scadaproj/code-reviews/Theme/findings.md

# Code Review — Theme

| Field | Value |
|-------|-------|
| Library | `ZB.MOM.WW.Theme/` |
| Packages | `ZB.MOM.WW.Theme` (Razor Class Library) |
| Component spec | `components/ui-theme/spec/SPEC.md` |
| Shared contract | `components/ui-theme/shared-contract/ZB.MOM.WW.Theme.md` |
| Status | Reviewed |
| Last reviewed | 2026-11-09 (re-review; baseline 2026-06-01 at `5f75cd4`) |
| Reviewer | Claude (2026-11-09 re-review, scadaproj#15; baseline: Claude automated) |
| Commit reviewed | `91fb2b4` |
| Open findings | 2 |

## Summary

**2026-11-09 re-review (`91fb2b4`, scadaproj#15).** Re-reviewed at `91fb2b4` (2026-11-09) against the 2026-06-01 baseline `5f75cd4`; the kit is now `0.5.0`.
New since the baseline:

- the CSS-only `<details>` app shell, with the `::details-content` fix and the mobile shell fix;
- `nav-state.js` (localStorage persistence of `NavRailSection`, active-link reveal, and a
  `MutationObserver` backstop for interactive render modes);
- `ThemeScripts`;
- the `TechButton` `Size`/`Class` extensions;
- the stateless `TreeTable<TRow>`/`ITreeRow`;
- the upstreamed `.btn` sizing block and the 0.5.0 presentation classes.

The C# components hold no state, implement no `IDisposable` and hold no JS interop references. There is
no `MarkupString` anywhere, and every caller-supplied string is attribute- or text-encoded by Razor.
`LoginCard` still delegates `ReturnUrl` validation and antiforgery to the host, with explicit docs.
`TreeTable` documents its Bootstrap Icons dependency.

The `MutationObserver` is never disconnected. That is fine: it is page-lifetime, created once per
document, and the observe→mutate loop is guarded. The `<details>` shell relies on author `display`
rules for browsers without `::details-content`. That could not be verified without those browsers, so it
is not raised.

All 60 bUnit/static-asset tests pass. There are two new findings:

- **Theme-007 (Medium):** `TechButton` silently discards a caller-supplied `disabled` attribute. The
  explicit `disabled="@Busy"` is written after the splat, so a button the caller meant to disable
  renders enabled. Verified with a bUnit probe.
- **Theme-008 (Low):** `nav-state.js` marks a section as wired with a DOM attribute. Blazor's
  enhanced-navigation DOM sync strips that attribute, so toggle listeners pile up on each navigation,
  and the extra listeners save the transient active-link reveal as the user's preference, which is the
  issue #2 regression. Verified by reading the code and simulating it in jsdom, with a control.

**2026-06-01 baseline (`5f75cd4`).** `ZB.MOM.WW.Theme` is a small, clean, well-scoped Blazor Razor Class Library: ten flat-namespace
SSR components, two stylesheets (`theme.css` 379 lines, `layout.css` 191 lines), three vendored
IBM Plex `.woff2` fonts, and two public enums. The component C# is correct and minimal — stateless
render logic, no concurrency surface, no secret handling, sensible `[Parameter]`/`EditorRequired`
choices, and `TreatWarningsAsErrors`. The 32 bUnit tests (20 `[Fact]` + 12 `[Theory]` cases) exercise
the public render contract of every component plus a static-asset check on the corrected font path.
The standout security-sensitive area — `LoginCard` (open-redirect + antiforgery) — is handled
correctly by delegating both concerns to the consumer with explicit XML docs and in-markup security
notes, matching the spec. No findings are Critical or High. The findings are concentrated in two
themes: (1) one functional gap — the mobile hamburger relies on Bootstrap's collapse JS, an
undocumented runtime dependency that contradicts the "no JavaScript" framing; and (2) CSS / token
hygiene — a token-fidelity drift on `.chip-idle`, hardcoded hex values that violate the spec's
"no hardcoded hex" rule, an undefined `.rail-ico` class, orphan/dead nav CSS, and a near-total
absence of XML docs on the public parameter surface.

