About NDP Search
NDP Search is an independent, non-commercial research tool for the Immigration and Refugee Board of Canada's National Documentation Packages (NDPs), the country-conditions evidence used in refugee hearings. It helps you find what is in an NDP: searches return short excerpts with pinpoint citations, and every excerpt links back to the original document on its publisher's website.
Why this exists
Refugee hearings decide whether someone can stay in Canada or must return to the place they fled. They are proceedings where the evidence can be a matter of life and death, yet that evidence is surprisingly hard to work with. The IRB's National Documentation Packages are the backbone of country-conditions evidence in nearly every claim, but they exist as long lists of links on a website: more than two hundred packages, each holding dozens or hundreds of documents, thousands of pages deep, scattered across publishers all over the world.
The packages are also replaced wholesale several times a year, with a difficult to follow record of what changed. A document you cited last month may have been updated, renumbered, or removed. There has been no practical way to find out what meaningfully changed short of reading both versions side by side.
There's a deeper problem, too. Refugee proceedings are private (for good reasons) but the shape that privacy has taken over time has created a serious structural information deficit. There is no body of published outcomes to learn from, so it is hard or impossible for practitioners and the public to know how the IRB tends to approach different kinds of claims, which evidence decision-makers rely on, or what adjudicative theories they need to answer. Many refugee claims are, consequently, prepared in the dark. And there is poor public visibility into the decisions of a system that plays such a significant role.
NDP Search is an attempt to close some of that gap with the public record we do have. It unifies the public sources — every NDP document, every version it can capture, and the published decisions that cite them — and gives them a modern research interface: full-text search with pinpoint citations, version-to-version comparison, and visibility into which evidence actually gets cited in decisions. None of this is secret information; it was always public. It was just unusable.
I'm a former refugee lawyer and adjudicator. I've used my experience of what matters in practice to design something on modern principles and technology. The fundamental goal is to improve access, transparency and fairness inside a really important public-information problem. I hope you find it useful.
A note of caution
This is an independent research project not associated with the Immigration and Refugee Board of Canada. NDP Search is an experimental tool under active development. It can and does contain errors and limitations. Treat it as a way to find evidence, never as the evidence itself. AI-generated material (summaries, litigation themes) is labelled where it appears and can be wrong. Always verify against the original sources, which are linked from every document.
How current is it?
The tool checks the IRB website for new and updated NDPs every week. When something has changed, it fetches each new or updated document from the original publisher's website and then converts the downloaded file into searchable text. For scanned documents, that conversion includes optical character recognition (software that reads text out of page images).
Each country page shows the date of the NDP version you're looking at, and the "Previous NDPs" tab shows what changed when a new version replaced an old one. (This feature is in development.)
Known limitations
- Text read from scanned pages can contain recognition errors, especially in tables, footnotes, and low-quality scans.
- Image-heavy documents and maps do not convert well to text. A map's searchable content may be little more than its title.
- Some documents resist automated download entirely (paywalls, bot protection, retired websites), so a small share of each NDP may not be searchable. Country pages mark these as "Not yet available".
- The whole pipeline is automated. Errors that a human editor would catch can slip through, which is why every excerpt links back to the original document.
What does search actually search?
The documents in each country's current NDP. Search matches both exact words and meaning: a search for "forced conscription" will also find passages about compulsory military service. Results are short excerpts — not full documents — each tied to an NDP item number and a link to the original document on the publisher's website.
How does the case law connection work?
The Decisions tabs search Canadian refugee decisions through the A2AJ database. The tool also scans those decisions for references to NDP items; that's where the "cited in X decisions" labels come from. They show how often decision-makers have relied on a piece of evidence. The counts are version-aware: "Cited in X decisions" counts decisions that cited the document you are looking at (including in earlier NDP versions where its content was unchanged, and following renumbering), while "Predecessors cited in X decisions" counts decisions that cited the items the current document replaced — or where the NDP version can't be determined — meaning a different text than the one shown.
