/* NoveltyBench — measurement report.
 *
 * The benchmark asks one question: sample a model ten times, how many answers
 * actually differ? So the page has one visual idea — a ten-cell track where a
 * cell is violet if the answer was new and grey if it repeated an earlier one.
 * It carries the hero and the distinct column, so the notation is learned once.
 * Everything else is type and hairlines. */

:root {
  --paper: #f5f6f8;
  --surface: #ffffff;
  --ink: #1b1c21;
  --ink-soft: #5f6270;
  --ink-faint: #9296a3;
  --rule: #dfe1e7;
  --rule-firm: #c8ccd6;

  /* The metric in two values: violet is an answer you had not seen, grey is a
     repeat. No third accent — the palette is the definition. */
  --new: #47376c;
  --new-lift: #6a5794;
  --repeat: #ccced6;

  /* Rank movement between metric versions. Low saturation, and always paired
     with a glyph so colour is never the only channel. */
  --rise: #3d6b4f;
  --fall: #8a5340;

  --serif: "et-book", Palatino, "Palatino Linotype", Georgia, serif;
  --data: "Archivo", -apple-system, BlinkMacSystemFont, "Segoe UI", Helvetica,
    Arial, sans-serif;

  --measure: 66ch;
  --gutter: clamp(1.25rem, 5vw, 4rem);
}

* {
  box-sizing: border-box;
}

html {
  -webkit-text-size-adjust: 100%;
}

body {
  margin: 0;
  background: var(--paper);
  color: var(--ink);
  font: 400 1.125rem/1.62 var(--serif);
  font-feature-settings: "kern" 1;
}

.wrap {
  max-width: 76rem;
  margin: 0 auto;
  padding-inline: var(--gutter);
}

a {
  color: var(--new);
  text-decoration-thickness: 1px;
  text-underline-offset: 0.14em;
}

a:hover {
  color: var(--new-lift);
}

:focus-visible {
  outline: 2px solid var(--new);
  outline-offset: 3px;
}

h1,
h2,
h3 {
  margin: 0;
  font-weight: 400;
  letter-spacing: -0.012em;
}

p {
  margin: 0 0 0.9em;
  max-width: var(--measure);
}

code {
  font: 0.86em/1.4 var(--data);
  background: #eceef2;
  padding: 0.1em 0.32em;
  border-radius: 2px;
}

/* ---------------------------------------------------------------- masthead */

.masthead {
  border-bottom: 1px solid var(--rule-firm);
  background: var(--surface);
}

.masthead .wrap {
  display: flex;
  flex-wrap: wrap;
  align-items: baseline;
  gap: 0.5rem 2rem;
  padding-block: 1.1rem;
}

.wordmark {
  font: 400 1.4rem/1 var(--serif);
  letter-spacing: -0.02em;
  margin-right: auto;
  color: var(--ink);
  text-decoration: none;
}

.masthead nav {
  display: flex;
  flex-wrap: wrap;
  gap: 1.35rem;
  font: 400 0.95rem/1 var(--data);
}

.masthead nav a {
  color: var(--ink-soft);
  text-decoration: none;
  padding-block: 0.2rem;
  border-bottom: 1px solid transparent;
}

.masthead nav a:hover {
  color: var(--ink);
  border-bottom-color: var(--rule-firm);
}

/* ------------------------------------------------------------------- hero */

.hero {
  padding-block: clamp(3rem, 8vw, 5.5rem) clamp(2.5rem, 6vw, 4rem);
  border-bottom: 1px solid var(--rule);
}

.hero h1 {
  font-size: clamp(2rem, 5.2vw, 3.4rem);
  line-height: 1.12;
  max-width: 30ch;
  text-wrap: balance;
}

.hero h1 em {
  font-style: italic;
}

.hero .stand {
  margin-top: 1.4rem;
  font-size: 1.16rem;
  color: var(--ink-soft);
  max-width: 54ch;
}

/* Live evidence: three systems, ten cells each. */
.samples {
  margin: clamp(2rem, 5vw, 3rem) 0 0;
  display: grid;
  gap: 0.65rem;
  max-width: 40rem;
}

.sample {
  display: grid;
  grid-template-columns: minmax(8.5rem, 12rem) 1fr auto;
  align-items: center;
  gap: 0.9rem;
}

.sample-name {
  font: 400 0.95rem/1.3 var(--data);
  color: var(--ink);
  overflow-wrap: anywhere;
}

.sample-name small {
  display: block;
  color: var(--ink-faint);
  font-size: 0.82rem;
}

.sample-count {
  font: 400 0.95rem/1 var(--data);
  font-variant-numeric: tabular-nums;
  color: var(--ink-soft);
  white-space: nowrap;
}

