Skip to main content
SemanticRisk • Public Interpretation Assurance Record
upwind.io
A 14-day evidence record showing what was captured from the website, how claims were interpreted, and whether that interpretation changed. This is not merely an AI visibility or mention count.
How to read this page: The score is generated from a deterministic weighted evaluation, so fixed score intervals and shared values can reflect the same evaluation outcomes. Score movement describes changes in that aggregate score. The timeline below shows one row per scan observation, while the company explorer calculates material claim drift across stored model comparisons. A single scored scan can produce separate Claude, Gemini, GPT and Grok comparisons. A held observation carries forward the previous valid score when normalized captured content is unchanged.
Why can claims move while the score stays the same? Aggregate scores summarize several evaluation dimensions. Claim-level differences can still occur in attribution, specificity, scope, classification, quantities, references, or wording without changing the total score. Not every extracted difference is material. See how SemanticRisk classifies differences →
1 • Public evidence

What the website supplied

Captured text, accessible page content, structured signals, response conditions, and normalized content provide the evidence base for this record.
2 • Extracted interpretation

What AI understood

SemanticRisk records extracted company, capability, audience, industry, scope, attribution, stance, and confidence claims where available.
3 • Interpretation drift

What changed — and why

Content change, extraction drift, stance drift, confidence drift, crawl regression, and unchanged-content variation are separated rather than treated as one visibility metric.
The SemanticRisk difference

AI visibility tells you whether a company appears. Interpretation assurance shows what was understood and the evidence behind it.

View sample reviewed report
Current score ⓘ
56.15 /100
Middle-scoring range
Latest score change ⓘ
No score change
Previous score: 56.15 /100
14-day scan observations with material flag ⓘ
No displayed scan observations
14-day score movement ⓘ
No newly scored comparisons

Latest interpretation comparison

What changed, how significant it appears, and the recorded likely driver.
No latest material flag
Classification ⓘ
Mixed
Likely driver ⓘ
Extraction Variation
Claims added / removed ⓘ
0 / 2
Stance / confidence changes ⓘ
0 / 0
A changed extraction does not automatically mean the company changed its position. It can reflect source wording, page structure, captured content, or interpretation variation. Review the evidence preview and content-capture status together. See the difference taxonomy →

Crawl and capture health

These signals describe whether enough usable website content was captured to support the interpretation record.
Preflight status ⓘ
Ok
TTFB ⓘ
406 ms
Estimated readable text ⓘ
14,639
Normalized content ⓘ
14,751

14-day score and interpretation record

Bars show the effective score out of 100. A carried-forward marker identifies observations where unchanged normalized content retained the previous valid score. Table rows show score movement separately from claim-level classification.
0 records
↔ indicates a carried-forward score: normalized captured content was unchanged and no new scored evaluation replaced the previous valid result.
Scan timeScore /100 ⓘScore change ⓘClaim classification ⓘSeverity ⓘStatusEvidence
No scan records were stored in this public window.
Turn this public record into a decision

Get a human-reviewed interpretation assessment for your own company.

The US$299 AI Interpretation Review adds a written assessment of current interpretation, extracted claims and evidence, model differences, capture conditions, prioritized recommendations, and a 30-minute findings call. Monitoring can then track whether the pattern changes.
Is this your company?

Get a human-reviewed AI Interpretation Review.

The public record shows monitoring evidence. The paid review adds model comparison, extracted claims, likely evidence drivers, prioritized recommendations, and a 30-minute findings call.
Diagnostic boundary: SemanticRisk reports observations produced by a controlled monitoring pipeline. Model interpretations and extracted claims are not independently verified company facts. See the canonical monitoring definition, see how semantic differences are classified, or read the score guide.