Ireland FSAI Food and Allergen Alerts

Private, build/test-ready Apify Actor. Reads official Food Safety Authority of Ireland food and allergen alerts. No Apify deployment or paid run is included in this package.

Run

Python 3.12. Install pip install -r requirements.txt, then python -m src.main using Apify's local INPUT key-value record, or build the Dockerfile. Actor configuration is .actor/actor.json. Offline tests: python -m unittest discover -s tests -v.

The direct smoke input in examples/input-direct.json requests one known alert. The discovery smoke input visits one food listing and emits at most two details. Runtime additionally requests robots.txt. Default input visits one page each of food and allergen listings and emits up to 20 alerts. Pages contain 10 cards at the time of verification. No browser, proxy, authentication, or challenge solving is used.

Input and bounded scope

• alertTypes: unique food / allergen values, default both.
• startUrls: up to 50 explicit HTTPS www.fsai.ie alert-detail URL strings; default empty.
• discover: default true. Set false for direct URLs only.
• maxListingPages: 1–5 per stream; default 1. Only actual next-page links are followed.
• maxItems: 1–100 output alerts; default 20.
• issueYears: optional issue-date years (1990–2100). Applied only to fetched details, with missing dates excluded. This does not search the archive or promise all alerts from a year.
• requestDelaySeconds: minimum serial request spacing, 2–60 seconds; default 2. A longer published robots crawl delay is honored.

Discovery visits the bounded listing pages before detail requests. At most 100 discovered listing-card URLs plus 50 direct URLs are candidates. Filters and duplicate records can reduce output. RUN_REPORT records visited URLs and the limited coverage statement. Malformed listing/details and unfiltered empty output fail visibly rather than pretending success. A valid year filter can legitimately produce zero rows. All detail reads complete and every output row validates before dataset writes begin.

Output and interpretation

One row per source alert, deduplicated by alert type and alert number or normalized detail URL. Food and allergen number streams remain distinct. issuedDate is parsed from the displayed issue date; observedAt is extraction time in UTC. Missing or unrecognized dates stay null with a warning. No issue date is inferred from observation time or URL.

Published product identification, batch code, message, nature of danger and business/consumer actions are retained as source text with whitespace normalized. sourceTables preserves all published table cells; productRows exposes tables with a Product column. packSize, approvalNumber, and useByDate extract explicitly labelled values only. All-best-before-date scope remains verbatim in batchCode. Where allergens are specified inside a product table, the summary field can say “Please see table below”; consult productRows and the full message. Advisory qualifications, including possible/unintentional presence and affected consumer groups, are preserved rather than reduced to a misleading allergen list. No medical interpretation is supplied. Always check the linked current official alert before acting.

Safety and source rights

Only HTTPS www.fsai.ie food/allergen paths are accepted, including redirect destinations. robots.txt must load successfully and allow each request; failures or denial stop the run. Requests are serial, bounded to 5 MB streamed response bodies, with finite timeout and redirect limits. HTTP 401/403/429 and recognizable challenges stop immediately without bypass or retry. No page images, attachments, tracking code, forms, or contact data are fetched.

Source: Food Safety Authority of Ireland (FSAI), © Food Safety Authority of Ireland. Reuse terms: https://www.fsai.ie/re-use-of-public-sector-information . FSAI permits free reuse subject to attribution, accurate reproduction, no misleading use, no principal advertising/promotional use, and no illegal, immoral, fraudulent or dishonest use. FSAI does not warrant currency or error-free information. No endorsement or exclusive rights is implied. Downstream users must preserve these conditions and attribution. The Actor's structured fields do not replace the source.

Fixtures are sanitized official HTML observations from 8 October 2026: two food alerts (2026.61 and 2026.60), one allergen alert (2026.A26), both genuine listings, and a wrong-route Latest News listing retained solely for negative testing. Scripts, forms, inputs, iframes, images and nonessential attributes were removed. Samples are reproducible fixture output, not a deployed Actor run. Official Apify input/dataset metaschemas are vendored under tests/metaschemas and used in tests. No live Apify spend has occurred.