DOCUMENTED CRYPTOASSET SERVICE ACCESS — PUBLIC REPRODUCIBILITY SUPPLEMENT
Author display: Vladislav Ch
Source / affiliation: Holder.io — https://holder.io/
Evidence window: 30–31 October 2026. Package prepared: 1 November 2026.
No additional reuse licence granted. See RIGHTS-AND-OMISSIONS.txt and
ESMA-REUSE-NOTICE.txt for source rights and attribution.

This reduced derivative preserves the accepted 1,200-query V2 dataset and the
frozen V1 comparison data, source metadata, 635 original authored evidence notes
(384 V2 + 251 V1 acquisition/review observations), numerical analysis, and the ten
new V2 figures in two languages. It is an offline computational supplement,
not a new evidence-collection round or an independent human review.

From this directory, using Python 3.10 or newer (standard library only):
  python3 analysis/statistics-v2/run.py --require-final
  python3 analysis/statistics-v2/verify.py
  python3 figures/validate_figures_v2.py
  python3 verify_public_package.py

The first command recalculates the statistical outputs. The second independently
recounts labels, checks 384 authored-note hashes, parses selected ESMA service
codes, evaluates the compression comparison, and repeats the calculation with
byte-identical outputs. The third checks supplied figure data/image hashes and
case-specific assertions. The last verifies all 635 note references, core
counts, omitted-capture boundaries, and publication-file hashes. A source-note
hash authenticates saved note bytes; it does not authenticate remote page bytes.

Optional figure rebuild requires numpy and matplotlib already installed:
  python3 figures/build_figures_v2.py
  python3 figures/validate_figures_v2.py
Recorded original versions: numpy 1.26.4, matplotlib 3.10.9, Python 3.10.5.
Graphics and fonts may differ across environments. Rebuild in a separate copy:
the public-file manifest binds the supplied release bytes, while a regenerated
figure manifest binds your new outputs. No Docker, account, service or network
request is needed for any command. No dependency installer is run automatically.

DATA AND INTERPRETATION
The fixed sample is 20 platform surfaces × 10 jurisdictions × 6 services. The
query profile is adult individual retail actual residence; onboarding concerns
a new client and the other services an existing client. Other geographic
attributes are unspecified. Country, entity, product and client restrictions
must be read together. Licence or registration evidence alone is not access.

V2 retains 1,198 labels without a new individual adjudication and changes only
Kraken derivatives for Germany and France from unknown to restricted. It has
384 source observations, 334 exact URLs, 329 claims, 200 document-layer rows,
118 regulatory observations and 102 conservative instrument groups. None of
these last two counts is a count of unique active exchange licences.

Status counts: available 12; unavailable 273; restricted 187; unknown 718;
inaccessible 10; conflicting 0. Country-only optimal repeated-label compression
has 185/1,200 disagreements in V2 (183/1,200 in V1). These measure saved-label
information loss, not legal accuracy or real account access. The 200 document
coverage rows are a different unit from 200 platform-country pairs.

The baseline data dictionary and schemas explain inherited fields. Canonical
records retain native/original records, provenance hashes, source IDs, dates,
retrieval failures and decision qualifications. Historical relative paths may
point to deliberately omitted acquisition/integration files; they are provenance
locators rather than claims that those files are distributed. Input/output
hashes inside original integration manifests refer to the frozen original V2.
Use provenance/file-manifest.json for this public derivative's file hashes.

BOUNDARIES
The package reproduces arithmetic from accepted data and the ten V2 figures.
It does not rerun original source acquisition, full canonical-data integration,
Holder's private corpus audit, the 22 earlier V1 figures, or DOCX/PDF layout.
Those boundaries are deliberate; no substitute private database or service is
used. The 14 selected factual ESMA rows reproduce 81 listed entity-service
pairs. The full-register size of 392 remains a recorded collection measurement
and cannot be reconstructed from this selected extract.

The original archive and source research tree were read only. The single local
screenshot pathname in V2 source metadata was replaced by an omission marker;
source URL, locator, dates, source ID, qualification and evidence-note hashes
remain. The figure builder's machine-specific dependency hook was removed.
No empirical label, legal assertion or numerical result was intentionally
changed. See provenance/transformation-manifest.json and verify the recorded
comparison against the original V2 results. Supplied images are the original
V2 bytes; inherited visual-QA claims are not republished as a fresh review.

Research preparation and this packaging used AI assistance. Automated internal
checks are not external peer review, independent factual collection, human
inter-rater reliability, legal advice, or confirmation of real transactions.
Researcher identifiers and funding claims are not part of this supplement.