NDP Search is not an official product of the Immigration and Refugee Board of Canada or the Government of Canada. It is an independent research tool. It is not affiliated with, endorsed by, or associated with the IRB, the Government of Canada, or any of the publishers of the public documents that appear in the NDPs.
Use of this website, its search tools, or its AI-assistant (MCP) interface constitutes acceptance of these terms.
- This tool is provided "as is", without warranty of any kind.
- It does not provide legal advice and does not replace professional legal counsel.
- Document text is extracted automatically and may contain errors. You are responsible for verifying everything against the original sources before relying on it.
- The AI-generated features (search summaries and litigation themes) are experimental. They may contain errors, omissions, or invented content, and must not be relied on without checking each statement against the cited source.
Copyright
NDP Search is a non-commercial project. Use of the materials on this site is exclusively for research, private study, and education.
The NDP compilation structure (item numbers, section organization, document selection) is © Immigration and Refugee Board of Canada. The underlying document content is © its respective publishers and remains available from those publishers. This tool does not republish full documents. The website displays short excerpts to support research, and links every excerpt back to the publisher's original source, where the document itself should be consulted. Connected AI assistants can additionally retrieve limited page-level passages for verification, subject to the limits on automated access below. AI-generated summaries are produced by this tool and do not represent the views of the IRB or any publisher. Rights holders who would prefer a document not appear can use the contact page.
Automated and AI-assistant access
Connecting an AI assistant or other automated tool to NDP Search is permitted for the same purpose as the website: legal research, with verification against original sources. You may not use automated access to bulk-download, reconstruct, or redistribute the document collection or any substantial part of it — as files, as a database, or as training data. To enforce this, connected accounts are subject to fair-use rate limits: a cap on requests per minute and a daily limit on how much full document text can be retrieved. These are set well above ordinary research use, reset each day, and exist to prevent automated bulk extraction; reaching them is a sign of automated harvesting rather than normal work. Citations and the accuracy of anything your assistant produces from this service remain your responsibility. Automated access is best-effort and may be limited or withdrawn for any account that degrades the service for others. Signing in to connect constitutes acceptance of these terms.
Using the NDP Search website requires no account. Connecting an AI assistant requires a one-time sign-in (see below). Either way, nothing about you is sold or shared. The service keeps standard server logs (pages visited, timestamps, network addresses) and anonymous usage counts, used only to keep it running and understand load. Search queries are processed to return results and are not tied to an identity. Finding and ranking the passages that match your search runs entirely on this server, so the text you search is not sent to any outside company to perform the search itself. (The optional AI summary is a separate step, described below.)
Third-party services
The site relies on a small number of outside services, each with its own terms and privacy practices that may affect you:
- Cloudflare routes the site's traffic, so your requests pass through its network.
- Anthropic (Claude) processes the optional AI features: when you use an AI summary or litigation themes, your query and the retrieved passages are sent to Anthropic's Claude service to generate the text.
- A2AJ receives case law searches made from the Decisions tabs.
- WorkOS operates sign-in for AI assistant connections. It holds your email address and issues the credentials your assistant uses to stay connected. No NDP content and no search activity passes through it.
AI assistant (MCP) connections
If you connect an AI assistant to NDP Search, your conversation with it never reaches this service. What gets sent to the server is determined by your AI tool, typically short search requests the assistant composes. The wording of those requests is not stored. All that's kept is anonymous usage and thematic counts. To be clear, this means nothing you type into an AI assistant or the terms it uses to engage with NDP Search are stored on this service.
Connecting requires a one-time sign-in that collects your email address, used to keep you signed in and to apply fair-use limits. Sign-ins are removed on request: your assistant simply loses access, and nothing else about you exists to delete.
Questions about any of this: the contact page.
Maintained by Clifford McCarten. Built with support from A2AJ.