## Checklist coverage

2026-11-09 re-review at `91fb2b4`:

| # | Category | Examined | Notes |
|---|----------|----------|-------|
| 1 | Correctness & logic bugs | ☑ | `TechButton` drops caller `disabled` → Theme-007. Duplicate toggle listeners after enhanced navigation persist the transient reveal → Theme-008. `TreeTable` (keyed rows, leaf ARIA omission, invariant-culture depth style) and the `NavRailSection` slug are correct. |
| 2 | Public API surface & compatibility | ☑ | Additive in 0.2.0–0.5.0: `ButtonSize`, `TechButton.Size`/`Class`, `TreeTable`/`ITreeRow`, `NavRailSection.Key`, `LoginCard.Heading`, `ThemeScripts`. No removals. The `CaptureUnmatchedValues` contract is not fully honoured for `disabled` (Theme-007). |
| 3 | Concurrency & thread safety | ☑ | Stateless components; the JS is single-threaded with idempotent `apply()`. No issues found. |
| 4 | Error handling & resilience | ☑ | `localStorage` access is wrapped in try/catch, and the script degrades when `window.Blazor` or `MutationObserver` is absent. No issues found. |
| 5 | Security & secret handling | ☑ | No `MarkupString` and no raw HTML. `Accent` goes into a `style` attribute but is developer-supplied. `LoginCard` `ReturnUrl` is echoed into an encoded hidden field, and the open-redirect check stays documented as the host's job. localStorage holds only open/closed flags. No issues found. |
| 6 | Performance & resource management | ☑ | The `MutationObserver` (whole-document `childList`/`subtree`) is page-lifetime and cheap per mutation. Toggle listeners accumulate one per enhanced navigation (Theme-008). No `IJSObjectReference` or `IDisposable` to leak. |
| 7 | Spec & shared-contract adherence | ☑ | No-JS shell holds: native `<details>`, no Bootstrap JS. Tokens and chips are unchanged apart from documented additions. No new drift found. |
| 8 | Packaging, dependencies & project layout | ☑ | 0.5.0 RCL, `FrameworkReference` only. `nav-state.js` is shipped and asserted by `StaticAssetsTests`. AngleSharp is pinned to 1.5.0 in the test project only. No issues found. |
| 9 | Testing coverage | ☑ | 60 tests pass. Gaps: no test passes `disabled` through `TechButton`'s splat (Theme-007), and nothing exercises `nav-state.js` under DOM sync (Theme-008). |
| 10 | Documentation & XML docs | ☑ | Theme-006 fix holds (XML docs on every component, parameter and enum). README documents `TreeTable`'s Bootstrap Icons and interactivity requirements. No issues found. |

<details><summary>2026-06-01 baseline checklist (<code>5f75cd4</code>)</summary>

| # | Category | Examined | Notes |
|---|----------|----------|-------|
| 1 | Correctness & logic bugs | ☑ | Component render logic correct; `open="@Expanded"` boolean-attribute binding behaves (tests confirm); chip/variant `switch` mappings complete. No issues found. |
| 2 | Public API surface & compatibility | ☑ | Flat `ZB.MOM.WW.Theme` namespace; minimal, intentional; `EditorRequired` applied; nullable annotations correct; no internal types leak. Surface matches contract. No issues found. |
| 3 | Concurrency & thread safety | ☑ | Stateless components, no shared mutable state, no `async`/statics/singletons. No issues found (UI kit). |
| 4 | Error handling & resilience | ☑ | No throwing paths; required params default to `string.Empty`; null `RenderFragment` slots are guarded (`is not null` / `IsNullOrEmpty`). No issues found. |
| 5 | Security & secret handling | ☑ | No secrets/PII. `LoginCard` open-redirect + CSRF correctly delegated to consumer with docs. `Accent` interpolates into a `style` attr but is developer-supplied (not untrusted). No issues found. |
| 6 | Performance & resource management | ☑ | No `IDisposable`, no allocations of note, fonts vendored once. No issues found. |
| 7 | Spec & shared-contract adherence | ☑ | Component/parameter API matches SPEC §4 and the shared contract. Drifts found: `.chip-idle` foreground (Theme-002) and "no hardcoded hex" rule (Theme-003). |
| 8 | Packaging, dependencies & project layout | ☑ | `net10.0` RCL, single package id `ZB.MOM.WW.Theme`, `FrameworkReference` only, version `0.1.0` (inherited from `Directory.Build.props`), static assets under `_content/`. No issues found. |
| 9 | Testing coverage | ☑ | 32 tests cover every component's render contract + corrected font path. CSS-class wiring (e.g. `.rail-ico`, orphan nav CSS) is not asserted, which is how Theme-004/005 slipped through. |
| 10 | Documentation & XML docs | ☑ | README/contract accurate. XML docs present only on `LoginCard`; absent on the other 9 components, all their parameters, and both enums (Theme-006). |

