Methodology

The document audit

Before any of the research could be assessed, the files it was built from had to be audited. That audit found problems in the material itself, not only in what it contained.

Every file supplied was read. None was unreadable, and none was present-but-not-attempted.
The files were not the format they claimed. Seven documents supplied with a PDF extension were actually zip archives containing per-page images and text layers; standard PDF tools failed on all of them. Text was extracted from the text layers in page order.
One supplied file has no connection to the subject at all — a generic research memo on an unrelated topic. It is recorded as present-and-irrelevant so that no future work spends effort on it.
The provenance chain has two branches that never met: two parallel investigations of the same subject, run from the same missing brief, that did not read each other. Where they cover the same fixtures they disagree, and this site renders that disagreement rather than merging it.
One corrections addendum is sourced overwhelmingly from Wikipedia: eleven of the twelve sources it lists. Wikipedia is never evidence on this site and never counts as independent confirmation, so that addendum’s sourcing is treated accordingly.

The full audit, file by file, is published at the document audit.

What the record is missing

This site publishes 128 records. 109 of them carry their original full research entry. The rest do not, and this is the most significant limitation of the whole record.

Partial records — the checksum audit of 18 July 2026

Three files supplied as original research dossiers turned out, on checksum, to be byte-identical duplicates of the correction addenda that accompany them. They are not dossiers. One phase’s base dossier is not in the corpus at all.

23 numbered research entries have no surviving original dossier text as a result. 19 of them are published here, each carrying a notice on the record itself saying so, and each reconstructed only from surviving correction and synthesis material. The remaining 4 are not published at all: no date survives for them anywhere, and inventing one was not permissible. They are listed individually on research status.

These 19 records are never described as fully audited or full-field. Where a field of theirs has no surviving content, the field is hidden rather than filled — an empty evidence panel would suggest the research looked and found nothing, when in fact the material that would have answered it is gone.

The affected entries are not evenly distributed. They fall entirely within the earliest phase of the record, which includes the childhood growth-hormone diagnosis, its prescribed treatment and the financing of that treatment — editorially sensitive ground where the site is working from correction-table fragments rather than the full entry.

What the research never established at all — as opposed to what it recorded incompletely — is a separate matter, set out on research status.

How claims are classified

Every claim and every timeline entry carries exactly one of seven labels. The label is a statement about the evidence behind the claim, not about how serious the claim is or how likely it seems.

Classification	What it means
Confirmed fact	Supported by a primary record, adjudicated finding or strong independent evidence
Credible reported allegation	Credible reporting or a formal allegation, not adjudicated as fact
Disputed interpretation	Evidence or qualified experts support more than one reading
Circumstantial connection	A documented proximity or overlap that does not prove causation or misconduct
Analytical inference	A reasoned interpretation derived from documented facts
Unsupported or false	Contradicted by evidence or lacking a credible basis after review
Not independently verified	Traceable claim with insufficient independent confirmation

A record can also carry secondary statuses alongside its classification — official finding, developing, corrected, superseded, and the sourcing states described below. These qualify a classification; they never replace one. A record whose position is still moving carries its primary label for what is established, plus the date on which that position was last accurate.

Officiating decisions carry one further, entirely separate axis recording which side a decision fell towards. It is not an evidence classification and is never mapped onto one: that a decision benefited a team is not evidence of anything about why it was made.

Every classification is readable without colour. Each renders a text label, an icon and a distinct border treatment, so the seven states are told apart with colour removed entirely. Colour accents the icon and border only, and the label text always uses the page’s foreground colour so its legibility never depends on the accent.

Every term used on this site is defined in the glossary.

Reconciling five vocabularies

The seven labels above are not the labels the underlying research was written in — at least, not all of it. This is published because a reader checking this site against the research will otherwise find labels that do not match, with no way to tell a relabelling from a disagreement.

Five classification vocabularies were in use across the eight research phases: a ten-value list, the governing seven-value list, one phase’s own ad-hoc set, a third list supplied by another phase’s brief, and a fifth used by the two stage documents.

