What the dataset is — and what it is not
The dataset behind PC Bottleneck Lab is the set of facts the calculator uses to judge a CPU and GPU pairing: what each part is, what it can do, and how pairings behave at different resolutions. It is not a secret formula, and it is not a single measurement. It is a structured collection of published specifications and published benchmark results, combined into model values that the calculator can use.
It is not a laboratory of our own measurements — yet. We do not own a test bench with every CPU and GPU, and this article does not pretend otherwise. What we publish today is built from other people's published data, checked and combined openly. The accuracy test described in our calculator test method is the plan for adding measured results of our own, and until that work is complete, the dataset is a curated collection, not a measurement programme.
That distinction matters for trust. A calculator that claims measurements it never made is worse than one that says plainly where its numbers come from. This article says where ours come from, how they are checked, and where the gaps are.
Sources and provenance
The catalogue was imported from TechAPI records. Those records contain component specifications, benchmark fields and source URLs where available. TechAPI data is attributed under CC-BY-SA-4.0. A separate import script reads a saved Will It Bottleneck snapshot to attach public gaming scores and ranks.
The calculator first checks its built-in pace tables, then uses a public gaming score, then falls back to a stored estimate derived from available benchmark fields. A part without a usable benchmark anchor remains in the catalogue but is excluded from the calculator selectors.
Source coverage is uneven. A source URL does not prove that every derived number was measured on that exact component. The hard-coded tables do not yet include a benchmark citation and test configuration for every value. Manufacturer specifications should be checked against the relevant product documentation before a purchase.
Reference checks and measurement validation
The database contains 18 saved outputs from other calculators. Our reference check runs the current calculation code, compares limiter labels exactly and allows a numerical difference of up to three percentage points. It currently reports 16 matching cases and two label disagreements.
For Ryzen 7 7700 with RTX 5070 at 1080p, the saved reference says 10.3% CPU-limited and our model says 0% balanced. For Ryzen 5 5500 with RTX 4060 at 4K, the reference says 1.6% GPU-limited and our model says 0% balanced. The differences follow from the models' pace values and thresholds; they are not evidence that either model has measured your PC.
The original reference observations are retained. Separate regression checks protect reviewed model behavior. Independent validation would require matched game scenes, settings and test-system measurements. That validation is not complete.
The calculator gives one percentage and FPS estimate per resolution. It does not calculate uncertainty intervals. Read the methodology for its formulas and limitations.
How corrections are recorded
Code and content changes are tracked in the project repository. A complete public, per-record corrections log is not currently available. When a specification is corrected, its original observation and the supporting source should remain distinguishable from derived model values.
To report a possible error, use the contact page and include the component name, the page, and a source or reproducible test. A calculator disagreement alone is not enough to establish which value should change.
Published features and remaining evidence gaps
Available now: searchable component catalogues, CPU and GPU detail pages, a calculator with estimates at three resolutions, and a comparison page using the same pairing model.
The calculator does not select a game or simulate RAM, thermals, driver versions or graphics presets. Its fallback values extend coverage but are estimates, not measurements. The proposed multi-calculator accuracy test and quarterly Index articles describe methods; they do not supply completed measured results.
The next validation work is to attach traceable sources and test configurations to pace values and compare model outputs with reproducible game measurements. Until then, treat the calculator as a planning aid and confirm important decisions with evidence for your games.
Frequently asked questions
Where does the data come from?
TechAPI catalogue records, an imported public gaming score scale, and derived model values. Source coverage and measurement details are incomplete for some values.
Are these your own game measurements?
No. The current dataset is imported and modelled. The proposed accuracy test has not established measured performance for these pairings.
What do the reference checks verify?
They compare the current model with 18 saved competitor outputs. Two limiter disagreements remain. Agreement is not proof of real-world accuracy.
Can I report an error?
Yes. Use the contact page with the component, page and supporting source. Repository history records changes; a complete public per-record corrections log is not available.