docs: track product context and UI critique snapshots

PRODUCT.md records the durable constraints the UI work keeps having to
rediscover: dark-first is binding (night reading), mobile is the target
rather than the starting breakpoint, and updated_at ordering is a
correctness rule the design must not decorate over.

The two critique snapshots are the scored before/after (21/40, then
26/40) the card fixes were driven from — worth keeping as the record of
which problems were already known, and which ones the fixes introduced.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

PRODUCT.md

@@ -0,0 +1,56 @@
+# Product
+
+<!-- impeccable:product-schema 1 -->
+
+## Platform
+
+web
+
+## Users
+
+Single user (self-hosted, no accounts, no multi-user planned). Reads manga on **asurascans.com** and **demonicscans.org** primarily via Bromite on mobile, also checks/updates from a desktop browser. The web UI is the cross-device view into progress captured by the userscript while reading.
+
+## Product Purpose
+
+Tracks read-progress ("last chapter read") per manga series across two otherwise-unrelated manga sites that each have their own separate `localStorage`. A Go backend unifies bookmarks into one store; the web UI is a password-gated browser view of that store for reviewing, favouriting, correcting, or removing bookmarks, and jumping back into a series to continue reading. A background poller also refreshes each series' latest-published-chapter so the list can flag "NEW" without the user visiting the site.
+
+## Positioning
+
+Not a public reading tracker or social app — a private, self-hosted sync layer purpose-built for two specific scraped sites, with no server-side account system (single bearer token + one password-gated session).
+
+## Operating Context
+
+- Primary reading device: Bromite (mobile Chromium), where a userscript captures progress automatically.
+- Web UI is a secondary surface: checking list state, correcting a wrong chapter number, removing dead bookmarks, jumping to "continue reading."
+- Manga cover art and titles come from the source sites' `og:image`/`og:title` — real content, not placeholders.
+- List order is driven by `updated_at`, which moves only on real reading progress (not favouriting, not a newly detected chapter) — a UI constraint the redesign must not break.
+
+## Capabilities and Constraints
+
+- Two tabs: All / Favourites. Search-filter by title (client-side, `filter.js`).
+- Card actions: continue (opens source site), toggle favourite, manual chapter override, delete (with confirm).
+- "Continue reading" horizontal strip for recently-progressed series.
+- htmx-driven partial updates (card re-render on favourite/chapter/delete), no client-side framework/build step — templates are Go `html/template`, `go:embed`-ed.
+- Mobile-first is a hard functional constraint (primary device is a phone), not just a starting breakpoint.
+
+## Brand Commitments
+
+- Name: **mangaBookmark**.
+- **Dark-first is binding**: current dark-by-default / light-follows-system-preference behavior must be preserved as a design constraint, not just a starting default, because reading happens at night.
+
+## Evidence on Hand
+
+- Live templates/CSS at `backend/templates/*.html`, `backend/static/style.css` — current implemented UI, functional but not yet treated as an intentional design system.
+- No logo, screenshots, or marketing copy exist; none should be fabricated.
+
+## Product Principles
+
+- Dark-first, night-reading-optimized — never regress to a light-default or high-glare surface.
+- Mobile is the primary target; desktop is an enhancement, not the design center.
+- Progress data integrity over visual flourish: `updated_at`/list-ordering behavior is a correctness constraint the UI must respect, not decorate over.
+- No accounts, no multi-tenant chrome — the whole product is for one reader.
+- Prefer native platform affordances (system dark/light, native touch targets) over custom widgetry — this is a lean self-hosted tool, not a product to demo.
+
+## Accessibility & Inclusion
+
+Explicit personal requirement: optimized for **low-light/night reading** — minimize glare and eye strain (true dark surfaces, restrained brightness/contrast on accents, no jarring pure-white flashes), beyond generic touch-target/contrast compliance.