.sample-count b {
  font-weight: 600;
  color: var(--ink);
}

/* the shared notation ------------------------------------------------------ */

.cells {
  display: grid;
  grid-template-columns: repeat(10, 1fr);
  gap: 2px;
  height: 1.5rem;
}

.cells span {
  background: var(--repeat);
  border-radius: 1px;
}

.cells span.is-new {
  background: var(--new);
}

.hero .cells span {
  transform: scaleY(0.34);
  transform-origin: bottom;
  animation: settle 0.5s cubic-bezier(0.2, 0.7, 0.3, 1) forwards;
}

@keyframes settle {
  to {
    transform: scaleY(1);
  }
}

@media (prefers-reduced-motion: reduce) {
  .hero .cells span {
    animation: none;
    transform: none;
  }
}

/* ---------------------------------------------------------------- sections */

section.band {
  padding-block: clamp(2.5rem, 6vw, 4rem);
  border-bottom: 1px solid var(--rule);
}

section.band > .wrap > h2 {
  font-size: clamp(1.5rem, 3vw, 2rem);
  margin-bottom: 0.5rem;
}

.lede {
  color: var(--ink-soft);
  max-width: var(--measure);
}

/* ------------------------------------------------------------- the regimes */

.regimes {
  display: grid;
  grid-template-columns: repeat(auto-fit, minmax(min(22rem, 100%), 1fr));
  gap: 1px;
  margin: 1.6rem 0 1.1rem;
  background: var(--rule);
  border: 1px solid var(--rule);
}

.regime {
  appearance: none;
  border: 0;
  margin: 0;
  text-align: left;
  background: var(--surface);
  padding: 0.95rem 1.1rem;
  cursor: pointer;
  font: inherit;
  display: grid;
  gap: 0.35rem;
  border-left: 3px solid transparent;
}

.regime:hover:not(:disabled) {
  background: #fbfbfd;
}

.regime.is-active {
  border-left-color: var(--new);
}

.regime:disabled {
  cursor: default;
  color: var(--ink-faint);
}

.regime-id {
  font: 500 1.02rem/1 var(--data);
  color: var(--ink-soft);
}

.regime.is-active .regime-id {
  color: var(--new);
}

.regime-spec {
  display: grid;
  gap: 0.16rem;
  font: 400 0.86rem/1.42 var(--data);
  color: var(--ink-soft);
}

.regime-metric {
  color: var(--ink);
}

.regime-count {
  font: 400 0.8rem/1 var(--data);
  color: var(--ink-faint);
}

.meter-key,
.note {
  font: 400 0.88rem/1.5 var(--data);
  color: var(--ink-faint);
  max-width: 62ch;
  margin: 0.6rem 0 0;
}

/* --------------------------------------------------------------- filtering */

.segmented {
  display: flex;
  flex-wrap: wrap;
  gap: 0.45rem;
  margin: 1.5rem 0 0.2rem;
  font: 400 0.92rem/1 var(--data);
}

.segment {
  appearance: none;
  background: var(--surface);
  border: 1px solid var(--rule-firm);
  border-radius: 3px;
  padding: 0.42rem 0.78rem;
  margin: 0;
  cursor: pointer;
  color: var(--ink-soft);
  font: inherit;
  transition: background-color 0.12s ease, border-color 0.12s ease;
}

.segment:hover {
  border-color: var(--ink-faint);
  color: var(--ink);
}

.segment.is-active {
  background: var(--new);
  border-color: var(--new);
  color: #fff;
}

.segment-count {
  font-variant-numeric: tabular-nums;
  margin-left: 0.45em;
  color: var(--ink-faint);
}

.segment.is-active .segment-count {
  color: #cfc7e0;
}

@media (prefers-reduced-motion: reduce) {
  .segment {
    transition: none;
  }
}

/* ------------------------------------------------------------------- table */

.scrollable {
  overflow-x: auto;
  margin-top: 1rem;
}

.table {
  width: 100%;
  border-collapse: collapse;
  background: var(--surface);
  font: 400 0.98rem/1.45 var(--data);
}

.table th,
.table td {
  padding: 0.72rem 0.85rem;
  text-align: left;
  vertical-align: middle;
  border-bottom: 1px solid var(--rule);
}

.table thead th {
  border-bottom: 1px solid var(--rule-firm);
  font-weight: 500;
  font-size: 0.86rem;
  color: var(--ink-faint);
  white-space: nowrap;
}

.sortable {
  cursor: pointer;
  user-select: none;
}

.sortable:hover {
  color: var(--ink-soft);
}

.sortable.is-sorted {
  color: var(--new);
}

.sort-icon {
  font-size: 0.78em;
  opacity: 0.7;
}

.col-rank {
  width: 3rem;
  text-align: right;
  color: var(--ink-faint);
  font-variant-numeric: tabular-nums;
}

