# Ireland (IE) — internal licensing notes

## Source research, 2026-08-29 — sellability + reproducibility audit

**Clean GO, and simpler than the prior FedEx-schema research anticipated.**
`fedex-agents-prototype/countries/IRL/research.json` (researched
2026-06-12, status REVIEW, never approved) already established the core
constraint: Ireland's Eircode system has two tiers, and only the coarser
one is openly reusable. What that research didn't check is that **GeoNames
itself already ships exactly the open tier**, as a single CC BY 4.0
source — no need for the gist + autoaddress.com combination that research
proposed and never got licence-checked.

### The constraint (established in the FedEx research, still holds)

The full per-address Eircode Address Database (ECAD, ~2.2M codes) is
operated by Eircode Ltd/Capita under a state contract and requires a paid
licence; bulk scraping (e.g. the public Eircode Finder) violates its
terms. **Only the 139 3-character routing keys** (e.g. `D02` → "Dublin 2")
are openly reusable — this is not a choice made for this build, it's the
actual boundary of what's legally available.

### The source — GeoNames IE (verified 2026-08-29, not just assumed from the readme note)

```
https://download.geonames.org/export/zip/IE.zip      # 3.7 KB, 139 rows
```

GeoNames' own `export/zip/readme.txt` states: *"For Ireland we have only
the first letters of the full postal codes (for copyright reasons)"* —
this is not a gap, it's GeoNames' own compliance with the exact same ECAD
licensing boundary. **Fetched and verified directly: exactly 139 rows, 139
distinct routing keys, all 4 provinces (Connacht/Leinster/Munster/Ulster)
present, every row has a coordinate.** `admin_name1`/`admin_code1`
(province, single-letter C/L/M/U) are already populated on every row.
`admin_name2`/`admin_name3` (county, post-town) are empty in this file —
GeoNames doesn't carry the routing-key→county join, matching the gap the
FedEx research already flagged (post-town→county is an approximate join
requiring a separate gazetteer, not present in any raw source found).

**This single GeoNames file supersedes the gist+autoaddress.com plan.**
The FedEx research's tier-5 sources
(`gist.github.com/ajoorabchi/...`, `autoaddress.com/articles/eircode-routing-keys/`)
were never licence-checked and don't need to be now — GeoNames already
ships the same 139-row routing-key list under an already-verified CC BY
4.0 licence, avoiding two unverified sources entirely.

**Licence: CC-BY-4.0**, verified against `export/zip/readme.txt` on
2026-08-29 (same file as every other GeoNames-sourced entry in this
catalogue) — Ireland's entry is the copyright-restricted row itself, not a
separate caveat blocking use of what IS included.

### County enrichment (admin_name2) — optional, not attempted this pass

Adding county would need a post-town → county join (Wikipedia's
`Counties_of_Ireland` page, CC BY-SA facts, same treatment as other
Wikipedia-derived admin facts in this catalogue) as sketched in the FedEx
research. Not required for a sellable product — the base 139-row file with
province already ships something real. Flag as a possible enrichment, not
a blocker.

## licence_status (proposed — needs human sign-off before shipping)

```json
{
  "channel": "postal-codes-r2",
  "source_url": "https://download.geonames.org/export/zip/IE.zip",
  "licence_id": "CC-BY-4.0",
  "collected_at": "2026-08-29",
  "decision": "compatible",
  "decided_by": "claude-session-2026-08-29",
  "decided_at": "2026-08-29"
}
```

## Proposed tier

**B ($49).** Single source normalised in place. Real derivation work
(admin_code1 already usable as-is, ISO 3166-2:IE prefix convention should
be checked — GeoNames' bare "C"/"L"/"M"/"U" vs `IE-C`/`IE-L`/`IE-M`/`IE-U`
per the FedEx research's own note) but no second source consolidated.
Genuinely thin (139 rows, only 2 of 13 columns beyond the base 5 populated)
but that thinness is the real shape of what's legally available for
Ireland, same as AL's bashki gap — not a build shortcoming.

## Verdict

**Build it.** No licensing obstacle. This is a smaller, cleaner build than
the FedEx-schema research anticipated — one already-verified source
instead of two unverified community ones. Not yet built for this
product — this note is the research/decision record only; no `data/`
folder exists yet for `ie`.

## Build — 2026-08-30 (local only, not published)

Built via `datasets/ie/scripts/build_postal_codes_product.py`. 139 rows, all
4 provinces, `admin_code1` upgraded to real ISO 3166-2:IE codes (verified
against Wikipedia's ISO_3166-2:IE page: C/L/M/U → IE-C/IE-L/IE-M/IE-U —
trivial, no ambiguity). `admin_name2`/`admin_name3` (county, post-town)
ship empty as disclosed — no county-enrichment join attempted, per the
research note's own scoping.

Frictionless validation: `valid: True`.

**Status: built and validated locally only.** Nothing published to R2,
nothing committed to git. `datasets/ie/README.md` and `datapackage.yml`
updated in place (tier B, `premium_source_key: ie/ie.zip`, real
`sources`/`licenses`, replacing the "Multiple sources" placeholder). Note:
`datapackage.yml` already had `is_premium: true` before this build — the
live R2 `ie.csv` (5,960 bytes per the old resource entry) was not
inspected in this pass; unclear what it currently contains or whether it
matches this rebuild.