evijit's picture
evijit HF Staff
Add new component files and align app to EvalEval design system
dbdd6d1
Raw
History Blame
17.5 kB
import Link from "next/link"
import { ArrowRight } from "lucide-react"
import { Navigation } from "@/components/navigation"
export default function AboutPage() {
return (
<div className="min-h-screen bg-background">
<Navigation />
<main className="mx-auto w-full max-w-[64rem] px-4 pb-24 pt-12 sm:px-8">
{/* HEADER --------------------------------------------------------- */}
<div className="kicker">About · Working paper v0.4</div>
<h1
className="mt-2 mb-7"
style={{
fontSize: "clamp(40px, 5.2vw, 56px)",
fontWeight: 700,
letterSpacing: "-0.03em",
lineHeight: 1.05,
color: "var(--fg)",
}}
>
A reporting layer for AI evaluation.
</h1>
<p className="mb-6 text-[19px] leading-[1.6] text-[color:var(--fg-muted)]">
<strong className="text-[color:var(--fg)] font-semibold">Eval Cards</strong> is a
structured registry of how AI models are evaluated — and, just as importantly, of
what is left undocumented. It composes existing evaluation infrastructure into a
single audience-agnostic reading surface. It is a research artifact of the{" "}
<strong className="text-[color:var(--fg)] font-semibold">EvalEval Coalition</strong>,
a community of academic and industrial labs working on broader-impact evaluation of
AI systems.
</p>
<p className="mb-5 text-base leading-[1.75] text-[color:var(--fg)]">
Benchmark scores are routinely reported without the context required to interpret
them: prompts, decoding parameters, evaluator identity, reproduction artifacts,
scope of validity. Eval Cards treats every published evaluation as a{" "}
<em>claim</em>, and every absent field as a claim <em>not made</em>. Neither is an
error — the distinction is what makes the public record useful.
</p>
<p className="mb-14 text-base leading-[1.75] text-[color:var(--fg)]">
The card format is audience-agnostic. A researcher and a policy analyst look at
different fields on the same record. Reader modes (Research · Policy) surface the
fields most load-bearing for each audience; the underlying data is shared.
</p>
{/* BUILT ON ------------------------------------------------------- */}
<section className="mb-14">
<div className="section-head">
<h2>What it is built on</h2>
</div>
<div className="grid gap-5 sm:grid-cols-2">
{[
[
"Auto-BenchmarkCards",
"A schema for benchmark-level metadata — what a benchmark measures, its splits, intended use, validity scope, and known limitations. Each benchmark family in this registry has an Auto-BenchmarkCard at the family root and a Policy Note compressed for plain-language reading.",
],
[
"Every Eval Ever",
"A run-level corpus of public evaluation results — (model, benchmark, metric-path, value, source) tuples extracted from papers, model cards and leaderboards. Provides the raw rows the registry canonicalises and joins.",
],
[
"IBM Risk Atlas alignment",
"Risk-domain annotations on benchmarks (capability, robustness, safety, agentic risk, fairness) so policy readers can locate which deployment-relevant property a number speaks to.",
],
[
"Six-level hierarchy",
"Family → Suite → Single benchmark → Split → Subtask → Metric. Every score resolves to an explicit path, so aggregate claims drill down to the evidence supporting them.",
],
].map(([h, p]) => (
<div
key={h}
className="border border-[color:var(--border-soft)] bg-[color:var(--bg-warm)] p-[22px]"
>
<h3 className="m-0 mb-2 text-base font-semibold tracking-[-0.005em] text-[color:var(--fg)]">
{h}
</h3>
<p className="m-0 text-[13.5px] leading-[1.6] text-[color:var(--fg-muted)]">
{p}
</p>
</div>
))}
</div>
</section>
{/* TWO READER MODES ---------------------------------------------- */}
<section className="mb-14">
<div className="section-head">
<h2>Two reader modes, one record</h2>
</div>
<p className="mb-5 max-w-[700px] text-[15px] leading-[1.7] text-[color:var(--fg-muted)]">
The same evaluation record renders differently depending on the question the
reader brings to it. Toggle in the topbar; the URL, the data and the citations
are unchanged.
</p>
<div className="grid grid-cols-1 sm:grid-cols-2 border border-[color:var(--border-soft)]">
<div className="p-6 sm:border-r border-[color:var(--border-soft)]">
<div className="kicker mb-2">Research</div>
<h3 className="m-0 mb-2.5 text-lg font-semibold text-[color:var(--fg)]">
Methodology read
</h3>
<p className="m-0 text-[13.5px] leading-[1.65] text-[color:var(--fg-muted)]">
Setup variants, n-shot, decoding parameters, evaluator identity, confidence
intervals, and the specific schema fields missing for reproduction are
foregrounded on every metric row.
