AI Research Methodology

Last updated: 26 July 2026

AI Explorer is a service and property of Giznova. This methodology explains how we research AI devices, verify capabilities, produce comparisons, and assign confidence scores.

1. Research Philosophy

Giznova AI Explorer exists to make AI technology easier to understand. We believe users deserve transparent, verifiable, and independent information about AI devices. Our methodology is designed to be repeatable, auditable, and honest about its limits.

2. Data Sources

Official Manufacturer Documentation

We prioritize official manufacturer sources including:

• Product specification pages and datasheets.
• Developer documentation and API references.
• Press releases and launch event materials.
• Official technical whitepapers.

When official documentation is available, it is treated as the highest-confidence source.

Public Technical Documentation

We also consult public technical documentation from:

• Chipset vendors (Qualcomm, MediaTek, ARM, etc.).
• Platform providers (Google, Microsoft, Apple, Samsung, Xiaomi, Honor, OnePlus, Lenovo, etc.).
• Industry standards bodies and open-source projects.

Independent Public Sources

When official documentation is incomplete, we cross-reference credible public sources including technical journalism, independent benchmarks, and developer community findings. These sources are used with appropriate confidence downgrades.

3. Feature Verification

AI capability research follows these steps:

1. Source collection: gather all available documentation and public sources.
2. Feature mapping: map documented features to standardized capability categories.
3. Execution mode classification: classify each capability as On-device, Hybrid, or Cloud based on available evidence.
4. Offline status assessment: determine whether the capability works without an internet connection.
5. Minimum software version: note the OS or app version required.

4. AI Capability Research

We maintain a standardized capability framework covering common AI workflows such as translation, transcription, summarization, image generation, search, writing assistance, and productivity automation. Each device is evaluated against this framework where documentation permits.

5. Comparison Methodology

Device comparisons use multi-dimensional scoring:

• AI Rating: derived from validated NPU TOPS values and platform maturity.
• Local AI capability: on-device vs. cloud-dependent processing strength.
• Capability coverage: number and type of verified AI capabilities.
• Platform intelligence: ecosystem maturity, privacy architecture, and AI workflow focus.

Weights and formulas are applied consistently. Manual overrides are not used to inflate or deflate specific devices.

6. Research Confidence Scoring

Each research item receives a confidence score:

• High: confirmed by multiple authoritative sources, including official vendor docs.
• Medium: supported by credible public sources, partially confirmed.
• Low: inferred from related platforms or limited sources.
• Research Pending: research has not yet been completed or verified.

Confidence scores are displayed where users can see them so they can weigh uncertainty accordingly.

7. Research Pending Meaning

Research Pending means that for a particular device, category, or capability, standardized research has not yet been completed. This is common for:

• Newly released devices with limited public documentation.
• Devices from smaller manufacturers with sparse technical disclosure.
• Emerging AI capabilities where standards are still evolving.

Research Pending is not a negative signal - it is a transparency signal. Users should not assume a Research Pending device lacks features; they should simply treat the relevant data as incomplete.

8. NPU TOPS and Performance Notes

NPU TOPS (Trillions of Operations Per Second) is a theoretical throughput metric. It does not fully predict real-world AI performance due to variables like memory bandwidth, thermal design, software optimization, and model architecture. We use TOPS as one input among many.

9. Editorial Review

Research outputs are reviewed against our Editorial Policy. Any material changes to methodology are documented and reflected in future research cycles.

10. Feedback

We welcome questions and corrections about our methodology. Please contact us at info@giznova.in or via our Contact page.

Back to Explorer