</details>

## Findings

### Theme-001 — Mobile hamburger toggle silently depends on Bootstrap collapse JS

| | |
|--|--|
| Severity | Medium |
| Category | Spec & shared-contract adherence |
| Status | Resolved |
| Location | `ZB.MOM.WW.Theme/src/ZB.MOM.WW.Theme/Components/ThemeShell.razor:5` |

**Description**

`ThemeShell` renders the narrow-viewport navigation toggle as a Bootstrap collapse trigger:
`<button ... data-bs-toggle="collapse" data-bs-target="#theme-rail" aria-expanded="false">`, with the
rail wrapped in `<div class="collapse d-lg-block" id="theme-rail">`. The expand/collapse behaviour
and the `aria-expanded` state transition are driven entirely by **Bootstrap's collapse JavaScript
bundle** at runtime. The shared contract states "No global JavaScript … `NavRailSection` is CSS-only"
and the spec emphasizes static-SSR/no-JS operation, while §5 only requires each app to keep its own
Bootstrap **`<link>`** (CSS) — Bootstrap *JS* is never mentioned. A consumer that includes only
Bootstrap CSS (a reasonable reading of the docs) gets a hamburger button that does nothing below the
`lg` breakpoint: the rail stays `display:none` (`.collapse`) and is unreachable on mobile/narrow
windows. There is no compile-time signal, and the hardcoded `aria-expanded="false"` never updates
without the JS, so assistive tech is also misinformed. This affects every consumer's small-viewport
navigation.

**Recommendation**

Either (a) document the Bootstrap-JS runtime requirement explicitly in the README/spec adoption steps
("the side-rail mobile toggle requires Bootstrap's collapse JS bundle"), or (b) make the toggle
JS-independent to match the "no JavaScript" promise — e.g. wrap the rail in a `<details>`/checkbox
CSS-only disclosure (the same pattern `NavRailSection` already uses), so collapse works in pure static
SSR. If keeping the Bootstrap-collapse approach, document that `aria-expanded` requires the JS to stay
accurate.

**Resolution**

Resolved in `544a6dd`, 2026-06-01 — removed the Bootstrap-collapse JS dependency: the shell is now a native CSS-only `<details>`/`<summary>` disclosure (no `data-bs-*`), force-shown on lg+; bUnit test asserts no Bootstrap collapse hooks.

### Theme-002 — `.chip-idle` foreground diverges from the documented token pairing

| | |
|--|--|
| Severity | Low |
| Category | Spec & shared-contract adherence |
| Status | Resolved |
| Location | `ZB.MOM.WW.Theme/src/ZB.MOM.WW.Theme/wwwroot/css/theme.css:177` |

**Description**

`DESIGN-TOKENS.md` ("Status tokens — tinted backgrounds") states the chip classes pair each tinted
background with its matching foreground token — "`--ok-bg` background with `--ok` foreground … Used by
`.chip-ok`, `.chip-warn`, `.chip-bad`, `.chip-idle`". The `--idle` token (`#868e96`) is correctly used
for `.s-idle` and the `.conn-pill .dot`, but `.chip-idle` uses `color: var(--ink-soft)` (`#5a6066`)
instead of `color: var(--idle)`:

```css
.chip-idle { color: var(--ink-soft); background: var(--idle-bg); border-color: var(--rule-strong); }
```

So the one idle chip variant does not follow the pairing rule the design-tokens doc spells out. It is
not a contrast or readability regression (`--ink-soft` is darker than `--idle`), but code and the
normalized token reference disagree, which is exactly the spec-drift category §7 asks to flag. Note the
`.chip-warn` foreground is likewise a hardcoded `#b56a00` rather than `var(--warn)` — see Theme-003.