backend/.impeccable/critique/2026-07-26T15-50-42Z__backend-templates-app-html.md

@@ -0,0 +1,52 @@
+---
+target: backend/templates (web UI)
+total_score: 21
+max_score: 40
+na_heuristics: 
+p0_count: 0
+p1_count: 2
+timestamp: 2026-07-26T15-50-42Z
+slug: backend-templates-app-html
+---
+Method: dual-agent (A: ab4734f79f4becbf2 · B: af2c412345f87a2d1)
+
+## Design Health Score
+
+| # | Heuristic | Score | Key Issue |
+|---|---|---|---|
+| 1 | Visibility of System Status | 2 | No hx-indicator on any htmx mutation — favourite/save/delete swap silently |
+| 2 | Match System/Real World | 3 | Icon-only actions rely on title= tooltips, which never fire on touch |
+| 3 | User Control and Freedom | 2 | Favourite-toggle and chapter-save are instant, no undo |
+| 4 | Consistency and Standards | 3 | Native unstyled confirm() for delete breaks from themed htmx-swap pattern |
+| 5 | Error Prevention | 2 | Chapter input has no max, no reference to LatestChapter while editing |
+| 6 | Recognition Rather Than Recall | 3 | Good persistent state; edit form doesn't show current latest chapter |
+| 7 | Flexibility and Efficiency | 1 | No bulk actions, no shortcuts, no filter for "has new chapter" |
+| 8 | Aesthetic/Minimalist Design | 3 | Lean; .meta/.actions rows lack visual grouping |
+| 9 | Error Recovery | 1 | No failed-request UI state anywhere |
+| 10 | Help and Documentation | 1 | Zero affordance explains the favourite/NEW-don't-reorder rule |
+
+Total: 21/40 — Acceptable band
+
+## Design Specificity Verdict
+Generic-competent. Nothing signals manga/night-reading/two-site-unification beyond the header text. site-{{.Site}} badge (the actual product differentiator) styled as a neutral pill, same weight as everything else.
+
+Deterministic scan: CLI static-HTML detector 0 findings (exit 0). Browser/live detector found: tight-leading (line-height 1.25, real), overused-font/single-font (100% Roboto), flat-type-hierarchy advisory (~1.5:1 ratio). One text-occlusion hit was a false positive (detector's own leftover overlay text).
+
+## Priority Issues
+P1 — Silent failure on every htmx mutation (no hx-indicator, no error UI). Fix: hx-indicator + htmx:responseError inline error text.
+P1 — NEW-chapter signal has no visual priority/filter despite poller existing for exactly this. Fix: elevate .new styling, add "Updated" quick-filter.
+P2 — Native confirm() on delete breaks dark theme, violates "no pure-white flash" a11y requirement. Fix: inline htmx confirm row.
+P2 — Chapter input has no upper bound/reference to LatestChapter. Fix: show "Latest: Ch N" inline, soft max warning.
+P3 — Single font, flat type scale, line-height under 1.3 floor. Fix: real typographic hierarchy.
+
+## Persona Red Flags
+Casey (mobile, primary): 4 stacked chrome regions before content; touch-only tooltips never fire; silent failures invite repeat-tapping.
+Riley (stress): webkit-only line-clamp with no fallback; no-cover cards render blank with no placeholder; filter.js untested at scale.
+
+## Minor Observations
+Continue-strip not tab-scoped. Fixed accent colors unverified for contrast against light theme. title= tooltips dead weight on touch.
+
+## Questions to Consider
+1. Why does the NEW-chapter signal carry less visual weight than the static site badge?
+2. Why is confirm() (guaranteed to break theme) still unreplaced given the explicit no-white-flash requirement?
+3. Does the ordering rule need an explanation, or does needing one mean the mental model fights the user?