</p>
</div>
<div className="p-6 bg-[color:var(--fg)] text-[color:var(--bg)]">
<div className="kicker mb-2" style={{ color: "var(--accent)" }}>
Policy
</div>
<h3 className="m-0 mb-2.5 text-lg font-semibold">Plain-language read</h3>
<p className="m-0 text-[13.5px] leading-[1.65]" style={{ color: "rgba(240,237,232,0.78)" }}>
Policy Notes (measures · caveat · intended for), risk-domain annotations,
first/third-party evaluator tags, and disclosure-gap flags are foregrounded;
metric configuration is compressed.
</p>
</div>
</div>
</section>
{/* FOUR SIGNALS --------------------------------------------------- */}
<section className="mb-14">
<div className="section-head">
<h2>Four interpretive signals</h2>
</div>
<p className="mb-5 max-w-[700px] text-[15px] leading-[1.7] text-[color:var(--fg-muted)]">
Computed over each <code className="font-mono text-[12px]">(model, benchmark, metric-path)</code> record and aggregated to the
corpus. Per-record instances appear on every model and benchmark page;
corpus rollups appear on the home page.
</p>
<ol className="list-none p-0 m-0">
{[
[
"Reproducibility",
"Can a third party run this evaluation and obtain a comparable number? Tracks setup-variant disclosure, prompt and decoding parameters, harness version, seed, and code/artifact availability.",
],
[
"Completeness",
"Does the record meet the standard report card for this class of model? Tracks coverage across capability, robustness, safety and fairness benchmarks expected for the model's claimed use.",
],
[
"Provenance & risk",
"Who produced this number, and which deployment-relevant property does it speak to? Tracks evaluator identity (first-party / third-party), source citation, and IBM Risk Atlas-aligned risk domain.",
],
[
"Comparability",
"Can two scores under the same benchmark be put side-by-side? Tracks split, subtask, metric variant, and unit harmonisation; flags rows that cannot be ranked together.",
],
].map(([h, p], i) => (
<li
key={h}
className="grid grid-cols-[50px_1fr] gap-5 border-b border-[color:var(--border-soft)] py-5"
>
<span
className="font-mono"
style={{
fontSize: 11,
color: "var(--accent)",
letterSpacing: "0.1em",
}}
>
0{i + 1}
</span>
<div>
<h3 className="m-0 text-[17px] font-semibold text-[color:var(--fg)]">{h}</h3>
<p className="mt-1.5 m-0 text-sm leading-[1.65] text-[color:var(--fg-muted)]">
{p}
</p>
</div>
</li>
))}
</ol>
</section>
{/* METHODOLOGY ---------------------------------------------------- */}
<section className="mb-14">
<div className="section-head">
<h2>Methodology</h2>
</div>
<ol className="list-none p-0 m-0">
{[
[
"Canonicalisation",
"Heterogeneous score reports — papers, model cards, leaderboards, blog posts — are normalised to (model, benchmark, split, subtask, metric, value, source) tuples. Model name aliases and benchmark version aliases are resolved against a curated mapping.",
],
[
"Source attribution",
"Each record cites the document of record with a line reference. Where multiple sources report the same configuration, the developer's primary source is preferred and discrepancies are flagged.",
],
[
"Evaluator identity",
"Two categories only: first-party (the model developer) and third-party (an independent evaluator). The two are tagged distinctly and never silently merged; if both have reported on a (model, benchmark) pair, both rows appear separately.",
],
[
"No imputation",
"Empty cells are empty. The registry never estimates, infers, or cross-fills missing values. Disclosure gaps are surfaced as such.",
],
[
"Snapshot discipline",
"Each release is a dated snapshot. Numbers are not back-edited; corrections add a new version with provenance preserved.",
],
].map(([h, p], i) => (
<li
key={h}
className="grid grid-cols-[50px_1fr] gap-5 border-b border-[color:var(--border-soft)] py-5"
>
<span
className="font-mono"
style={{
fontSize: 11,
color: "var(--fg-subtle)",
letterSpacing: "0.1em",
}}
>
M.{i + 1}
</span>
<div>
<h3 className="m-0 text-[17px] font-semibold text-[color:var(--fg)]">{h}</h3>
<p className="mt-1.5 m-0 text-sm leading-[1.65] text-[color:var(--fg-muted)]">
{p}
</p>
</div>
</li>
))}
</ol>
</section>
{/* PRINCIPLES ----------------------------------------------------- */}
<section className="mb-14">
<div className="section-head">
