québec included — among them the Québec corpus and its 12,409 articles. Québec texts are reproduced with attribution and a notice of non-official status: the authoritative version remains Québec’s official publication, and that publication prevails in case of any discrepancy.

Because the law of a matter is the law of its date. Property disposed of in 2023 is read under the 2023 version of the Act; an importation of March 2024 is charged the tariff in force that day; a folio is cited only with its date. The record keeps the vintages and the dated chronologies so an answer stands on the text in force at the triggering event, not on a blended present.

It depends on the source, and Carter says so. The case law of the Canadian International Trade Tribunal is read on the merits. The case law of the Tax Court of Canada, by contrast, is served as an index of notices — reference, subject, link — not as a full-text corpus: the connector points you to the decision, which you read at the source before relying on it. Taking the index for the text would be a mistake; Carter marks the difference.

From the page of the desk you want, click “Start your trial”: claude.ai opens with the connector prefilled, name and address included. Confirm the addition, open a conversation, ask your question — the whole thing takes two minutes. You can also enter the address by hand, in Claude’s or ChatGPT’s settings (Connectors):

Carter is a subscription service — a professional plan, with a free trial to start. During the launch period the connector is open: the trial needs no payment and no card. A sign-in protects access; beyond it, the only account in play is your Claude or ChatGPT account, the one you already talk to. Pricing will be published on this site before any billing begins, and adding a connector will stay just as simple.

In Claude: claude.ai in the browser, Claude Desktop and the mobile apps, wherever your plan supports custom connectors. And in ChatGPT: the same address is added under Settings → Connectors, and the work is the same there. This site is only the front door; the work happens in your conversation.

As for the model: Carter has the assistant read long statutory passages, walk classification descents and follow chains of dates, and that reading rewards a capable model. Claude Opus and GPT-5 Thinking do the best work; lighter models will answer, with less patience for long texts.

From the served text, never from memory. An inclusion rate, a duty rate, a correction or objection deadline is read in the section, in its dated version — because a figure recited from memory would be wrong the moment the law changed. If an answer states a rate or a deadline, it cites the section and the version it comes from; absent a source, it does not state it.

The date first. The record reads texts as of the governing date: make sure the question states the entry date or the tax year you mean; an answer read as of today can differ, correctly, from an answer read as of 2021. Then the citation: substantive answers carry their source; open the cited section or decision and read it. Where a reading and the official text disagree, the official text wins. The record is maintained on the connector’s side: there is nothing to refresh on yours.

Yes. Federal law is enacted in both languages, and both are equally authoritative. Carter serves both versions of a provision and flags the gap when they diverge. A divergence is resolved by the shared-meaning rule — the meaning common to both versions (R. v. Daoust, 2004 SCC 6) — and the conclusion stays with the lawyer. For Québec texts, Québec’s official version prevails, as noted above.

The site is static: no accounts, no forms, no analytics, no advertising cookies. Type is served from Google Fonts, so your browser fetches font files from Google when a page loads; that is the site’s only third-party request. Everything else — adding and using the connectors — happens inside Claude or ChatGPT.

Your conversations live in your Claude or ChatGPT account and are governed by Anthropic’s or OpenAI’s policies. The publisher of Carter cannot see your exchanges. What reaches the connector is the specific lookups the assistant issues while consulting the record — a phrase, a section number, a date — processed to return the requested texts. Beyond what authentication requires, the connector collects no personal data. As with any research tool, give a question the facts it needs and no more. The Privacy Policy spells all of this out.

Federal law is reproduced under the Reproduction of Federal Law Order (SI/97-5), which permits reproduction without permission, provided that due diligence is exercised in ensuring accuracy and that the reproduction is identified as not being official. Québec texts are reproduced with attribution and a notice of non-official status. In both cases only the official publication is authoritative: where a served text and the official publication diverge, the latter prevails, and the record is corrected. The detail is in the Legal Notice.