backend/.impeccable/critique/2026-07-26T17-08-41Z__backend-templates-app-html.md

@@ -0,0 +1,51 @@
+---
+target: backend/templates (web UI)
+total_score: 26
+max_score: 40
+na_heuristics: 
+p0_count: 1
+p1_count: 0
+timestamp: 2026-07-26T17-08-41Z
+slug: backend-templates-app-html
+---
+Method: dual-agent (A: a67675dd3039e8156 · B: aa64e48aafb9db799), regression independently re-verified live
+
+## Design Health Score
+
+| # | Heuristic | Score | Key Issue |
+|---|---|---|---|
+| 1 | Visibility of System Status | 3 | htmx dimming/error-inline verified live and working |
+| 2 | Match System/Real World | 4 | "Continue reading," "Updated," site names all read naturally |
+| 3 | User Control and Freedom | 3 | Cancel works; no undo after delete (acceptable, single-user tool) |
+| 4 | Consistency and Standards | 2 | Disclosure-toggle pattern consistent; bare-emoji icons vs. tuned typography mismatch |
+| 5 | Error Prevention | 2 | Chapter input still has no max, hint is prose-only |
+| 6 | Recognition Rather Than Recall | 2 | The hint meant to aid recall is what causes the overflow bug |
+| 7 | Flexibility and Efficiency | 3 | Search, tabs, direct Continue link present |
+| 8 | Aesthetic/Minimalist Design | 2 | Clean tokens undercut by a live horizontal-scroll bug and flat no-cover placeholder |
+| 9 | Error Recovery | 3 | "No connection - try again" verified firing on network abort |
+| 10 | Help and Documentation | 2 | No help affordance beyond hover tooltips |
+
+Total: 26/40 - Acceptable band. Trend: 21 -> 26.
+
+## Design Specificity Verdict
+Real measured progress: site-tinted badges and --new ribbon hue verified via computed contrast (5.3-9.0:1), genuine product-specific signal. Working against it: bare OS-emoji icons (star/pencil/trash/play) clash with tuned typography; no-cover placeholder still a flat contentless box.
+
+Deterministic scan: CLI 0 findings (clean). Live detector: overused-font/single-font expected (deliberate single-typeface Operate surface). flat-type-hierarchy already accepted. monotonous-spacing and text-occlusion confirmed FALSE POSITIVES (detector's own accumulated debug overlays). edge-flush-cards on Continue-reading strip likely false positive (deliberate scrollable carousel with edge-bleed). ai-color-palette flagging new site-tint hues - plausible false positive (contrast-verified, restrained) but unconfirmed.
+
+## Priority Issues
+P0 - Chapter-edit form causes 97-127px horizontal page overflow on mobile, Save button pushed off-screen. .chapter-form has display:flex with no wrap, now holds 3 children (hint/input/button) in one unwrapped row. Confirmed live at 390px and 360px. Regression from the harden command - hint text was verified to render but never checked for container overflow at mobile width. Fix: flex-wrap:wrap + .hint{flex:1 0 100%}, same pattern already used in .actions/.confirm-row.
+P3 - Icon-glyph inconsistency: bare emoji vs tuned type/color system.
+P3 - No-cover placeholder reads as stuck loading skeleton, not designed empty state.
+P3 - .confirm-row text wraps awkwardly ("Remove/this?") - cosmetic only, no overflow.
+
+## Persona Red Flags
+Casey (mobile, primary): chapter-form overflow directly hits her core task (one-handed chapter correction), unexpected sideways scroll needed.
+Sam (accessibility): well-served overall, but no custom :focus-visible on icon/action buttons.
+
+## Minor Observations
+Continue-reading strip still not tab-scoped. updated_at ordering verified live via actual favourite-toggle round-trip. ai-color-palette on site tints worth a second glance, unconfirmed either way.
+
+## Questions to Consider
+1. Did the harden command's verification ever check mobile viewport width with the hint text present, or only that it rendered?
+2. Is the emoji-icon look deliberate "undesigned" for a single-user tool, or just expedient?
+3. Is a missing og:image common enough in practice to deserve real empty-state design?