Data

Using the data

All 797 records are open to read, one by one or a page at a time, with no account and no key. That is not a trial and it will not be withdrawn: a register nobody can cite is not doing its job, and the people this exists for — a pharmacist checking a batch, a regulator confirming what another authority did — should never have to ask anyone’s permission.

Taking a copy of the whole register is a different request, and that one we ask about. It is the work described on the about page: notices read, batch numbers pulled out of PDFs and checked, dates decoded, every source archived.

Free to consult, and we mean any of it. Every record, every field, every filter — up to 50 at a time and as far as the first 200 matches of a query, which is everything we hold for every country except Egypt. Beyond that, or as a single file, write to us. Institutions, researchers and authorities are exactly who it is for.

Endpoints

GET /api/v1/notices?cc=ZM&action=recall&from=2026-01-01

Regulatory actions, filterable on every classification axis. Add format=csv for a spreadsheet.

GET /api/v1/notices/{alert_id}

One record with its batches, version history and context.

GET /api/v1/batches/WL25024

Batch lookup across every authority at once. The highest-intent query on the service.

GET /api/v1/manufacturers/{slug}

Every recorded action naming a manufacturer, continent-wide.

GET /api/v1/events/{slug}

Cross-border clusters with the propagation lag computed.

GET /api/v1/authorities · /api/v1/coverage

Reference data, including the coverage matrix as JSON.

Sending us records

for authorities

POST /api/v1/submissions

For an authority whose own system already holds its recalls. Post them rather than retyping them. Reading needs no key; sending does, because a key names an organisation and can be withdrawn.

A submission is stored, not published. What you send is kept exactly as sent and waits for a person to read it against the notice. Nothing posted here reaches the register until they have. The response says received, and that is the whole of what it claims.

A submission is recorded at tier 2 — confirmed to us directly by the authority — or tier 3, a named credible secondary source. Tier 1 means the authority published the notice on its own channel and we captured it there, which a POST is not. Tier 0 is the rank only a named officer of the issuing authority can create, and it is refused here in as many words.

Every submission in a request is checked before any of it is stored, so your system and this register never disagree about what was accepted. A refusal names each field and says what is wrong with it. Where a submission is later set aside, the row keeps its place and states the reason.

Send the notice with it. Post as multipart/form-data with a submission part holding the JSON and a document part holding the file — PDF, Word, JPEG, PNG, HTML or plain text, up to 4 MB. One document to one submission: a batch export is a list of actions with no single notice behind it, and a form that pretended otherwise would attach the wrong file to the wrong recall. The response quotes the SHA-256 we computed on receipt, so you can check it is the file you meant before it is archived.

A submission with no document is still one we want. It will say on its face that no published notice was found, which is a legitimate thing for a record to say. Silence is not.

Do not send us a link to your own notice. If you have published it, that is a tier 1 record and belongs to us to capture from the channel that published it — so source_url is refused here. Send the address of that channel to notices@drugeyeafrica.org instead and we will capture this notice and every one after it.

Two other doors exist and need no software at all: an email to notices@drugeyeafrica.org, or the export your system already produces, sent as a CSV. Both go through the same reading.

Record shape

openFDA-aligned

{
  "alert_id": "DEA-KE-2026-0015",
  "cc": "KE",  "country": "KE",  "regulator": "PPB",
  "regulator_ref": "REC/2026/015",
  "date_published": "2026-07-20",
  "product_name": "Flazart",
  "inn": null,  "manufacturer": null,
  "batch_numbers": ["WL25024","WL25025","WL25026"],
  "action_type": "recall",
  "classification": null,
  "defect_category": null,
  "verification_tier": 1,
  "source_url": "https://web.pharmacyboardkenya.org/...",
  "no_published_notice_note": null,
  "source_archive_url": null,
  "source_archive_held": false,
  "source_sha256": null,
  "source_retrieved_at": null,
  "status": "active",  "version": 1,
  "publication_tier": "B",  "noindex": true
}

Nulls are returned, not omitted. A consumer can tell the difference between “the authority did not state this” and “we did not capture it”. Every record carries verification_tier, source_url, no_published_notice_note and source_archive_url, whatever their values.

source_archive_url is null unless the archived copy is on storage you can fetch. A copy on a machine you cannot reach is not evidence, and a path naming somebody’s home directory is not an address; the register does not publish one as though it were. Where a copy exists but is not yet reachable, source_archive_held is true and source_sha256 identifies it exactly.

Verification tiers

Tier 0 — entered and published by a named officer of the issuing regulator. The portal that would create these is not built, so no record carries this rank yet. It is the top of the scale, not a rank on offer.
Tier 1 — published by the regulator and captured by us
Tier 2 — confirmed to us by the authority, but not published by them
Tier 3 — reported by a named credible secondary source
Tier 4 — reported and unconfirmed, published as an open question

Publication tiers and indexing

A record reaches search engines only with a date, a product name, and either a batch number or a stated reason. Everything else is kept out of them until that is filled in — but it is stored, searchable and readable here from the moment it lands. Nothing is held back; it is only kept from being found by somebody who would read a half-finished record as the whole story.

Feeds

Atom, so a search you build here is a subscription you can keep. Every filter the register offers works on the feed: build the view you want on the register, then put the same query string on /feed.xml.

/feed.xml — every published action, newest first
/feed.xml?cc=EG&action=recall — any register search, saved
/country/<slug>/feed.xml — one market
/manufacturer/<slug>/feed.xml — one company, across every market we read

A feed carries what the register lists, including records too incomplete to be given a public URL — the entry says which fields the authority did not state. It never carries a superseded record. Each entry’s identifier is its alert ID, which is permanent, so a record corrected or superseded is not re-delivered as something new.

Also here

Coverage matrix — what we monitor and where we are blind
Corrections log — timestamped and public
The register — every filtered view is a URL