method
Verification framework, how claims are assessed
Every claim carries a status and two independent ratings, source independence and reproducibility. Numeric confidence scores are not used. The verifier is named and is not independent of the business.
| Version | 1.0 |
|---|---|
| Updated | 2026-09-11 |
Who runs this
This knowledge base is published by Navigator Multimedia about itself. The records were compiled and assessed by an AI agent operated by Navigator, named on every claim as the verifier. That agent is not a third party. Its role is to be transparent: to record what was checked, how, when, and against which source, so that anyone else can repeat the check. Where a record says a fact is independently supported, that refers to the source of the evidence, never to the agent.
Two axes, not one scale
A single "level 0 to 5" scale conflates two different questions. This framework separates them.
Source independence answers "who says so?"
| Value | Meaning |
|---|---|
| self-asserted | The business says it; no document or record. |
| first-party | A record created by the business (a dated blog post, an account export, a signed statement). |
| third-party | A record created by someone unrelated to the business (a review, a news article, a directory, a commercial index). |
| official | A government registry, a certifying body, a platform's own system, or a public protocol such as DNS. |
| multiple-independent | Two or more third-party or official sources that did not get the fact from each other. |
Reproducibility answers "can someone else check it?"
| Value | Meaning |
|---|---|
| not-reproducible | Take our word for it. |
| with-access | Someone with access to a private system or paid index could confirm it. |
| public-method | Anyone can repeat the check using the documented method (a URL, a DNS query, a registry number). |
| independently-reproduced | A named party outside the business has repeated the check and said so. |
A claim can be first-party and publicly reproducible (the hosting count: private total, public per-domain check), or third-party and not reproducible (a review that was later deleted). Recording both is more useful to a reader, human or machine, than one blended number.
Status vocabulary
| Status | Rule |
|---|---|
| verified | Official or multiple independent sources, and a reproducible method. Requires a verified_at date. |
| supported | Third-party evidence exists but is not independent of the business's own account, or a reproducible check exists but only for part of the claim. |
| partially-supported | Part of the claim is evidenced, part is not, and the record says which. |
| unverified | Only the business says so, or the evidence does not reach the claim. Not an accusation. |
| disputed | Sources conflict and the conflict is preserved, not resolved by choosing one. |
| rejected | The claim cannot be assessed as stated (undefined terms, no methodology) or is contradicted by evidence. |
| retired | Formerly made, no longer made; kept for history. |
How many sources make "multiple". At least two, from unrelated parties, each describing the fact from their own direct experience or record, each publicly reproducible. A source that repeats what the business says, or that describes a service menu rather than work received, does not count toward the total. This threshold is stated here because it was previously applied by judgement, which is not good enough for a record that asks to be trusted.
Validation rules enforce these: a claim cannot be "verified" on first-party evidence; a comparative or superlative claim must reference a methodology; every evidence record must name which part of the claim it supports and what its limitations are.
Evidence classes
first-party (the business's own records), third-party (unrelated sources), official (registries, platforms, protocols), derived (analysis performed on other evidence, always naming who performed it). A company assertion is never entered as third-party evidence, and sponsored content on a news site is classed as first-party.
Time
Time-sensitive claims carry valid_from, valid_until, expires_at and a review cycle. Observations (a review count, a hosted-site count) are appended with dates, never overwritten. A status change is appended to the claim's history with a note. Nothing is deleted; superseded records are marked retired.
Vocabulary
JSON-LD on every page uses schema.org where it fits and a small nav: vocabulary where it does not; see the vocabulary page.
Confidence scores
Not used. A number without a documented derivation is decoration.
What this framework cannot do
It cannot make the business's own agent independent. It cannot verify a fact for which no independent record exists (the 1993 founding is the standing example). It does not know whether AI systems weigh any of this; that is the experiment, described in the experiment protocol.