Methodology

Source first

Published records pass through source capture, normalization, validation, review and publication. Unknown facts stay unknown.

Why is my Performance Index unavailable?

No population suitable for a normalized performance index is currently installed. CPU, GPU and system indices are unavailable; this does not mean your PC is slow. Specification facts alone cannot produce a defensible 0–100 score.

A future relative index requires comparable measurements, reusable source rights and a documented normalization method. Labels and percentiles require a representative population. No percentile, performance band or sample count is fabricated.

Browser benchmark prototype

Suite rigprism-browser-v1 runs a short single-thread CPU test in a browser Web Worker (integer mixing, floating-point matrix multiplication and array scanning), or WebGPU integer compute. Each workload has one warm-up and three measured runs. The local result shows real duration median and sample variance without submitting anything. Cancel, background tab, navigation, device loss or the 12-second budget discard incomplete results.

WebGPU timings measure command submission through queue completion, including browser/driver overhead. Browser engines, memory, drivers and power states affect both tests. A self-selected model is not hardware identity verification. These tests are not stress tests, gaming FPS or a 0–100 Performance Index.

Threshold policy browser-thresholds-v1: fewer than five independently verified devices is insufficient; five is early data. Public aggregates require at least 30 independent devices within the same component, suite and browser/backend cohort, with a relative 95% median confidence interval width of at most 20%. High confidence requires at least 100 and at most 10%. These sample floors are conservative product policy, not universal statistical guarantees. Three repeated runs on one device still count as one device.

The median interval uses exact binomial order statistics, assuming independent observations from the same population. It does not remove selection bias or make volunteer devices representative. No percentile, population-wide index or gaming winner is inferred. Submitted results require validation, variance/foreground checks and operator-confirmed independent-device evidence before acceptance. Public readers cannot access raw submissions.

Sharing defaults off and requires selecting a model plus a separate submit action. Collection can be paused. Ratings are separate opinions on a five-point scale; no average appears before five actual anonymous votes. Anonymous browser tokens are duplicate prevention, not proof of a unique person or verified playtime.

References: Median confidence regions in a nonparametric model · W3C WebGPU specification.

How is balance determined?

CPU/GPU balance requires comparable benchmark indices and workload context. It is currently insufficient evidence. A difference between relative indices would not represent a percentage of lost FPS. We do not show a universal bottleneck percentage.

Compatibility

Rules use Compatible, Incompatible, Unknown and Manual check required. A matching socket alone does not prove firmware, chipset or generation support. Every known check shows the relevant specification fields and source. Full compatibility is never implied by missing facts.

Power planning

Only documented component power is summed. Complete power facts are required for a system estimate. The current rules add 30% planning headroom, respect a documented GPU system PSU recommendation and round up to 50 W. A partial subtotal is not total system draw or measured wall power.

Replacement checks

Changing one CPU or GPU reruns compatibility with the remaining selected components. Candidates are catalog entries, not ranked performance upgrades. No performance improvement or price-based recommendation is inferred. Unknown and manual checks still need verification.

Evidence, confidence and freshness

Specification confidence is the lowest reviewed confidence label among selected records. It is separate from benchmark confidence, which requires accepted measurements. The oldest selected specification verification date and individual source links are shown. These dates do not establish benchmark freshness.

Compatibility rules version: build-rules-2026-10-03-v1. There is no performance scoring methodology version yet.

Is this FPS?

No. Requirements and FPS data are currently unavailable. No performance score, compatibility status or power estimate is an FPS prediction.

Help improve RigPrism

Collection is currently paused. Build sharing defaults off and would require a separate positive action for each submission. Feedback is a separate action. Missing-component drafts stay in the current browser view; search queries are not stored in contribution tables.

Analyze your build · Contribution privacy · Gaming gear methodology