Internet Quality Barometer (IQB) Overview

• Internet Quality Barometer (IQB) is a composite metric that defines Internet quality through user-centric use cases and explicit network performance thresholds.
• It employs a multi-tiered methodology that maps raw speed-test measurements to use-case requirements using expert-defined weights and percentile-based thresholds.
• The framework aggregates performance scores into a normalized index, enabling cross-region benchmarking and informed policy analysis.

Searching arXiv for papers on the Internet Quality Barometer and closely related Internet-quality/QoE frameworks. The Internet Quality Barometer (IQB) is a user-centric framework for assessing Internet quality beyond “speed” alone. It defines Internet quality in terms of representative use cases, maps network requirements to those use cases through weights and quality thresholds, and leverages publicly available Internet performance datasets to calculate the IQB Score, a composite metric intended to reflect the quality of Internet experience. In its implemented form, IQB transforms raw speed-test measurements into use-case-specific scores and then aggregates them into a composite index ranging from 0 to 1, so that Internet quality can be compared across locations and over time (Ohlsen et al., 23 Sep 2025, Sermpezis et al., 9 Jun 2026).

1. Concept and rationale

IQB was proposed in response to the limitation of equating Internet quality with throughput or bandwidth alone. The framework starts from the observation that contemporary Internet use spans web browsing, audio streaming, video streaming, online backup, video conferencing, and gaming, and that these activities depend on different combinations of download throughput, upload throughput, latency, and packet loss. Throughput matters more for large downloads or backups; latency matters more for interactive applications such as gaming or conferencing; packet loss can degrade real-time communication and gameplay (Ohlsen et al., 23 Sep 2025).

This formulation replaces the question “What is the speed?” with the question “Can the connection support the user’s intended activity at a useful quality level?” The resulting IQB Score is therefore not a direct measurement of a single network primitive. It is a composite index intended to summarize whether measured network conditions satisfy the requirements of representative activities. The 2026 implementation paper makes this operational by mapping raw speed-test measurements to network requirements, mapping those requirements to representative use cases, and aggregating the resulting use-case scores into a single score in the unit interval (Sermpezis et al., 9 Jun 2026).

The framework is explicitly presented as redefining Internet quality beyond raw speed, and it is described as being inspired by composite indices such as a credit score or Nutri-Score. This does not remove the underlying multidimensionality; rather, it compresses it into a transparent summary metric built from explicit factors, thresholds, and weights (Ohlsen et al., 23 Sep 2025).

2. Tiered structure and use-case model

The framework is organized as a three-tier system in which values are aggregated upward from datasets to requirements to use cases to the final score (Ohlsen et al., 23 Sep 2025).

Tier	Content	Role
Use cases	Web Browsing; Streaming Audio; Streaming Video; Online Backup; Video Conferencing; Gaming	User-centric definition of quality
Network requirements	Download throughput; Upload throughput; Latency; Packet loss	Technical mapping from activities to metrics
Datasets	M-Lab NDT; Cloudflare Radar/Aggregated Internet Performance Metrics; Ookla Global Fixed and Mobile Network Performance Maps	Measurement basis for scoring

The first tier defines Internet quality through six representative use cases: Web Browsing, Streaming Audio, Streaming Video, Online Backup, Video Conferencing, and Gaming. The second tier associates each use case with network requirements. The third tier provides the measurements from which those requirements are evaluated (Ohlsen et al., 23 Sep 2025).

For each use case and each requirement, IQB uses both quality thresholds and importance weights. The thresholds distinguish minimum quality from high quality. The weights encode how much each requirement matters for a given use case. These thresholds and weights were developed through expert input; the framework reports engagement with more than 60 experts from network research, public policy, digital inclusion, ISPs, and content providers (Ohlsen et al., 23 Sep 2025). The 2026 implementation paper states that the original design phase relied on workshops and interviews, prior studies, and domain knowledge, and that the implementation paper constitutes a second phase focused on software and preliminary data-driven evaluation (Sermpezis et al., 9 Jun 2026).

The reported requirement weights illustrate the intended semantics. Latency is weighted 5 for gaming, indicating its critical role. Upload is weighted 1 for audio streaming, indicating comparatively low importance there. Video conferencing assigns weight 4 to download, upload, latency, and packet loss, reflecting a more balanced sensitivity profile. This weighting scheme makes IQB context-dependent rather than metric-agnostic (Ohlsen et al., 23 Sep 2025).