**Recommendation**

Either change `.chip-idle` to `color: var(--idle)` to honor the documented pairing, or, if the darker
`--ink-soft` ink is the deliberate choice, amend `DESIGN-TOKENS.md` to record that `.chip-idle` uses
`--ink-soft` foreground (as it already does for the `Info` variant footnote).

**Resolution**

Resolved in `544a6dd`, 2026-06-01 — changed `.chip-idle` foreground from `var(--ink-soft)` to `var(--idle)`, honoring the DESIGN-TOKENS pairing rule; static-asset test asserts the pairing.

### Theme-003 — Hardcoded hex values in CSS contradict the "no hardcoded hex" rule

| | |
|--|--|
| Severity | Low |
| Category | Spec & shared-contract adherence |
| Status | Resolved |
| Location | `ZB.MOM.WW.Theme/src/ZB.MOM.WW.Theme/wwwroot/css/theme.css:174` (and theme.css:96-98,145-150,175-176,251-254,281,290-291,335,348,368-369,377; layout.css:123,128,184) |

**Description**

`SPEC.md` §1 and `DESIGN-TOKENS.md` both assert the rule: "Components carry **no hardcoded hex values**
— everything references these tokens" / "no hardcoded hex values appear in component markup or CSS".
In practice both stylesheets contain numerous literal hex colors outside the `:root` token block —
chip/agg border tints (`#c6e6cd`, `#eec3c3`, `#efd6a6`, `#bfe3c6`, `#f0d9ab`, `#f0c0c0`), chip/notice
foregrounds (`#b56a00`, `#8a5a00`), zebra/hover fills (`#fbfbf9`, `#f3f6fd`, `#eef2fc`), and the
info/dir tints (`#e7ecfb`, `#cdd9f7`). These are derived shades (lighter borders, hover washes) with no
corresponding token, so they are not re-themable via the token block and any per-app accent override
will not reach them. The rule as written is therefore aspirational, and the `DESIGN-TOKENS.md` Info
footnote itself already references a raw `#e7ecfb`, so the doc partially contradicts its own rule.

**Recommendation**

Reconcile code and doc: either promote the recurring derived shades to named tokens (e.g.
`--ok-border`, `--hover-bg`, `--info-bg`/`--info-border`) and reference them, or soften the spec wording
to "no hardcoded hex in **component markup**; CSS derives border/hover shades from tokens where
practical" so the rule matches reality. This keeps the "token changes are breaking" SemVer guarantee
honest.

**Resolution**

Resolved in `544a6dd`, 2026-06-01 — promoted every derived border/ink/wash shade to named `:root` tokens (`--ok-border`, `--warn-ink`, `--info-bg`, `--hover-bg`, `--active-bg`, `--zebra-bg`, …) and referenced them; no hex literal remains outside `:root` in either stylesheet, asserted by a new test.

### Theme-004 — `NavRailItem` emits a `.rail-ico` span that no stylesheet defines

| | |
|--|--|
| Severity | Low |
| Category | Correctness & logic bugs |
| Status | Resolved |
| Location | `ZB.MOM.WW.Theme/src/ZB.MOM.WW.Theme/Components/NavRailItem.razor:4` |

**Description**

When the optional `Icon` slot is supplied, `NavRailItem` wraps it: `<span class="rail-ico">@Icon</span>`.
Neither `theme.css` nor `layout.css` defines a `.rail-ico` rule (confirmed by grep). The icon therefore
renders as an unstyled inline span — no sizing, alignment, or gap relative to the label text — so an
icon + text rail link will not line up the way the other rail styling implies. It is not a crash, but
the component ships a class hook with no backing style, which is a latent presentation bug for any
consumer that uses the `Icon` parameter (and none of the 32 tests cover the icon path).