<h2>Principles</h2>
</div>
<ol className="list-none p-0 m-0">
{[
[
"We do not impute.",
"If a developer did not publish a score, the cell is empty. We do not estimate, infer, or cross-fill.",
],
[
"Every number cites its source.",
"Each reported score resolves to a specific document — paper, model card, blog post — with a line reference.",
],
[
"Evaluator identity matters.",
"First-party and third-party results are visually distinct and never silently merged. When both have reported on the same (model, benchmark) pair, both rows are kept side by side.",
],
[
"Gaps are data.",
"Undisclosed fields appear alongside disclosed ones. Silence about a safety benchmark is itself information.",
],
[
"Aggregates resolve to evidence.",
"Every corpus-level claim drills down to the (model, benchmark, metric-path) records that support it. No black-box scores.",
],
[
"Corrections are welcome.",
"Each record links a correction path. The registry is a living artifact; coverage improves as developers publish.",
],
].map(([h, p], i) => (
<li
key={i}
className="grid grid-cols-[60px_1fr] gap-6 border-b border-[color:var(--border-soft)] py-6"
>
<span
className="font-mono"
style={{
fontSize: 12,
color: "var(--accent)",
letterSpacing: "0.1em",
}}
>
0{i + 1}
</span>
<div>
<h3 className="m-0 text-xl font-semibold text-[color:var(--fg)]">{h}</h3>
<p className="mt-2 m-0 text-[15px] leading-[1.65] text-[color:var(--fg-muted)]">
{p}
</p>
</div>
</li>
))}
</ol>
</section>
{/* OUT OF SCOPE --------------------------------------------------- */}
<section className="mb-14">
<div className="section-head">
<h2>What this registry does not do</h2>
</div>
<ul className="list-none p-0 m-0 text-[14.5px] leading-[1.75]">
{[
"Produce a single capability ranking. Metrics across benchmarks are heterogeneous and not commensurable; rolling them into one score throws away the information that makes evaluation useful.",
"Evaluate models. Eval Cards reports on what others have already evaluated. New runs go through the upstream Every Eval Ever pipeline, not this surface.",
"Endorse a benchmark. Inclusion in the registry is a statement about disclosure prevalence, not benchmark quality. Policy Notes describe limitations; reading them is part of using the registry.",
"Replace model cards or system cards. Eval Cards complements them — it is the cross-model, cross-benchmark reading surface that individual cards alone cannot provide.",
].map((t, i) => (
<li
key={i}
className="grid grid-cols-[24px_1fr] gap-4 border-b border-[color:var(--border-soft)] py-3.5 text-[color:var(--fg-muted)]"
>
<span className="font-mono text-[color:var(--fg-subtle)]"></span>
<span>{t}</span>
</li>
))}
</ul>
</section>
{/* CITATION & CORRECTIONS ----------------------------------------- */}
<section className="mb-8">
<div className="section-head">
<h2>Citation &amp; corrections</h2>
</div>
<div className="grid gap-5 sm:grid-cols-2">
<div className="border border-[color:var(--border-soft)] p-[22px]">
<div className="kicker mb-2.5">Cite as</div>
<p className="font-mono m-0 text-[12px] leading-[1.7] text-[color:var(--fg)]">
EvalEval Coalition. (2026). Eval Cards: a reporting layer for AI evaluation
(Working paper v0.4, snapshot 18 Apr 2026). evalcards.evalevalai.com
</p>
</div>
<div className="border border-[color:var(--border-soft)] p-[22px]">
<div className="kicker mb-2.5">Submit a correction</div>
<p className="m-0 mb-2 text-[13.5px] leading-[1.65] text-[color:var(--fg-muted)]">
Each record links a correction path. Disclosure gaps close as developers and
third parties publish; we accept patches against any (model, benchmark,
metric-path) tuple with a citation.
</p>
<span className="font-mono text-[color:var(--fg-subtle)] text-[11px] uppercase tracking-[0.1em]">
corrections@evalevalai.com
</span>
</div>
</div>
</section>
{/* CTA ROW -------------------------------------------------------- */}
<section className="mt-12 flex flex-wrap gap-3 border-t border-[color:var(--border-soft)] pt-10">
<Link href="/" className="btn-ec">
Back to home
<ArrowRight className="h-3.5 w-3.5" aria-hidden />
</Link>
<Link href="/models" className="btn-ec outline">
Browse models
</Link>
<Link href="/evals" className="btn-ec outline">
Browse evaluations
</Link>
</section>
</main>
</div>
)
}