.table td.col-rank {
  text-align: right;
}

.col-source {
  width: 7rem;
}

/* Provenance is a label, not a measurement: one hairline treatment for all
   three, so no source reads as ranking above another. */
.source {
  display: inline-block;
  font-size: 0.82rem;
  color: var(--ink-soft);
  border: 1px solid var(--rule);
  border-radius: 2px;
  padding: 0.03rem 0.36rem;
  white-space: nowrap;
}

.source-paper {
  color: var(--ink);
  border-color: var(--rule-firm);
}

.source-unknown {
  color: var(--ink-faint);
  border-color: transparent;
}

.col-open {
  width: 4rem;
  text-align: center;
}

.table td.col-open {
  text-align: center;
}

.col-distinct,
.col-utility {
  width: 12rem;
}

.col-shift {
  width: 6rem;
}

.table tbody tr:hover td {
  background: #fbfbfd;
}

.table tr.is-unscored td {
  color: var(--ink-faint);
}

.unscored {
  font-size: 0.9rem;
}

.num {
  font-variant-numeric: tabular-nums;
  margin-left: 0.6rem;
  font-size: 0.95rem;
}

/* Divided track for a count, continuous for a score. */
.meter {
  position: relative;
  display: inline-block;
  width: 6.6rem;
  height: 0.82rem;
  vertical-align: -0.1rem;
  background: #edeef2;
}

.meter i {
  position: absolute;
  inset: 0 auto 0 0;
  background: var(--new);
}

.meter-units {
  background-image: repeating-linear-gradient(
    to right,
    transparent 0,
    transparent calc(10% - 1px),
    var(--surface) calc(10% - 1px),
    var(--surface) 10%
  );
  background-position: 0 0;
  background-size: 100% 100%;
}

.meter-units::after {
  content: "";
  position: absolute;
  inset: 0;
  background-image: inherit;
  pointer-events: none;
}

.shift-rise {
  color: var(--rise);
}

.shift-fall {
  color: var(--fall);
}

.shift-flat,
.shift-none {
  color: var(--ink-faint);
}

.col-shift .shift-rise,
.col-shift .shift-fall {
  font-variant-numeric: tabular-nums;
}

.open-yes {
  color: var(--rise);
}

.open-no {
  color: var(--ink-faint);
}

/* -------------------------------------------------------------- model cell */

.model-name {
  line-height: 1.3;
}

.model-base,
.tag {
  display: block;
  font-size: 0.82rem;
  color: var(--ink-faint);
}

.tag {
  display: inline-block;
  margin-top: 0.2rem;
  color: var(--ink-soft);
  border: 1px solid var(--rule);
  border-radius: 2px;
  padding: 0.03rem 0.36rem;
  margin-right: 0.3rem;
}

.model-citation {
  color: var(--ink-faint);
  font-size: 0.88rem;
}

.model-link {
  color: var(--ink-faint);
  margin-left: 0.25rem;
}

.group-row th {
  background: #f0f1f5;
  border-bottom: 1px solid var(--rule-firm);
  font-weight: 500;
}

.group-name {
  color: var(--ink);
}

.group-blurb,
.group-count {
  color: var(--ink-faint);
  font-weight: 400;
  margin-left: 0.7rem;
}

.empty-row {
  color: var(--ink-faint);
  padding-block: 2rem;
}

.is-hidden {
  display: none !important;
}

/* ------------------------------------------------------------ judge prompts */

details.prompts {
  margin-top: 1.6rem;
  border-top: 1px solid var(--rule);
  padding-top: 1rem;
}

details.prompts summary {
  cursor: pointer;
  font: 400 1.05rem/1.4 var(--data);
  color: var(--new);
}

details.prompts > * {
  margin-top: 0.9rem;
}

.prompt-label {
  font: 500 0.92rem/1.4 var(--data);
  color: var(--ink);
  margin-bottom: 0.3rem;
  max-width: none;
}

pre.prompt {
  font: 400 0.88rem/1.55 var(--data);
  background: var(--surface);
  border: 1px solid var(--rule);
  border-left: 3px solid var(--new);
  padding: 0.9rem 1rem;
  overflow-x: auto;
  white-space: pre-wrap;
  margin: 0 0 1.2rem;
}

/* ------------------------------------------------------------------ footer */

footer {
  padding-block: 2.5rem 4rem;
  font: 400 0.92rem/1.6 var(--data);
  color: var(--ink-faint);
}

footer .wrap > * {
  max-width: var(--measure);
}

pre.cite {
  font: 400 0.82rem/1.5 var(--data);
  background: var(--surface);
  border: 1px solid var(--rule);
  padding: 0.9rem 1rem;
  overflow-x: auto;
  color: var(--ink-soft);
}

