Methodology

Where does program information come from?

From official sources: the government, agency, municipality or program operator that runs the program. Every program page links its primary official source and the date we last checked it, and — where we recorded it — quotes the line from that page we relied on.

Research is AI-assisted. AI research assistants search for programs, read the official pages and draft each record — figures, conditions, the catch — usually with a quoted evidence line. A record only goes live with a link to its official source. Anything that could not be confirmed from an official source stays in research and is not published.

AI can misread a page. That is why every record shows its source, quote & check date, and why you should read the official page before you act.

What do the program statuses mean, and what is hidden?

• Active. Open to applicants under its published rules.
• Open with restrictions. Open, but only to some applicants or in limited rounds.
• Waitlist. Accepting interest or a waitlist rather than full applications.
• Paused & expired. Not taking applicants. These move to the Archive with their catch and the reason they closed; they are never ranked or matched.
• Research. An unconfirmed lead. Not shown anywhere on CasaSedora: not in Explore, matches or our sitemap, and its page returns “not found”.

Programs shown to be misleading (for example, a widely shared “grant” the official source does not support) are kept in the Archive marked “Misleading claim”, so you can see why.

When does a program show a Verified badge?

Only when all three hold: its key facts were confirmed against the official source (or the program operator's own page) when the record was researched; our automated check read that official page within the last 24 hours and found nothing that needs a person; and no “Rule change detected” alert is open. Otherwise the badge says when the official page was last confirmed instead. A program whose official page blocks automated checks or fails to load cannot show Verified, however good its record.

• Last confirmed …. The record was verified, but our automated check has not confirmed the official page in the last 24 hours. Treat the figures with a little more care and check the official page.
• Needs review. Partly confirmed, marked low confidence, or flagged by the automated check as possibly changed. Treat the figures as a starting point.
• Not verified. Rests on press or community reports rather than an official page. Never featured.

What do the source details on a program mean?

• Official source. The primary page the record is based on, plus a secondary reference when one helps.
• Evidence. The line from the official page that supports the key facts.
• Last checked. The date the record was last compared with its source.
• Confidence (high, medium, low). A judgement recorded at each check of how firm the key figures are. High: the official source states them clearly. Medium: confirmed, but some reading of the source was needed (for example, an unclear rule or an amount given only in part). Low: we could not pin them down — a low-confidence record never shows as verified.
• Open to movers from abroad. Whether a non-citizen moving from abroad can realistically use the program: yes; “conditions apply to non-citizens”; or “citizens/residents only”, which is never shown as a match unless your citizenship fits.
• Money in US dollars. Estimates converted at a dated exchange rate, for comparison only. The official currency amount is what counts.

How often are programs re-checked?

Continuously. Our Rule-Change Radar re-reads every program's official pages at least once a day, and about every hour for programs someone has saved or applied to, for countries with a raised travel advisory, and for programs with an open alert. It reads politely (one request at a time per website, respecting robots.txt) and compares the page's text with the last version it read. Hourly checks depend on a scheduler outside our hosting plan; if it is not running, checks fall back to once a day.

It also watches official travel advisories — the US State Department and the UK Foreign Office (FCDO) — and the GDACS disaster-alert feed. When the US raises a country to Level 4 “Do Not Travel”, or the UK advises against all travel to the whole country, every program in that country is flagged at once (a level we see for the very first time, or several countries at once, waits for a person instead, as that is more likely a feed fault than news). Natural-disaster alerts only make us check that country's programs more often.

When a page changes, an AI model classifies the change, and a second, independent AI check (a different model with different instructions, which never sees the first answer) classifies it again. The page text is treated as data, never as instructions, and a serious verdict only counts if the model quotes the page word for word.

• Serious changes. The page says the program is closed, suspended or paused, the deadline has passed or the quota is used up, it reports a conflict or emergency, or the page has gone (missing on two checks in a row) — or a top-level travel advisory, or the application deadline on our record has passed. Within the hour of the check, the program shows a red “Rule change detected — confirming” banner, loses its Verified badge (programs whose page says paused move to the Archive), and everyone who saved or applied to it gets an alert in the app — and by email or phone notification when both AI checks agree (or the signal needs no AI), at most three a day. No person approves this first, so the alert can be a false alarm: always check the official source.
• Confirming. A later check that finds the same thing — and, for page changes, the second AI check agreeing — turns the banner to “confirmed”. If the page goes back to how it was, or the advisory is lowered on two checks in a row, the banner clears and the program's earlier status returns on its own.
• Smaller changes. A change to fees, thresholds, documents or steps is published as a one-line note in the program's “Rule changes” section only when both AI checks agree it is a rule change; people who saved the program are told. When the checks disagree or are unsure, the change waits for a person. Changes to wording, layout or page dates are filed as noise.
• Undo. Every automated action is logged with its evidence. A person reviews the log after the fact and can undo any action in one step: the earlier status comes back, the banner clears, and everyone alerted is told it was withdrawn.