3. Formal score construction

The formal construction is hierarchical. Let $U$ denote the set of use cases, $R$ the set of requirements, and $D$ the set of datasets. For each dataset $d$, use case $u$, and requirement $r$, the framework defines a binary requirement score $S_{u,r,d} \in \{0,1\}$, where $1$ indicates that the relevant threshold is met and $0$ otherwise (Ohlsen et al., 23 Sep 2025).

The first aggregation stage computes a requirement agreement score across datasets:

$$S_{u,r} = \frac{\sum_d w_{u,r,d}\cdot S_{u,r,d}}{\sum_d w_{u,r,d}} = \sum_d w'_{u,r,d}\cdot S_{u,r,d}.$$

This is a weighted average of binary success or failure scores across datasets. If all relevant datasets indicate that the requirement is met, then $R$0; if none do, then $R$1; intermediate values arise when datasets disagree or receive different weights (Ohlsen et al., 23 Sep 2025).

The second stage computes the use-case score:

$R$2

Substituting the first equation yields a nested weighted average over requirements and datasets (Ohlsen et al., 23 Sep 2025).

The final IQB Score is then

$R$3

Equivalently, the overall score can be written as a hierarchical weighted average across datasets, requirements, and use cases. The 2026 implementation paper describes the output of this process as a composite index normalized to $R$4, where $R$5 corresponds to very poor or unmet requirements, $R$6 corresponds to full satisfaction of the chosen requirements, and intermediate values indicate partial fulfillment (Ohlsen et al., 23 Sep 2025, Sermpezis et al., 9 Jun 2026).

This construction means that IQB is fundamentally threshold-based and weighted. It does not directly predict subjective MOS; instead, it composes many small pass/fail judgments into a single index. A plausible implication is that interpretability depends strongly on the thresholds, weights, and percentile rule used upstream.

4. Measurement inputs, percentile aggregation, and implementation

In the implemented framework, IQB is computed from crowdsourced speed-test measurements, especially M-Lab-style data. The implementation paper emphasizes that these data are noisy and biased because users often run tests when they think the connection is bad, sampling is geographically uneven, participation correlates with income, urbanization, and device type, and M-Lab’s server placement can create longer-than-normal paths (Sermpezis et al., 9 Jun 2026).

For a country, region, city, or ISP, the measurements are first aggregated into a distribution, and IQB then selects a percentile threshold of that distribution. The initial recommendation is the 95th percentile, intended to approximate infrastructure capacity or best observed performance rather than average user experience. The implementation also evaluates the 75th, 50th, and 25th percentiles (Sermpezis et al., 9 Jun 2026). The earlier framework paper likewise states that IQB uses the 95th percentile of dataset measurements to evaluate a metric, and gives the example of using the 95th percentile packet loss for users in a region when assessing whether the region meets the packet-loss criterion for high-quality gaming (Ohlsen et al., 23 Sep 2025).

The selected percentile values for throughput, latency, and packet loss are compared with the minimum thresholds for each use case. Each use case then receives a score in $R$7, and these are aggregated into the final composite IQB Score (Sermpezis et al., 9 Jun 2026).

The implementation paper reports an open-source IQB library hosted at https://github.com/m-lab/iqb and a prototype web application at https://iqb.mlab-staging.measurementlab.net/. These tools support the computation and visualization of IQB scores at the level of countries, regions, cities, and, where data allow, individual ISPs. Reported outputs include global maps, country/region/city comparisons, per-use-case score distributions, percentile-versus-score curves, and longitudinal comparisons across time (Sermpezis et al., 9 Jun 2026).

Several concrete examples are given. Under the 95th percentile, nearly all countries appear very high-performing on the global map; under the 75th percentile, lower IQB values emerge in parts of Africa and Central Asia. The United States reaches a perfect IQB Score already at the 75th percentile, while Zambia reaches a perfect score only at the 90th percentile. For use-case-specific outputs, more than 50% of countries achieve perfect scores for web browsing, while only about 10% do so for gaming, and nearly half fall below 0.5 for gaming (Sermpezis et al., 9 Jun 2026).