**Recommendation**

Add a `.rail-ico` rule to `layout.css` (e.g. `display:inline-flex; width:1rem; margin-right:0.4rem;
color:var(--ink-faint);`) alongside the existing `.rail-link` rules, and add a bUnit test asserting the
`.rail-ico` span is emitted when `Icon` is supplied.

**Resolution**

Resolved in `544a6dd`, 2026-06-01 — added a `.rail-ico` rule to `layout.css` (inline-flex, sized, gapped, `--ink-faint`); bUnit tests assert the span is emitted when `Icon` is supplied and omitted otherwise.

### Theme-005 — Orphan and unstyled nav CSS classes in `layout.css`

| | |
|--|--|
| Severity | Low |
| Category | Documentation & XML docs |
| Status | Resolved |
| Location | `ZB.MOM.WW.Theme/src/ZB.MOM.WW.Theme/wwwroot/css/layout.css:75` (`.rail-eyebrow`), `:103` (`.rail-eyebrow-chevron`); `ZB.MOM.WW.Theme/src/ZB.MOM.WW.Theme/Components/NavRailSection.razor:5` (`.rail-eyebrow-label`) |

**Description**

`NavRailSection` renders `<summary class="rail-eyebrow-toggle"><span class="rail-eyebrow-label">@Title</span></summary>`
and `layout.css` provides the actual collapse chevron via `.rail-section > summary::before` (lines
178-181). That leaves several mismatched class hooks: `.rail-eyebrow` (line 75, non-toggle) and
`.rail-eyebrow-chevron` (lines 103-108) are defined in CSS but emitted by **no** component (dead CSS),
while `.rail-eyebrow-label` is emitted by the component but has **no** CSS rule (it merely inherits from
`.rail-eyebrow-toggle`). The result is two parallel chevron mechanisms (the styled-but-unused
`.rail-eyebrow-chevron` and the actually-used `::before`) and stylesheet cruft that misleads future
maintainers about how the section header is themed.

**Recommendation**

Remove the dead `.rail-eyebrow` and `.rail-eyebrow-chevron` rules (or wire `.rail-eyebrow-chevron` into
`NavRailSection` and drop the `::before` approach — pick one), and either add a `.rail-eyebrow-label`
rule or drop the now-redundant wrapper span. Keeping one chevron mechanism removes the ambiguity.

**Resolution**

Resolved in `544a6dd`, 2026-06-01 — removed the dead `.rail-eyebrow` and `.rail-eyebrow-chevron` rules and dropped the redundant `.rail-eyebrow-label` wrapper span in `NavRailSection`, leaving one chevron mechanism (`summary::before`).

### Theme-006 — Public component/parameter surface lacks XML documentation

| | |
|--|--|
| Severity | Low |
| Category | Documentation & XML docs |
| Status | Resolved |
| Location | `ZB.MOM.WW.Theme/src/ZB.MOM.WW.Theme/Components/ThemeShell.razor:23` (and `BrandBar.razor`, `NavRailItem.razor`, `NavRailSection.razor`, `StatusPill.razor`, `TechButton.razor`, `TechCard.razor`, `TechField.razor`, `ThemeHead.razor`, `ButtonVariant.cs`, `StatusState.cs`) |

**Description**

This is a packaged library (`IsPackable=true`, `0.1.0`) consumed by three separate apps, and
`REVIEW-PROCESS.md` §2.10 calls for "XML doc accuracy and presence on the **public** API (these are
libraries — public docs matter)." Only `LoginCard` carries XML docs (on `ReturnUrl` and `ChildContent`).
The other nine components, all of their `[Parameter]` properties, and both public enums
(`ButtonVariant`, `StatusState`) have no `///` documentation, so consumers get no IntelliSense/tooltip
help on the parameter contract (e.g. that `ThemeShell.Accent` overrides the `--accent` token, that
`NavRailItem.Match` defaults to `Prefix`, or what each `StatusState`/`ButtonVariant` member maps to).
The README and shared contract document all of this, but that prose is not surfaced at the call site.

