How we calculate impact

The crawler, PageSpeed strategies, hosting and discovery collectors operate independently. A blocked crawler does not invalidate usable PageSpeed evidence. Missing evidence stays unavailable; it is never silently replaced with zero. Laboratory Lighthouse results and real-user CrUX observations remain separate.

Assessment evidence and calculations are retained with integrity hashes. A hash helps detect changes; it does not prove that the original measurement was accurate. Public explanations do not require exposing private evidence, credentials or complete captured page content.

Impact is normalized to 1,000,000 pageviews for comparison. This is a scenario scale, not a claim about actual traffic. New annual portfolio and country scenarios assume 50 000 pageviews per website per year, shared equally across its selected canonical pages. For each page: annual impact = impact per million × 0.05 ÷ selected page count for that website. Missing evidence retains its unvalued traffic share. This is assumed traffic, not a measured global average; the scanned sample is not proven representative of every page on a website. Reference statistics and historical snapshots retain their recorded scale. Only pages with both current and supported avoidable values enter a paired comparison.

Mobile Largest Contentful Paint (LCP) is a loading-time indicator, not a measurement of every user's experience.

Total seconds = LCP milliseconds ÷ 1,000 × pageviews

Avoidable delay seconds = max(LCP seconds − 2.5, 0) × pageviews

The improvement reference is a modelled comparison. Total Blocking Time is not added to LCP. A new assessment is needed to confirm actual improvement.

GB = transferred bytes ÷ 1,000,000,000 × pageviews

Avoidable transfer uses supported byte-denominated savings evidence, bounded by total transfer. Milliseconds are never converted into bytes. Missing savings evidence is different from a measured zero opportunity.

The HIC-200 registry uses Sustainable Web Design Model v4 segment constants for device, network and data-centre operational and embodied energy. These are modelled impacts, not meter readings.

Segment energy = transferred GB × registered segment kWh/GB

Segment CO₂e = segment kWh × registered carbon intensity

Unit conversions and segment sums follow the recorded calculation. Hosting evidence does not automatically create a renewable-hosting discount. See the model documentation; the report retains the version and constants actually used.

Recorded Lighthouse main-thread milliseconds can be converted to seconds and scaled by pageviews. This is runtime under the recorded test configuration, not watts, GPU usage or billable CPU-hours. The HIC-200 blocking-time proxy is a separate metric.

AI content size uses retained UTF-8 text bytes. Readability evidence does not establish inference cost, guarantee citations or predict AI ranking. Unsupported compute and AI monetary values remain unavailable.

Scenario value = supported physical quantity × explicitly recorded valuation coefficient

Time values convert seconds to hours; carbon values convert kilograms to tonnes. Every coefficient needs a source, geography, currency, price year and version. Purchasing-power dollars are not interchangeable with nominal USD.

Global references and operator-country pricing scenarios answer different questions. A website's declared operator country does not establish its visitors' location. Country-based valuation must disclose that assumption.

Data subscription-equivalent value is not guaranteed bill savings. Time value is not necessarily saved wages. Carbon shadow value is not cash income. Retail data fees may already include network electricity costs, and compute energy may overlap the environmental model. Combined totals require an explicit non-overlap policy and compatible price bases. Partial data values are not complete economic totals.

The Index selects one current valid assessment per canonical page. Page-level statistics weight pages equally; domain-normalized statistics first take each domain's median, then summarize those medians. Each metric discloses its available sample count. Percentiles use the recorded interpolation method.

The sample describes pages observed by HIC, not the entire web. Immutable snapshots retain observation references, source versions and hashes. Historical methodology versions must not be silently mixed.

For a questioned result, retain its report URL, assessment or snapshot identifier, methodology version and selected scenario. The report's equations, sources, evidence coverage and calculation references are the starting point for review.