5. Sensitivity, interpretation, and limitations

The preliminary data-driven evaluation identifies percentile choice as the main source of sensitivity. Higher percentiles produce higher scores; the 95th percentile often yields values close to 1 for most countries; the 75th percentile still produces many near-perfect scores but reveals more variation; the 50th percentile yields a much more dispersed distribution; and the 25th percentile shifts scores lower still (Sermpezis et al., 9 Jun 2026). The practical distinction is explicit: higher percentiles emphasize best-case infrastructure capacity, while lower percentiles are closer to more typical user experience (Sermpezis et al., 9 Jun 2026).

Robustness is partial rather than absolute. The reported Pearson correlation between country-level scores computed with the 75th versus 50th percentile is 0.8, which suggests preservation of broad ordering, but not invariance. Rankings can still change, so percentile choice can materially affect prioritization and allocation decisions (Sermpezis et al., 9 Jun 2026).

The framework also reports weak dependence of country-level IQB score on sample size, with Pearson correlation 0.15. At the same time, sparse-data warnings remain necessary. The implementation paper tentatively suggests that several hundreds of measurements are likely needed for meaningful estimates, notes that there is no clear minimum sample-size threshold, and warns that year-over-year differences greater than 0.2 may be suspicious and could reflect low sample counts rather than actual change (Sermpezis et al., 9 Jun 2026).

The conceptual framework paper describes several design choices as provisional: the selected weights, the thresholds, and the use of the 95th percentile aggregation rule. It also notes implied limitations: dependence on the availability and comparability of public datasets; differences in how datasets measure performance; expert-informed but still subjective thresholds and weights; and the possibility that a single composite score oversimplifies complex quality dimensions. The implementation paper adds the measurement biases of crowdsourced speed tests and the absence, at that stage, of a full formal uncertainty model or external ground-truth validation (Ohlsen et al., 23 Sep 2025, Sermpezis et al., 9 Jun 2026).

A common misconception is that IQB is merely a more elaborate speed score. The framework’s design contradicts that interpretation: it is defined through use cases, multiple requirements, multiple datasets, and weighted aggregation. Another misconception is that the reported score directly reflects average user experience. The use of the 95th percentile in the initial recommendation means that the score may instead approximate infrastructure capacity or best observed performance (Sermpezis et al., 9 Jun 2026).

6. Relation to adjacent research and broader significance

IQB belongs to a broader family of efforts that translate raw Internet measurements into user-meaningful or policy-meaningful indicators. A closely related open-source, machine-learning-driven framework for QoE assessment in multimedia networks automates data collection, generates MOS labels using ITU-T P.1203, and predicts user satisfaction from delay, jitter, packet loss, bitrate, and throughput. That system is described as “IQB-style” in the sense that it continuously collects network-quality data, converts them into standardized QoE labels, and predicts user satisfaction from live Internet measurements (Panahi et al., 2024). The distinction is methodological: the QoE framework predicts MOS for multimedia sessions, whereas IQB constructs a composite Internet-quality index centered on use cases and threshold satisfaction.

Other adjacent work addresses problems that IQB must confront in practice. Neighborhood-scale analysis of approximately 170 million crowdsourced Ookla speed tests shows that internet quality is locally configured, that sampling bias requires correction, and that download speed, upload speed, and latency can have different socio-economic drivers once areas are stratified by population density (Kalamadi et al., 29 May 2026). Statewide mobile-quality estimation from Ookla data shows that heavily imbalanced spatial sampling necessitates adaptive estimation methods such as self-tuning kernel regression if one aims to construct geographically faithful quality surfaces (Jiang et al., 2023). These studies do not define IQB itself, but they identify representativeness, granularity, and spatial imbalance as central empirical problems for any barometer-like instrument.

Within this broader context, IQB’s principal significance lies in its attempt to formalize a shift from raw network statistics to “fit for purpose” Internet quality. It combines what users do online, what those activities require from the network, and what publicly available measurements show. This suggests a general role for IQB in benchmarking, cross-region comparison, policy analysis, and Internet experience assessment, provided that the encoded definition of quality—especially percentile, thresholds, and weights—is made explicit (Ohlsen et al., 23 Sep 2025, Sermpezis et al., 9 Jun 2026).