What this does not cover: pages that block automated checks, need a login or are built entirely in JavaScript; changes published somewhere other than the program's official page (for example in a national gazette — there is no reliable machine-readable feed for most countries); emergencies not reflected in a US or UK top-level advisory; PDFs, which we can only tell changed, not how. AI can misread a page, so a banner or note is a prompt to check the official source, not a ruling. When AI classification is unavailable, only missing pages, passed deadlines and top-level advisories act on their own and every other change waits for a person.

How is the language practice content made?

Phrases, First Words prompts, Field Drills, culture quiz questions and street missions are drafted with AI. Before anything is published, a separate AI reviewer for each language — working independently of the drafts — checks grammar, natural phrasing, formal vs informal address, the regional variety (for example European vs Brazilian Portuguese), cultural safety, and that the translation matches the English. Items that fail stay hidden. Culture quiz questions also need a source the reviewer actually opened.

This is not a native-speaker review. Published items carry the label “AI-checked · native review pending” until a native speaker has reviewed them. Every item has a “Report a mistake” button (reporting needs a free account); reports go to a review queue, and an item can be unpublished in one step.

Official language requirements (levels, exams) come from our sourced records, not from AI.

Where else does CasaSedora use AI?

• Del, your relocation assistant. Answers questions using only CasaSedora's own sourced program records, cites the official source for each fact, and declines legal, immigration, tax and approval-odds questions. It never promises an outcome. When an answer cannot be sourced, it says so. You must be 18 or over and accept how Del works before first use.
• Del's memory (optional). Only if you opt in, AI picks out short notes about your move from what you tell Del — budget, income, family, pets, work, languages, timeline, climate, lifestyle, dream home, deal-breakers. Income, budget, family and anything health-related is only proposed until you confirm it; nothing is copied to your profile unless you choose to. Confirmed notes give Del context for its answers but are never a source. View, edit or delete them under “What Del knows about you” on your profile.
• Del's daily digest. With memory on, a daily in-app digest lists new programs that match your profile (the same match score as Explore), published rule changes on programs you saved or are applying to, and housing or caretaker programs that fit your dream-home notes. The items are chosen by fixed rules, not by AI, and each links to its program page. AI writes only the one-line summary at the top, from the item titles; your notes are not sent for it.
• Guide, the conversation partner. Helps you practise a language in role-play scenarios. It does not answer immigration, legal or tax questions; it points you to the official source instead.
• Practice feedback. Some practice features use AI to give feedback on what you say or type; pronunciation checks send your voice recording for scoring. It is a practice aid, not a language certificate.
• Rule-Change Radar. Classifies changes to official program pages (two independent checks) and can raise a “Rule change detected — confirming” alert on its own. It only reads public official pages — never anything about you. See “How often are programs re-checked?” above.
• Menu & Sign Decoder. Reads a photo you upload (a menu, sign or form) with AI and explains it. Don't upload photos showing personal documents or other people's details.

These features run on third-party AI services (currently Anthropic's Claude for text & photos, and Microsoft Azure speech for pronunciation checks) and only when they are switched on. AI does not decide your eligibility: match results come from the program rules on record, and only the official authority can decide an application.

How do I report an error, and what happens next?

Use “Report an error” at the foot of any program page (it opens an email to hello@casasedora.com with the program filled in), or “Report a mistake” on any practice item. An official link showing the correct information helps most.

We re-check the official source, correct the record and update its check date. If a program has closed or turns out to be misleading, it moves to the Archive with the reason rather than disappearing. We aim to reply within two business days.

Is this legal or immigration advice?

No. CasaSedora is a research tool. Rules change and individual cases differ; confirm with the official authority, and for anything binding speak to a qualified immigration lawyer or tax adviser.