Exploratory. Six ways the Edo pigments could dress the public site, each with a light and a dark variant, shown as the same small page fragment twelve times so they can be compared like for like. Swatches are reference, underneath, not the main event.
How to read this. Every fragment is identical markup — nav, hero, body paragraph with an inline link, primary and secondary button, a bordered card with two semantic states, a muted caption. Only the nine colour values change. Hover the links and buttons: the hover colour is live, not just a swatch.
The surround is a constant neutral on purpose, and does not follow your OS. Judging light specimens against a light page and dark ones against a dark page is not a comparison.
Contrast is computed, not eyeballed — WCAG 2.x relative luminance, in the page, from the same hex values the fragment renders, so the two cannot drift apart. Anything under 4.5:1 is flagged FAIL.
Nothing here touches bristlenose/theme/colors/palette-edo.css. That file
remains the shipped palette and the source of truth for the app.
Asked for candour rather than six polite paragraphs. Here it is.
KEEPWave Indigo is the one I would build. It is the only candidate where the page itself carries pigment — a bero-ai wash with paper cards floating on it in light, a genuine deep Prussian field in dark — so the two modes read as one idea rather than two unrelated themes that happen to share a name. Everything else in the set goes warm-neutral in light and warm-black in dark, which is a convention, not a palette. The risk is real and worth saying: a blue-washed page reads "considered SaaS" to anyone who does not already know the print it came from. The Edo reference is invisible; only the quality survives the translation.
KEEPPassport Paper works and is boring, which for a product site may be exactly right. It is essentially the shipped light theme wearing website chrome: Prussian link at 6.7:1 on near-white paper, navy ink, parchment rules. A stranger would call it a well-made neutral site and never guess at an exhibition. If the goal is that the site not fight the product, this is the answer. If the goal is that the site be remembered, it isn't.
FLAWCinnabar Seal works, with one structural defect that should decide it. Vermillion is the only pigment in the whole Edo set with enough chroma to behave like an accent against a warm ground, and it has the widest hue separation from the body ink of anything here — 19° in light, 41° in dark, against 0–4.5° in light for the three below that do not. That is why the links actually look like links. But if vermillion is the accent then danger cannot also be red: in the dark variant the failure state and the primary button are 1.1:1 apart, i.e. the same colour. On a marketing site with no destructive actions that is survivable. It is not a palette the app could ever take, so choosing it forks site and product.
MUDDYSail Parchment is the muddiest
thing here and it is instructive about why. Accent hue 74°, body ink hue 78°: a bronze link, on a
brown-black ink, on a tan page, is three lightnesses of one hue. Nothing distinguishes a link
from a word except that it is paler — and on an already-warm ground paler reads as
fading, not as emphasis. Ochre has to be darkened to #7d5410 just to clear
4.5:1 against its own paper, at which point it has stopped being gold. The dark variant is
decent, but a dark ground with a gold accent is not parchment; it is any dark theme.
MUDDYRokusho Court fails the same way,
greener. Accent 183°, ink 185°. Verdigris at its real value (#7ba8a0) cannot carry
a link on a pale ground — it lands near 2:1 — so it has to be deepened to a dark teal, and a
dark teal beside a teal-black ink is the ink's sibling, not a signal. The dark variant is the
best-looking thing in this document that is not Wave Indigo: the true pigment on
#121a18 is genuinely handsome. A palette that only works in one of its two modes is
half a palette.
DROPSumi Snow is not a palette, it is
a typographic position, and it should be judged as one. The accent is the body text —
0° apart, zero chroma delta, by construction — so a link is carried entirely by its underline and
the only colour event on the whole page is the Bengara hover. That can be beautiful, and it is
the most honestly Edo thing in the set. It also puts every affordance on
text-decoration, leaves nothing for buttons, diagrams or state, and gives the site
no colour to be recognised by. Ink on paper is a book. A website has to do more jobs than a
book.
Not premature — just narrower than six. The honest finding is that the Edo source is overwhelmingly a paper-and-ink palette: five of its pigments are low-chroma warm earths, and the two with real saturation (bero-ai, cinnabar) are precisely the two that sampled badly off aged prints — which is why the shipped theme already had to reach past the masstone for Prussian's undertone. Low-chroma earths make lovely application chrome, where colour's job is to get out of the way of the researcher's own content. They make a weak marketing surface, where colour's job is to be remembered. Note what that implies: Wave Indigo and Cinnabar Seal are the two that work, and both work by being less faithful to the source.
So the choice is smaller than it looks — Wave Indigo if the site should look like the theme, Passport Paper if it should look like the product — and it sits underneath a question this exercise did not ask. The site and the app do not have to match, and "the public site is dressed in the app's optional non-default palette" is a slightly odd promise to make a visitor. Worth settling that before tuning any of these further.
One limit on this verdict, stated plainly: the contrast numbers and the hue and chroma separations behind these judgements are measured, but I have not looked at the rendered page — the house rule is that Claude Code preview tools are not used for QA here. Open it and the muddy three should be obvious in about four seconds; if they are not, distrust this section before you distrust your eyes.