Methodology / Public standard

What we know. What we infer. Where we stop.

A measurement is useful only when its method and limits remain attached.

Version 1.0 · 20 August 2026

01 / Claim discipline

Four labels. Four different promises.

York Studio does not use correlation language to imply causation. The label determines what the reader may reasonably conclude.

01

Observed

The recorded answer, citation, claim, or recommendation existed under the stated conditions.

What did the system return?
02

Inferred

An analysis estimated a pattern, relationship, category, sentiment, or likely explanation from observations.

What does the evidence suggest?
03

Attributed

An experimental or causal design supports the claim that an intervention contributed to a measured change.

What changed because of the intervention?
04

Opinion

A clearly labelled judgement or recommendation that goes beyond what the evidence alone establishes.

What does York Studio think should happen?

02 / Evidence record

Every number opens back into an observation.

A chart is a view of the evidence—not a replacement for it.

Observation / ys_obs_02418Comparable
01Prompt + prompt version
02Full answer or permitted evidence
03Citations + source URLs
04Surface + collection method
05Market + language
06Timestamp + run identifier
07Model or product identifier
08Request settings + status
09Extraction version + confidence
10Analyst correction history
Integrity hash retainedMethod: labelled

03 / Comparability

Do not blend unlike surfaces quietly.

An official API, a search-grounded API, a licensed consumer feed, and a manual capture can all answer useful questions. They are not assumed to reproduce one another.

Every series retains its collection method. A cross-surface comparison states what was held constant, what could not be held constant, and why the comparison is still useful.

Coverage gaps stay visible.

Failed runs, missing markets, unavailable metadata, and provider changes are reported. They are not silently substituted or removed from the denominator.

04 / Operating policies

The unglamorous parts of trust.

Research quality also depends on corrections, versioning, independence, and knowing when not to publish a convenient conclusion.

01

Limitations are part of the result

Every study and method note includes the boundary conditions a reasonable reader needs.

02

Negative and inconclusive results remain publishable

Commercial inconvenience is not a reason to hide a correctly designed result.

03

Material changes receive versions

Methods, prompt sets, extractors, and definitions are versioned so a historical comparison keeps its context.

04

Corrections leave a trace

Material factual or analytical errors are corrected visibly, with the reason and date.

05

Commercial relationships are disclosed

Sponsors, design partners, customers, or products relevant to a piece of research are stated alongside it.

Corrections

Good research can be challenged.

If you find a factual, methodological, or interpretive error, send the exact claim and the evidence that contradicts it.

Submit a correction