Press Room / TDM case
Live case · data through 24 September 2026TDM reservation, machine access and a governance experiment in public
Since 19 July 2026 ENTIA has been measuring what automated agents request, what is served, which rights representation is active at that moment, and how counterparties respond when a machine-readable reservation is brought to their attention. This page now reports the system, the evidence and the chronology without ranking or publicly scoring individual operators.
3,410,966third-party machine requests · 19 Jul–24 Sep
1,099,503distinct ENTIA objects requested
1,481,104distinct URLs requested
55.9 GBorigin egress measured
The project has moved from asking “who crawled?” to a harder question: what can be proved about machine access, which policy representation was actually served, and how should rights governance work when one operator may use different bot families for search, retrieval, grounding, training or user-directed access?
Measurement contract · data through 24 Sep 2026
What the telemetry actually says
Forensic boundary. This proves automated access, transfer, breadth and repetition. It does not by itself prove training, embeddings, storage, indexing, commercial exploitation or intent. Those limits are part of the measurement contract, not a disclaimer added afterwards.
Frozen historical baseline · signed 2 Sep
The original record remains intact
2,279,997third-party machine requests · 19 Jul–1 Sep
170agent families in the frozen family census
1,431,698record-level entity requests served
1,158,743distinct entity records served
Policy history · one legal scope, several evidence layers
From V1 to V3.1
V1Article 4(3) reservation goes live while broad positive permissions for public search, citation and fresh retrieval remain.
V2The reservation is formalised into objective technical-use lanes and reserved-use classes, while the broad search/retrieval permissions remain.
V3The prior broad positive permissions for automated search, indexing, linking, citation, fresh retrieval and grounding are prospectively withdrawn. The Article 4(3) reservation remains.
V3.1 evidence boundary
robots.txt becomes a canonical first-contact representation bound to V3 by exact bytes, SHA-256 and manifest. V3.1 is not a fourth policy and creates no new legal T0.Why V3.1 exists. Policy effective time and observed serving time are different facts. V3.1 makes it possible to associate an observed request with the exact first-contact representation served, without claiming remote parsing, understanding, acknowledgement, compliance, training or downstream use.
European Commission · documented chronology
What happened institutionally
Case submittedENTIA presents the machine-readable reservation and the broader access-governance problem to the AI Office and DG CNECT copyright unit.
Commission replyArticle 4(3) prescribes no single technical method; the Commission does not validate one protocol; the Code process remains open.
Workshop requestENTIA asks to participate in the Measure 1.3 copyright workshops.
Written invitationThe workshops are closed; the Commission formally invites a written contribution for consideration.
Contribution and updateThe contribution is submitted on 24 August and updated on 2 September with the first consolidated evidence package.
Exact scope. The Commission has not assessed or endorsed ENTIA's implementation and has taken no position on any operator. The documented fact is the invitation to contribute to an open process.
Counterparty communications · evidence preserved, names removed here
The purpose of the outreach
The original outreach covered twelve counterparties across model providers, search operators and crawler services. The correspondence has produced different outcomes: substantive replies, routing replies, remediation actions, commercial or legal channels, and non-responses. Those records are preserved, but this page no longer turns them into a public scoreboard.
Question 1 · identification
How does an operator identify a machine-readable Article 4(3) reservation when access is performed by several bot or crawler families with different declared purposes?
Question 2 · purpose
Which declared purposes correspond to search, indexing, user-directed retrieval, grounding, evaluation, dataset construction or model training, and where is that mapping technically documented?
Question 3 · evidence
What can be established from request telemetry and policy delivery, and what remains internal to the operator and therefore cannot be inferred from outside?
Question 4 · licensing
If a use is reserved, what machine-readable path should carry the request from discovery of the reservation to rights resolution, licence scope and auditable permission?
Architecture question
Why robots.txt is no longer enough on its own
Discovery is not authority
robots.txt is useful as a first-contact discovery surface, but it was designed for crawler exclusion, not for authenticated rights state, licensing, temporal evidence or purpose-specific permissions.
One operator, several purposes
The same organisation may publish different agent families for search, user requests, model-related uses or other functions. A binary crawl instruction cannot express the full purpose graph.
Integrity and time matter
If a policy changes, is cached or is modified, the relevant evidence is not merely what the file says today. It is what representation was actually served to a particular request at a particular time.
Rights resolution needs a layer above
The experiment therefore separates first contact, authoritative policy, enforcement and evidence. That is the architectural direction behind V3.1.
What the evidence supports
- That automated requests arrived at recorded times under specific declared agent names and network conditions.
- What resource was requested and what status was served.
- Aggregate breadth, repetition, transfer and distinct-resource counts under a declared measurement contract.
- Which policy version was legally effective and, from the V3.1 boundary, which canonical first-contact representation can be associated with an observed request when direct evidence exists.
- What was communicated to counterparties and what positions were returned through preserved channels.
What the evidence does not support
- That any access was used for training, embeddings, storage, indexing or commercial exploitation.
- That a User-Agent string alone establishes legal operator identity.
- That access forms a contract or creates an automatic fee, debt, damage or infringement.
- That V3.1 proves remote parsing, understanding, acknowledgement or compliance.
- That the European Commission has validated ENTIA's implementation.
19 Jul to 25 Sep 2026
Chronology
V1 effective. Machine-readable reservation enters production and per-request evidence begins.
The case is presented to the AI Office and DG CNECT copyright unit.
Commission reply: no single prescribed method and no protocol validation.
Written contribution invited. ENTIA submits it the same day.
Twelve counterparty evidence packages and licence proposals are dispatched.
The first consolidated record is signed and the Commission update is submitted.
V2 effective. Technical-use lanes and reserved-use classes are formalised.
V3 effective. Broad prior positive permissions for automated search/retrieval uses are prospectively withdrawn.
V3.1 evidence boundary. The canonical
robots.txt first-contact representation is bound to V3 by exact bytes, hash and manifest.Live telemetry closes the day with 3,410,966 third-party requests across the post-reservation window under the current measurement contract.
This page is rewritten. Named operator scoreboards are removed; the case is presented as an evidence, governance and methodology record.
Primary public sources
Recompute before interpreting
- entia.systems/machine-crawl-censusLive telemetry driven by the public measurement contract, plus the frozen signed family census.
- entia.systems/access-report/acta-consolidada.pdfFrozen consolidated record for 19 July to 1 September, signed on 2 September.
- entia.systems/.well-known/tdm-policy-history.jsonCanonical V1 → V2 → V3 policy chronology and V3.1 evidentiary boundary.
- entia.systems/.well-known/tdm-policy.jsonCurrent machine-readable policy.
- entia.systems/eu-submissionContribution submitted to the European Commission and its supporting material.
- European Commission · GPAI Code of PracticeOfficial copyright chapter and related public materials.
“The objective is not to infer what happens inside a model provider. It is to make the external boundary measurable: access, policy, evidence, declared purpose and rights resolution.”Fernando Vilches · Founder & Architect of ENTIA