Gaming evidence

The Builder can attach selectionSource: "gaming-evidence-v1" and a gamingDecision object to a gaming recommendation, built on gpu-index-v2 and cpu-index-v2. This page documents what that evidence actually is: external, published benchmark research, re-indexed by RigRoute — never something RigRoute itself ran, and never a promise about a specific FPS in a specific game.

In short: an index score is a relative estimate from third-party benchmark research, re-indexed by RigRoute. It is not a measured FPS, and RigRoute does not guarantee a specific frame rate in a specific game on a specific machine. claimsMeasuredFps: false on every estimate built from these two indices — with no exception.

What the indices are

gpu-index-v2 and cpu-index-v2 are relative tables, not measurements. Each GPU entry carries a score at 1080p, 1440p and 4K: the card's geometric-mean average FPS on a published multi-game suite at that resolution, divided by the fastest GPU's (GeForce RTX 5090 = index 100), stored at the source's own 0.1-point precision. A rounded 5-point display value exists for UI purposes only. Each CPU entry carries one 1080p score on the same logic, tested with a flagship GPU so the GPU is not the limiting part.

In plain terms: an index score tells you where a card or chip lands on a published test suite relative to the other cards on that same suite. It does not predict frame rate in a specific game on a specific machine, and none of the underlying benchmark runs were performed by RigRoute. The indices carry no price — a Danish price is combined with an index score only inside Market's price lookups, when a build is assembled.

Sources

TechPowerUp's relative-performance figures were deliberately left out: the site answers automated requests with a bot check, which RigRoute's own notes record was not bypassed. TechSpot itself returns HTTP 403 to automated requests; the text above was retrieved from publicly archived copies on web.archive.org, not by bypassing any access control.

Confidence levels

Known limits

• Only one of the two sources covers the full catalogue for either index; the second source is partial and pairwise (specific card-versus-card comparisons stated in article text, not a complete independent table). Many entries carry MEDIUM rather than HIGH confidence simply because no second source exists to check them against.
• Older CPUs are built as a linear bridge from the publisher's own legacy hierarchy page, fit on the CPUs present on both pages. Every bridged entry is labelled LOW, with a stated uncertainty.
• Name matching is imperfect. RigRoute's own integration notes record real mismatches: a card listed as “Radeon RX 9070XT” with no space resolves to the RX 9070 entry instead of RX 9070 XT (understating a faster card); “Ryzen 5 5600XT” resolves to the 5600X entry, “Ryzen 7 7700X3D” to 7700X, and “Ryzen 9 9950X3D2” to 9950X3D. Some of these are plausibly harmless (an F-suffix part mapped to its K-suffix sibling); the X3D and XT cases are not, and remain an open item, not a hidden one.
• The reference suite runs at native resolution, high/ultra settings, no ray tracing, no upscaling and no frame generation. A card's place on this suite says nothing about its ray-tracing or upscaled performance.
• The indices carry no price information at all.
• The underlying research reflects one point in time (retrieved 2026-10-08); driver and game-patch maturity differ between the primary and cross-check sources, part of why some pairs disagree by more than a handful of points.

What happens for UNKNOWN hardware

A part with no entry for the exact SKU is recorded with the literal status UNKNOWN. It is never turned into a guessed number.

Where the build generator still needs to say something about an UNKNOWN part, it uses a separately labelled, cruder fallback (a coarse tier or core-count heuristic) with a stated penalty — never silently raised to the same confidence as an evaluated entry.

More importantly, and separately from the above: even where both the GPU and CPU index entries are HIGH confidence, the Builder's gaming recommendation (gamingDecision) can only ever report RECOMMENDED, MET or NOT_MET from an admitted, whole-system, same-game, same-settings measured benchmark record. As of this writing, no such record has been admitted, so gamingDecision reports EXPLORE/UNKNOWN on every request, regardless of how confident the underlying GPU/CPU index entries are. A high-confidence index score is not the same claim as a RECOMMENDED recommendation, and the two should never be read as one and the same.

Refresh policy

RigRoute's own refresh procedure states a cadence directly: review the indices “every new GPU/CPU generation, or at least quarterly.” A changed reference suite or reference card gets a new index version (for example a future gpu-index-v3) instead of a silent edit of gpu-index-v2.

A status change at three specific, previously reviewed scenarios triggers a human review before a refreshed index ships. Where RigRoute's own notes do not specify something — for example an exact calendar date for the next refresh — this page says so plainly rather than inventing a date.

Active versions, for a technical reader

• Index files: gpu-index-v2, cpu-index-v2 (schema version 2).
• Builder selection source: selectionSource: "gaming-evidence-v1", attached only to an automatic, EXPLORE-mode gaming recommendation, behind the feature flag GAMING_EVIDENCE_V1 (off by default).
• Decision object: gamingDecision, with its own recommendation: RECOMMENDED, MET, NOT_MET, UNKNOWN or EXPLORE — governed entirely by the rule above about admitted, measured evidence, independent of GPU/CPU index confidence.

How this relates to RigRoute Labs' own measurements

This page describes external, third-party benchmark research, re-indexed by RigRoute. It is not a RigRoute Labs measurement, carries no E1-E4 level, and is not entered in Labs' own admission registry.

RigRoute Labs' own methodology, standard-ai-v0.1, measures a different thing entirely (local LLM throughput) and has not itself tested gaming workloads. See the Methodology page for what that methodology does and does not establish.

Related

• the Methodology page.
• The Builder's own gaming-recommendation surface lives in RigRoute's separate product, not on this site. A user-facing explainer there is a cross-repo follow-up, not something this page can add.
• No dedicated Garage gaming-assessment page exists on this site today, so there is nothing to link here yet.

Said once more, plainly: a gaming index score is a relative estimate from published third-party benchmark research, re-indexed by RigRoute. It is not something RigRoute itself measured, and it is not a promise about frame rate in a specific game on a specific machine.