.authors {
  color: var(--ink-soft);
}

@media (max-width: 40rem) {
  body {
    font-size: 1.05rem;
  }

  .sample {
    grid-template-columns: 1fr auto;
    grid-template-areas: "name count" "cells cells";
    gap: 0.35rem 0.9rem;
  }

  .sample-name {
    grid-area: name;
  }
  .sample-count {
    grid-area: count;
    text-align: right;
  }
  .cells {
    grid-area: cells;
    height: 1.15rem;
  }
}

.model-date {
  display: block;
  font-size: 0.78rem;
  color: var(--ink-faint);
  font-variant-numeric: tabular-nums;
  margin-top: 0.15rem;
}

/* --------------------------------------------------------- title & prose */

.hero .stand {
  font-size: 1.28rem;
  color: var(--ink-soft);
  margin-top: 0.5rem;
}

.hero .authors {
  margin-top: 1.4rem;
  font: 400 0.95rem/1.6 var(--data);
  color: var(--ink-faint);
  max-width: 60ch;
}

.figure {
  display: block;
  width: min(100%, 34rem);
  margin: 1.6rem 0;
}

section.band details {
  border-top: 1px solid var(--rule);
  padding-top: 1rem;
  margin-top: 1.4rem;
}

section.band details summary {
  cursor: pointer;
  font: 400 1.08rem/1.4 var(--data);
  color: var(--ink);
  margin-bottom: 0.6rem;
}

section.band details summary strong {
  font-weight: 500;
}

section.band ul {
  max-width: var(--measure);
  padding-left: 1.1rem;
  margin: 0 0 0.9em;
}

section.band li {
  margin-bottom: 0.5em;
}

footer h2 {
  font-size: 1.15rem;
  color: var(--ink-soft);
  margin-bottom: 0.4rem;
}

/* ------------------------------------------------------------ metric names */

/* Two rows, the metric named in the same face and colour it wears as a column
   header, so the definition and the column read as the same thing. */
.metrics {
  display: grid;
  gap: 0;
  margin: 1.5rem 0 1.2rem;
  max-width: 62rem;
  border-top: 1px solid var(--rule);
}

.metrics > div {
  display: grid;
  grid-template-columns: minmax(9rem, 12rem) 1fr;
  gap: 0.4rem 1.4rem;
  padding: 0.85rem 0;
  border-bottom: 1px solid var(--rule);
}

.metrics dt {
  font: 500 1.02rem/1.4 var(--data);
  color: var(--new);
}

.metrics dd {
  margin: 0;
  color: var(--ink);
  max-width: 58ch;
}

@media (max-width: 34rem) {
  .metrics > div {
    grid-template-columns: 1fr;
    gap: 0.2rem;
  }
}

/* -------------------------------------------------- leaderboard front matter */

.notes {
  display: grid;
  grid-template-columns: repeat(auto-fit, minmax(min(24rem, 100%), 1fr));
  gap: 0.2rem 2.6rem;
  margin: 0 0 0.4rem;
}

.notes p {
  max-width: none;
  color: var(--ink-soft);
}

.metrics {
  max-width: none;
}

@media (max-width: 48rem) {
  .notes {
    gap: 0;
  }
}

/* ----------------------------------------------------------------- article */

/* Headings sit in a rail beside their prose, so a comfortable measure is a
   deliberate column rather than a line that stops short of the table above. */
.article .row {
  display: grid;
  grid-template-columns: minmax(9rem, 15rem) minmax(0, 66ch);
  gap: 0.6rem 3rem;
  padding: 1.9rem 0;
  border-bottom: 1px solid var(--rule);
}

.article .row:first-child {
  padding-top: 0;
}

.article .row:last-child {
  border-bottom: 0;
  padding-bottom: 0;
}

.article h2,
.article h3 {
  font-size: 1.3rem;
  line-height: 1.25;
  color: var(--ink);
  text-wrap: balance;
}

.article .prose > :last-child {
  margin-bottom: 0;
}

.article .figure {
  margin: 1.4rem 0 0;
  width: 100%;
}

@media (max-width: 52rem) {
  .article .row {
    grid-template-columns: 1fr;
    gap: 0.7rem;
  }
}

/* The analysis page's guide reuses the article rail, with a rule above it to
   separate it from the charts it explains. */
.guide {
  border-top: 1px solid var(--rule-firm);
  padding-top: 2rem;
  margin-top: 3rem;
}

.guide ul {
  max-width: none;
  padding-left: 1.1rem;
  margin: 0;
}

.guide li {
  margin-bottom: 0.5em;
}