Summary
- The RIPE NCC's transfer-statistics page defines the fields a row contains — Original Block, Transferred Block, country codes, Transfer Type, Date — and never defines a unit in which a reader should count transfers. That work is done in Rene Wilhelm's bylined RIPE Labs analyses of 6 May 2015 and 29 June 2017, where a "block" is fixed as a contiguous range that need not be a single CIDR prefix and a second unit, the transaction, is added on top.
- The 2017 analysis prints a counted pair, 125 blocks and 375,040 IP addresses, without restating the definition, and it is also the page that names a boundary the dataset page does not: transfers of legacy resources within the RIPE Registry are excluded "as the details of these are not necessarily shared with the RIPE NCC". A reader holding only the file holds columns; a reader who opens the article holds the counting rule.
The page that tells you what a row is
Read on 22 November 2026, the registry's transfer-statistics page opens by routing its own scope through policy: "According to section 4.0 of 'RIPE Resource Transfer Policies', the RIPE NCC publishes a list of all Internet number resources transferred according to the policy" (RIPE NCC transfer statistics). What follows is a field glossary. An "Original Block" is the "Initial IPv4 or IPv6 address block from which fully or partially address space was transferred". A "Transferred Block" is the "Subdivided prefix or partial block derived from the original block or resources transferred". "From Country Code" and "To Country Code" give the country where the holding and receiving organisations were legally registered when the transfer was processed, according to ISO 3166. "Transfer Type" separates policy transfers from "transfers due to changes to an organisation's business structure (such as a merger or acquisition)", and adds that transfers "are permanent, unless temporary transfers are specified". "Date" is the "Date on which transfers were processed".
Every entry describes a cell. Not one describes a total. The page does not say whether a reader should count rows, prefixes, addresses or counterparty pairs, and it does not say what one transfer is worth as a unit of analysis. That is not a defect in a glossary — a glossary is the wrong instrument for the job. It does mean the file, taken alone, cannot produce a number that means anything until somebody else supplies the rule.
What the 2015 paper adds, and how carefully
On 6 May 2015 Rene Wilhelm supplied it, and his sentence is a model of scoping: "In this context, a 'block' is a contiguous range of IP addresses which may or may not correspond to a single CIDR prefix" (Wilhelm, 2015). Two words do the work — "in this context" — because the definition is offered for that analysis, not promulgated as registry vocabulary. It never appears in the field glossary.
The same article refuses to stop at one unit. It also counts "transactions (the combined transfers between same parties on the same date)" and plots "the distribution of such transactions as a function of time". The reason is visible in the same text, which records that "multiple smaller blocks were transferred, to reach to the required size": when a buyer assembles a holding out of small pieces, block counts and party-level activity part company, and only one of the two numbers describes how many deals happened.
Then the pair, printed plainly: "Between 17 October 2012 and 1 May 2015 a total of 2,252 unique address blocks have been transferred", with a summed 14,526,720 IPv4 addresses, "equivalent to 86.6% of a /8".
Two checks, mine, run on that article's own inputs. 14,526,720 divided by 16,777,216 — one /8 — is 0.8659, which is the printed 86.6%. And 14,526,720 spread across 2,252 blocks averages 6,450.6 addresses per block, a quotient neither page prints; it is consistent with a definition that lets a block be something other than a single CIDR prefix, and it says nothing about any individual transfer.
The word that changes meaning
By 29 June 2017, when Wilhelm published "Trends in RIPE NCC Service Region IPv4 Transfers", the second unit had a different job. "In total we count 125 blocks/375,040 IP addresses that have been re-transferred by the original recipient" (Wilhelm, 2017) — a counted pair, printed with no definition attached. The inspected text does not restate the 2015 block sentence, and "transaction" now carries its corporate sense: "another transaction (e.g. one acquiring part or all of the other's business)".
So the same word covers a combined same-day transfer between two parties in 2015 and a corporate deal in 2017. The 2017 text also explains why transfer lengths are not clean powers of two: "Transfers with non-integer lengths correspond to cases where the total number of IP addresses transferred is not a power of two; this happens when a single transfer involves multiple differently-sized blocks."
Neither sentence is a contradiction. Read as a series, though, the two papers are not one instrument: a correction unit appears in one and not the next, and the vocabulary shifts under a noun that looks stable. A total from 2015 and a total from 2017 belong to the paper that printed each.
The boundary that never reached the file page
Here is the sentence I would put on the dataset page myself: "As requested by the RIPE community, we publish raw statistics on transfers conducted via the transfer policies. These statistics include the transfer of allocated PA, assigned PI as well as inter-RIR transfers. Transfers of legacy resources within the RIPE Registry are not included, as the details of these are not necessarily shared with the RIPE NCC."
That is coverage, stated with its own reason, and it is printed in the 2017 analysis. The excluded class is not trivial. Legacy resources are the oldest holdings in the registry, and transfers among them happen without the registry necessarily holding the details. Anyone adding up "all RIPE transfers" from the file is adding up policy transfers — allocated PA, assigned PI, inter-RIR — and nothing else. Wilhelm's sentence says so. The field table does not.
No concealment is being alleged, and nothing in the reviewed record supports reading it that way. The dataset page routes its scope to the policy, and a registry that never receives the details of a transfer cannot publish them. The point is narrower and harder to shrug off: the boundary exists, it carries the date of a published analysis, and it did not travel to the surface that hands out the file.
The 15 July 2025 seam, and the model the counting rule never got
One change did get a dated sentence on the page. "On 15 July 2025, country codes were added to transfer statistics", the registry announced (RIPE NCC notice, 15 July 2025). Within the service region "country codes of both parties involved in the transfer are visible"; for inter-RIR transfers "the country code of the party within the RIPE NCC service region is displayed". Country codes follow ISO 3166 and "reflect the information available at the time the transfer was processed", and the statistics now also "indicate whether a transfer was temporary".
The dataset page carries the consequence in its own words: "The Country Code is marked as undefined '-' for transfers processed before 15 July 2025."
That is how a schema change should be disclosed — a date, the affected rows, a marker in the data. It is also the sharpest contrast in the record. The country field has a printed cut-off, so a reader knows that a map built before July 2025 and one built after it are not the same instrument. The counting unit has no such date, because it was never on that page to begin with.
What a stranger can re-derive
A reader who opens four documents in the right order gets a coherent instrument: the glossary for the cells, the 2015 paper for the unit and its correction unit, the 2017 paper for a worked pair and the coverage boundary, the 2025 notice for the field cut-off. A reader who opens only the file gets columns and a download.
The row-level check stays open. The published machine-readable transfer files were not opened as documents for this piece, and the charts inside the 2015 analysis were not inspected, so whether a stranger can reproduce a printed total from the file, and how many rows carry the undefined country marker, are questions I am not answering here.
What can be said without the file is enough for a working rule: when a transfer total turns up in a slide, a valuation or a negotiation, ask which paper it came from. The unit is an author's decision, published on a stated date, scoped by two words to one article — and it is the most portable thing in the whole record.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






