Restore the dark theme's warmth, and the terracotta accent - #801
Merged
Conversation
The token pass neutralised the dark ground. The assistant prose card went from chroma 0.017 to 0.0075, at hue 88 rather than the old ramp's 67-77, and a palette that read as brown started reading as grey. Its LIGHTNESS barely moved (0.232 vs 0.256) — worth stating because the complaint that prompted this was "further from black", and the measurement says the opposite. Warmth, not depth. `--surface` and `--surface-raised` are restored to the pre-token UI's values, sampled from a screenshot of it rather than recovered from the config, then verified against rendered pixels: the page paints #141210 and the prose card #28221a, byte-identical to the reference. Accent text was lifted 26% toward white, which turns terracotta into salmon (#c2603c -> #d58a6f) — the change people mean when they say "the orange changed"; `--accent` itself never moved. The raw accent cannot simply come back: it measures 3.77:1 on the prose card and 3.5:1 on the active surface, both below AA, so restoring it literally would make contrast worse rather than better. The lift is now 14%, the least that clears 4.5:1 against EVERY dark surface — `--surface-active` is the binding constraint and 12% misses it by 0.05. Fills and borders carry no text and are restored exactly, which needs the mix done in sRGB rather than OKLab: the old sub-agent strip was `bg-accent/10` composited over the page, and only an sRGB mix of the same two endpoints lands on that same #251a14. `resolveColor` gains `in srgb` so the contrast suite can evaluate them. The three interaction surfaces are deliberately NOT restored to their old hexes. The old `paddock-800` was a border colour that some places reused as a fill; adopting it as `--surface-active` drops `--text-subtle` to 3.75:1. They keep the ladder's lightness and take only its warmth. All 1721 web tests pass, including the 79 contrast pairs. Light mode untouched. Co-Authored-By: Claude <noreply@anthropic.com>
Deploying paddock with
|
| Latest commit: |
e41fc3f
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://b9544861.paddock-7u2.pages.dev |
| Branch Preview URL: | https://fix-restore-warmth.paddock-7u2.pages.dev |
Merged
edspencer
pushed a commit
that referenced
this pull request
Aug 10, 2026
root-home, adopt-row, adopt-modal, project-settings and the trigger tool picker were all shot before the design work landed (#763/#768/#769/#780/#801). Every one showed chrome, surfaces, type and accent the redesign replaced, on pages whose prose is in the present tense. Shot from the docs-media rig at v0.69.0, Foundation/dark, hue null, tint 0 — the out-of-the-box appearance. capture.mjs now pins the theme via addInitScript (the keys are read by a pre-paint script, so a post-goto evaluate catches a mid-swap frame) and asserts it applied.
edspencer
added a commit
that referenced
this pull request
Aug 11, 2026
Inline `code` in a chat message had no visible background and blockquotes had gone grey. Two different causes, both from the token migration. The chip borrowed `--surface-active`. That worked until a surface moved: #801 restored `--surface-raised` to the pre-token prose card (#28221a) and the gap between the chip and the card it sits on closed to 1.036:1. The chip never changed; the card rose to meet it, and every contrast assertion stayed green because no test pairs a fill with the surface it sits ON. It cannot simply ride `--surface-active` again. That token is pinned near the top of the interaction ladder by `--text-subtle`, which must clear 4.5:1 on it; at the lightness the chip wants (L 0.319) `--text-subtle` measures 3.95:1. A chip only ever carries `--text` (10.5:1 there), so it can afford a lightness the interaction fills cannot — which is the argument for a token of its own rather than a re-tuned ladder. `theme-foundation.css` already refuses the same lift for the same reason. Blockquotes are simpler: they take `--text-muted`, and the new ramp holds the old hue (88 vs 86.5) at a quarter of the chroma (0.015 vs 0.059). Warm to grey is the whole of what "the blockquote lost its colour" means. So: `--code-fill`, `--quote-text` and `--quote-border`, set to the MEASURED values of the pre-token UI rather than re-derived — #3a3125, #a4926a, #5b4d39 in dark, sampled from a v0.66.1 screenshot the way the dark surfaces were. The painted pixels now match that screenshot exactly. Light mode restores the warmth only and not the lightness, because the old light muted step was one of the ones that failed AA at 3.75:1. Parchment, Terminal and Sci-Fi alias the three back to the ladders and are byte-identical before and after — verified by reading computed styles for all four themes in both modes, not by inspection. Parchment's chrome slab restates them alongside the rest of the semantic set it redeclares. Guarded in both `tokens.test.ts` and `themes.test.ts`: the chip must clear 1.15:1 against `--surface-raised` in every theme and mode. Confirmed to FAIL at the old `var(--surface-active)` value, dark mode only — which is exactly where the bug was. Co-authored-by: HomeLab Agent <homelab-infra@valfenda.net> Co-authored-by: Claude <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The 0.67.0 token pass neutralised the dark ground. This puts the warmth back, measured against the UI it replaced rather than eyeballed.
What was actually wrong
#1f1d19-family, chroma 0.006 @ hue 88#141210#28221a— chroma 0.017 @ hue 74.8#d58a6f— accent lifted 26% to white#cd7758— lifted 14%The old values were recovered by decoding a screenshot of the previous UI and sampling pixels, not by reading the old Tailwind config — so what landed is what was on screen. Verified the same way afterwards: the running build paints
#141210and#28221a, byte-identical to the reference.Three things worth knowing before reviewing
1. The prose card's lightness barely moved. The report was "further from black", but 0.232 → 0.256 says otherwise. What halved was chroma. The problem was warmth, not depth, and fixing the wrong one would not have helped.
2. The old orange cannot come back as text, and this is the interesting constraint. Raw
#c2603cmeasures 3.77:1 on the prose card and 3.5:1 on--surface-active— below AA. Restoring it literally would have made contrast worse, which is the opposite of what was asked for. So the lift is now the least that clears 4.5:1 against every dark surface: 14%, giving#cd7758, much closer to the original terracotta than the salmon it replaces.--surface-activeis the binding constraint — 12% misses it by 0.05, which only the full sweep catches. Note--accentitself never changed in 0.67.0; only its derivations did.3. Fills had to be mixed in sRGB. The old sub-agent strip was literally
bg-accent/10over the page, and 10% of#c2603cover#141210composites to#251a14— the exact pixel in the reference. An OKLab mix of the same two endpoints lands somewhere else. Matching the mix space matters as much as the percentage.resolveColorgainsin srgbsupport so the contrast suite can evaluate these tokens instead of throwing.Deliberately not done
paddock-800(#3e3528) as--surface-activeand broke six contrast tests — it was a border colour in the old UI that a few places reused as a fill. Restoring it drops--text-subtleto 3.75:1.--surface-sunkenis not the oldpaddock-950, which was lighter than the page it sat on — not what a sunken surface means in this system.#a84e2f); white on the old#c2603cis 4.17:1, below AA on the most-clicked control in the app. Restoring that is a taste-versus-standard trade and deliberately left as a separate decision.Verification
🤖 Generated with Claude Code