The governing seven-value vocabulary was correctly established as primary early on, with an instruction that every later phase be notified — precisely so that a vocabulary change would not later be mistaken for a factual conflict.

The notification was to reach six later phases. It reached two of them, and explicitly did not reach three: one of those was still applying the superseded ten-value list; one was applying a third list its own brief supplied, and flagged the divergence itself; one never adopted any of the project’s vocabularies. What happened with the sixth is not recorded.

The mapping below is the reconciliation. It is published rather than applied silently, because a reader checking this site against the underlying research will otherwise find labels that do not match and have no way to tell a relabelling from a disagreement.

As written in the research	As classified here
VERIFIED FACT · CONFIRMED · INDEPENDENTLY-CORROBORATED	Confirmed fact
OFFICIAL-FINDING	Confirmed factPlus the secondary status “official finding”, which is preserved rather than absorbed.
REPORTED ALLEGATION · CHARGED-OR-FORMALLY-ALLEGED	Credible reported allegation
UNVERIFIED · NULL · REPORTED-SINGLE-SOURCE · NOT-FOUND · “Not established on retrieved record”	Not independently verified
ANALYSIS · “Commercial inference”	Analytical inference
ATTRIBUTED RUMOUR	Not independently verified, or Unsupported or falseDecided case by case. There is no automatic mapping, because the two outcomes are materially different and the source vocabulary did not distinguish them.
CONTRADICTED · NOT-SUBSTANTIATED · “Unsupported allegation”	Unsupported or false
“Disputed interpretation”	Disputed interpretation
DEVELOPING	No equivalent — kept as a secondary statusIt could not be mapped onto any primary label without loss. A developing record carries its primary classification for what is established, plus the date on which that position was last accurate.

How sources are counted

A claim’s strength is not the number of places it appears. Most of the counting rules below exist to stop a large reference count from reading as strong support when every reference traces to one report.

Repetitions collapse into their origin: Syndicated wire copy, aggregation, translations and outlets citing an original report are all the same source. Each repetition is linked to the report it derives from, and the whole chain counts once. Ten outlets echoing one wire story are one independent source, not ten.
A chain is not independence: A long repetition chain is not a strong claim — usually the reverse. The site shows the chain and the independent count separately so the difference is visible.
Independence is a property of sourcing, not of outlets: Two accounts by co-participants in the same event corroborate each other without being independent of each other. The record contains such a case, and counts it as two sources with a note that they are not mutually independent.
Wikipedia is excluded from every count: It may appear in internal discovery notes. It is never documentary evidence and never independent confirmation, and a Wikipedia-derived reference is excluded from the independent count regardless of how it is flagged in the data.
Both numbers are always shown: Independent sources and total references appear together. Showing only the larger of the two would misrepresent exactly the cases this rule exists to catch.

Every source behind the record, with its category, access state and independence relationship, is in the source library.

What this site will not do

Score anything

There is no numeric confidence score, bias score or corruption score anywhere on this site, and there will not be one. A number of that kind would imply a precision the underlying evidence does not support, and would invite comparison between records whose evidentiary positions are not commensurable. The classification, together with the record’s own sources and their independence, is the confidence system.

Call an absence of evidence a disproof

Where the record does not support a claim, this site says so in those terms — not supported by evidence in this record. It does not say “disproven”, “debunked” or “false”. Failing to find evidence for something is not the same as establishing that it did not happen, and the site does not blur the two in either direction.

The exact wording used is: “Not supported by evidence in this record”.

Merge separate matters

The material in this record is multiple separate issues, not one coordinated story. Distinct legal proceedings are kept in distinct chapters even where they share people, and where two records conflict the conflict is shown rather than resolved into a single authoritative version. Three incompatible accounts of the same fixture are published as three accounts.

Translate a legal term into a different legal system

Spanish and Argentine procedural terms are shown in their original wording with an explanation, never swapped for the nearest English word. The nearest English words name different things, and using them would assert a procedural equivalence that does not exist. These terms are set out in the glossary.

Correct itself quietly

Corrections are append-only and visible. The superseded wording stays on the record beside the corrected wording, and the full register is published at corrections.