Summary

  • Two successful downloads on 3 September contained identical bytes: 13,569 records, 13,546 distinct prefixes and 23 prefixes appearing twice each.
  • Nine pairs repeat exactly the same row. Fourteen differ in location fields, including four pairs with different country codes; the remainder must not be counted as country conflicts.
  • RFC 8805 treats repeated exact prefixes as an error but permits nested networks. Longest-prefix matching cannot choose between two identical /24s.
  • The evidence establishes a structural conflict, not an incorrect physical location or a harmed customer. Aggregation can become more accountable without making the registry an arbiter of geography.

At line 2,631 of LACNIC’s public geofeed, 170.51.97.0/24 is assigned the country code UY, the region UY-MO and the city Montevideo. At line 12,633, the same /24 appears with PY, PY-ASU and Asuncion. The country changes from Uruguay to Paraguay. The address range does not change at all.

The second entry is not a smaller network inside the first. Nor do the two rows carry an edit timestamp that establishes an update sequence. A consumer retaining the first occurrence would keep Uruguay. A consumer overwriting a previously seen key would keep Paraguay. Both could report that their answer came from the same file.

Those are illustrative loading policies, not findings about a particular vendor or LACNIC’s own software. They show why the distinction matters. A deterministic answer can emerge without any additional evidence about the world. The program has settled a tie; it has not located a network.

A narrow rule, and an important exception

RFC 8805 supplies a useful dividing line. Section 2.1.3 treats duplicate addresses or prefixes as an error and recommends logging them for administrative review. It separately permits nested prefixes and recommends the longest match. The location fields themselves are optional.

The difference is visible in the same feed. It contains both 177.126.48.0/22 and 177.126.48.0/24. A statement about the smaller /24 can refine a statement about its enclosing /22. For an address inside both ranges, prefix length provides a distinction. Two occurrences of 170.51.97.0/24 offer no such distinction.

This rules out an easy but misleading diagnosis: not every overlap is an error. A collector that rejected all nested entries would discard a useful part of the format. A collector that demanded a city from every contributor would invent a requirement and could encourage needless precision. The problem here is the repeated exact key, not the mere existence of a hierarchy or an empty field.

It also rules out an easy repair claim. Loading a file into a structure that allows only one value per key can eliminate visible duplication while concealing the choice that eliminated it. The destination becomes tidy. The evidence has not necessarily improved.

Twenty-three pairs are not one kind of problem

The first successful request began at 10:31:23 UTC on 3 September. Another began at 10:45:12 and completed at 10:45:48. Both returned the same 438,145 bytes. Two intervening attempts timed out. This is a record of two successful samples, not a service-availability study or proof that the condition persisted outside the observation window.

The file contains 13,569 nonblank, non-comment records: 10,782 IPv4 rows and 2,787 IPv6 rows. Parsing the address and prefix length, including normalization of IPv6 representations, produces 13,546 distinct networks. Twenty-three prefixes appear twice each, so 46 rows belong to duplicate pairs and there are 23 excess rows relative to one entry per prefix. A separate grouping of the literal prefix strings gives the same result.

The classification is more informative than the headline count:

Duplicate class Pairs What the sample establishes
Identical complete rows 9 The same declaration is emitted twice
Same country, different locality fields 10 Four pairs differ in region or city; six differ in available detail
Different country codes 4 Two country codes compete for the same exact prefix

For example, 200.123.12.0/24 appears twice consecutively with identical Lima, Peru data. That is not two independent confirmations of Lima. It is repetition in the output.

For 200.80.255.0/24, one row includes Argentina, its region code and Buenos Aires; the other supplies Argentina alone. The less detailed declaration is not automatically wrong. Equally, a fuller row is not automatically newer. More populated cells do not supply an edit history.

Another pair, 200.63.248.0/21, names Ibarra and Quito, with different region codes but the same country, Ecuador. That differs from the Uruguay–Paraguay case. The other three country-conflicting pairs—181.238.0.0/24, 181.239.32.0/24 and 186.182.0.0/24—each combine PY with AR, Paraguay with Argentina.

These are counts of records and prefixes. They are not counts of subscribers, organizations, autonomous systems or observed traffic paths. Nor do they establish that all other locations are accurate. The result is a small, reproducible set to investigate, not an accuracy score for the region.

The service deserves a defence on its own terms

LACNIC’s service description says that administrators of member organizations can voluntarily submit country, region and city information through MiLACNIC. The public CSV makes those declarations available in one place. The description was published in 2020; its retrieval in September 2026 is not evidence of a new launch.

The institution’s geolocation explanation gives the reason for a separate service. Whois contact geography may reflect an organization’s contractual details rather than where its addresses are used. Geofeeds let operators describe use and subdivide blocks accordingly. LACNIC also distinguishes its role from responsibility for commercial geolocation databases.

That is a sound, limited purpose. A common declaration channel need not measure every endpoint to be useful. It can make operational information easier to exchange while leaving its accuracy open to verification. Demanding that the registry certify every place would change the service’s nature and vastly enlarge its responsibilities.

But declining to certify geography does not prevent an aggregator from detecting an exact-key collision. It can distinguish repeated copies from competing statements. It can preserve the original evidence and expose a correction path. These are properties of publication, not claims to know where every device is.

The strongest defence therefore supports a narrow improvement. Keep the declaration service; make its handling of disagreement inspectable. The alternative is to pass the ambiguity to every consumer, each of which must invent a preference or reject more data than necessary.

What a cleaner export would—and would not—prove

Identical rows offer a relatively straightforward case for a canonical output that emits the declaration once while retaining its provenance. Differing rows require more care. A correction process might establish that one is stale, that a submission was duplicated or that a distributed deployment needs a different expression. None of those causes was established by this investigation.

The file does not identify who submitted the competing rows, when each was edited or which country is correct. No member account, confidential history, active geolocation measurement or commercial database was examined. A legitimate multi-location explanation remains possible; it would not, by itself, make two competing equal-key rows unambiguous.

Consumers can treat a location declaration as a hint, compare it with appropriate evidence and decide how much uncertainty their application can tolerate. What they should not do is silently convert their own tie-breaking rule into an apparent decision by the registry. If a country feeds a content-selection or access decision, that distinction may matter. No such downstream decision or customer harm was measured here.

A sensible export repair would keep unresolved disagreements distinguishable from single declarations and ordinary repetition. It would preserve valid unrelated records rather than treating the whole file as unusable. It would not promise that removing all duplicates makes every remaining location true.

The finding ends where the evidence ends: two identical samples contain 23 exact-prefix duplicate pairs, including four country-code conflicts. Correcting that bounded publication problem requires an account of what happened to the competing statements. It does not require an institution to decide geography by authority, or a researcher to pretend that the map has already been checked.