diff --git a/traces/src/components/player/replay-stage.tsx b/traces/src/components/player/replay-stage.tsx
index b6ac074..b5ded9d 100644
--- a/traces/src/components/player/replay-stage.tsx
+++ b/traces/src/components/player/replay-stage.tsx
@@ -159,10 +159,12 @@ export function ReplayStage() {
       {recording && viewport ? (
         /*
           10px, and allowed to be: an instrument readout in the corner of the frame, not text to read.
-          `muted` rather than `faint`, though — measured at 3.09:1 against `bg-base`, which is below the
-          4.5:1 this size needs, and this is the one readout that sits alone on the frame a recording
-          spends its whole runtime pointed at. `muted` is 6.07:1. Same size, same swap, same reason as the
-          ruler's tick labels in `timeline.tsx`, which is the precedent rather than a new judgement.
+          `muted` rather than `faint`, though — this is the one readout that sits alone on the frame a
+          recording spends its whole runtime pointed at, and it is read against replayed page content as
+          often as against `bg-base`. Both tones clear the 4.5:1 this size needs now that `faint` has been
+          raised for AA (`muted` 6.07:1, `faint` 5.26:1 against `bg-base`), so what picks `muted` here is
+          the margin over an unknown ground rather than the threshold. Same size, same swap, same reason as
+          the ruler's tick labels in `timeline.tsx`, which is the precedent rather than a new judgement.
         */
         <p className="absolute bottom-1 right-2 font-mono text-micro tabular-nums text-muted">
           {viewport.width}×{viewport.height} · {Math.round(scale * 100)}%
diff --git a/traces/tailwind.config.ts b/traces/tailwind.config.ts
index e761cf2..3e7b800 100644
--- a/traces/tailwind.config.ts
+++ b/traces/tailwind.config.ts
@@ -57,10 +57,19 @@ const config: Config = {
           strong: '#343b45',
         },
 
-        // Text.
+        // Text. `faint` is pinned by contrast rather than by taste: it is used on eleven different
+        // grounds, and at this value it clears WCAG AA (4.5:1) on all of them. `raised` is the binding
+        // one, being the lightest, at 4.51:1 — anything dimmer fails there. The 3:1 large-text
+        // allowance is no help, because `fontSize` below tops out at 15px and nothing here is bold.
+        // The cost is that it now sits 11 per channel from `muted` instead of 48, so the two read as
+        // one tone in places; that is the most separation the pair can have with both passing.
+        //
+        // One site is still short of AA, and no value here can fix it: the rejected hypothesis card in
+        // `hypothesis-cards.tsx` carries `opacity-60`, which veils text and ground together, so no
+        // foreground can outrun it — 2.61:1 for `faint`, and 2.90:1 for `muted` in the same card.
         ink: '#e6e8eb',
         muted: '#8b929c',
-        faint: '#5b626c',
+        faint: '#808791',
 
         // Authorship. These two carry meaning, not decoration: every marker, hypothesis and
         // activity entry is tinted by who created it. Do not reuse them for anything else.