Research
Evidence register
Every result we publish comes with the method that produced it and the conditions it holds under and whether another party has reproduced it.
Measured
14 claims are recorded. Each states its object, scope, evidence class and the check that settles it.
| Claim | Object | Scope | Evidence class | Method / evidence | Recorded |
|---|---|---|---|---|---|
| Measurement reaches the analytics property from the platform and not from the visitor's browser: a page read is forwarded only when the visitor allowed analytics, it carries a pseudonymous reference and the page it was recorded on, and it carries no address, session or user agent. | Implementation property | The navigation route and the measurement forwarder of the release under test, for a visitor with no choice, a refusal and a grant. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 17 November 2026 |
| An installation made by one release is upgraded in place by the next, and the upgrade leaves everything the earlier release recorded intact and verifiable. | Implementation property | An upgrade from the newest installed release to the release under test. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 16 November 2026 |
| Every response, including an error and an unknown address, carries a content policy that permits scripts, styles and images only from the same origin, forbids inline scripts and framing, requires transport security and prevents content type guessing; the contact page alone also permits image tiles from tile.openstreetmap.org, for its map. | Implementation property | Every response of the release under test, including error and unknown address responses, and the contact page, which alone names the map tile host. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 13 November 2026 |
| A usage event is stored only when the latest analytics choice of the visitor is granted, and it carries a campaign only when their latest marketing choice is granted too; an event from a visitor who refused either is not stored. | Implementation property | The navigation event route of the release under test, for a visitor with no choice, a refusal, an analytics grant and an analytics grant with and without marketing. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 13 November 2026 |
| A repeated request that carries the same idempotency key is measured once, and an operation that fails after it was measured is cancelled by a reversal and not charged. | Implementation property | Metered requests carrying an idempotency key, including a retry and a failure after measurement. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 9 November 2026 |
| An API key with a monthly budget is refused with status 402 when a request would take the month past it, nothing is performed or charged, and requests made at the same time cannot together pass the budget. | Implementation property | The metered operations of an API key with a monthly budget, including concurrent requests at the threshold. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 9 November 2026 |
| Erasing an account removes its saved documents and projects, and keeps its metered usage as a financial record that holds no email address and no content. | Implementation property | An account erasure through the platform's data subject request path. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 9 November 2026 |
| A record that is a draft, archived or not public is not served at its address and is not listed in the sitemap. | Implementation property | Every address the release under test serves and the sitemap it publishes, for a record that is a draft, archived or not public. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 24 October 2026 |
| The role the application connects as can post a journal entry but cannot write a ledger balance; balances are derived from posted entries. | Implementation property | The database role the application connects as; balances are derived from posted journal entries. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 24 October 2026 |
| The role the application connects as cannot change or delete an entry in the evidence chain or the audit record. | Implementation property | The database role the application connects as, for every entry in the evidence chain and audit record. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 24 October 2026 |
| An operation is refused when the caller holds no current grant for that operation on that resource. | Implementation property | Every operation the platform authorises, for a caller with no current grant on that resource. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 24 October 2026 |
| A passkey sign in assertion can be used once, and one assertion cannot create two sessions when it is presented concurrently. | Implementation property | Passkey sign in on the release under test, including concurrent presentation of one assertion. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 24 October 2026 |
| The metrics surface refuses an anonymous caller and a signed in caller without the metrics grant, and its labels carry no identifier of an account, a session or a request path. | Implementation property | The metrics surface of the release under test, for anonymous and signed in callers without the metrics grant. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 24 October 2026 |
| A changed or removed entry in the evidence chain is detected when the chain is verified, including a truncated chain and a consistently rewritten chain when an anchor has been recorded. | Implementation property | The chain verification of the release under test, including a truncated chain and, where an anchor exists, a consistently rewritten one. | Measured | Automated and repeatable: the release's own verification suite runs this check and a failure blocks the release. | 24 October 2026 |
How the register is kept
Every recorded action on this platform forms a numbered chain: each entry carries the fingerprint of the one before it. Nothing is rewritten; a change is a new entry naming its predecessor. The register states its position plainly:
- Register and chain
- 14 claims recorded. The platform's audit record is kept as one hash chain per account; the most recent anchoring attests position 18 of its chain, placed 17 November 2026.
How a claim is recorded and graded
We record five axes separately for each claim because one label cannot carry all of them. The object says what kind of thing is claimed: a definition, theorem, model, hypothesis, observation, implementation property or objective. The method is what settles it. The scope is the condition and boundary within which the recorded check established it. The provenance is who authored it and who checked it and whether an independent replication exists. The lifecycle is its status in the record.
The evidence class says what kind of evidence is attached and it is not a rank because we judge a derivation by its definitions and inference and an implementation property by the check that exercised it and a commercial capability by the deployed workflow. None of those becomes stronger or weaker because a third party has or has not signed it.
- Measured
- A value or behaviour observed in a stated check, with its conditions.
- Reproducible
- A result that another person can obtain by following the stated method.
- Reported
- A result reported by a named party whose method is held by that party.
- Hypothesis
- A proposition recorded with the test that would settle it and the test has not been run.
The method column names the check that our harness recorded for that claim in its own words and that is the test or script or command that ran. Where no verification row exists the column says so instead of filling the gap with a category. A superseded claim keeps its entry and names its successor and a withdrawn claim stays readable.
Technical evaluation
Follow the evidence
Evaluate a system through its specification, verification method and recorded result. Each register entry identifies the type of evidence and the method used.
For JSON, request this address with Accept: application/json. The same filters and access checks apply.
Security and assurance · Limitations · Research publications · Discuss your evaluation
Last reviewed 13 November 2026.