**Recommendation**

Add brief `///` summaries to the public components, their parameters, and the two enum types — at
minimum the non-obvious ones (`ThemeShell.Accent`, `NavRailItem.Match`, `TechButton.Busy`,
`TechCard.Title` vs `Header` precedence, and each enum member). Consider enabling
`GenerateDocumentationFile` so missing-doc warnings surface under the existing `TreatWarningsAsErrors`.

**Resolution**

Resolved in `544a6dd`, 2026-06-01 — added `///` summaries to all nine remaining components, every public `[Parameter]`, and both enums (`ButtonVariant`, `StatusState`); enabled `GenerateDocumentationFile` (CS1591 left off the error set for Razor-generated members) so docs build under `TreatWarningsAsErrors` with 0 warnings.

<!-- 2026-11-09 re-review (scadaproj#15) findings below. -->

### Theme-007 — `TechButton` silently discards a caller-supplied `disabled` attribute

| | |
|--|--|
| Severity | Medium |
| Priority | P2 |
| Category | Correctness & logic bugs |
| Status | Open |
| Location | `ZB.MOM.WW.Theme/src/ZB.MOM.WW.Theme/Components/TechButton.razor:3` |

**Description**

The button is rendered as
`<button @attributes="AdditionalAttributes" type="@Type" class="@ButtonClass" disabled="@Busy">`.
In Blazor, an attribute written after an `@attributes` splat overrides the same-named splatted
attribute. `disabled` is not a declared parameter, so a caller's `disabled="…"` is captured into
`AdditionalAttributes` and then overwritten by `disabled="@Busy"`. When `Busy` is `false`, Blazor drops
the attribute entirely. 0.5.0 added a merge for `class` (`ButtonClass` folds in the splatted `class`),
but `disabled` was not merged. The XML doc on `AdditionalAttributes` says "arbitrary additional HTML
attributes splatted onto the underlying `<button>`". The existing test
`Class_composition_keeps_busy_type_and_click_behavior` covers only the opposite direction
(`Busy=true` beating caller `disabled=false`).

**Failure scenario**

A page renders `<TechButton Variant="ButtonVariant.Danger" disabled="@(!CanDelete)" @onclick="Delete">Delete</TechButton>`
for a user who cannot delete, or for a form that is not yet valid. The output is
`<button type="button" class="btn btn-danger">Delete</button>`, with no `disabled`, so the button is
clickable and fires `Delete`. A bUnit probe in the scratchpad confirmed this:
`Render<TechButton>(p => p.AddUnmatched("disabled", true))` produced
`<button type="button" class="btn btn-primary">Delete</button>`. No current family call site passes
`disabled` (20 `.razor` files use `TechButton`), so this is latent. It is still a natural use of the
documented splat, and any server-side authorization check would be the only backstop.

**Recommendation**

Compute the effective state as `Busy || callerDisabled`, where `callerDisabled` is read from
`AdditionalAttributes["disabled"]` with the same case-insensitive lookup used for `class`. Better still,
add an explicit `[Parameter] public bool Disabled`. Then render `disabled="@(Busy || Disabled)"`. Add a
bUnit test for caller `disabled=true` with `Busy=false`.

**Resolution**

_Unresolved._

### Theme-008 — `nav-state.js` "wired" guard is a DOM attribute that Blazor enhanced-navigation DOM sync strips, so toggle listeners accumulate and the transient active-link reveal is persisted

| | |
|--|--|
| Severity | Low |
| Priority | P3 |
| Category | Correctness & logic bugs |
| Status | Open |
| Location | `ZB.MOM.WW.Theme/src/ZB.MOM.WW.Theme/wwwroot/js/nav-state.js:7`, `:17-18`, `:39`, `:65-66` |

**Description**

`wire()` marks a section as initialised with the attribute `data-zbnav-initialized`. `apply()` uses
`hasAttribute(INIT_ATTR)` as its only guard against adding a second `toggle` listener (commit `9e4dedc`,
"guard nav-state.js against duplicate toggle listeners"). After an enhanced navigation, Blazor's DOM
sync (`blazor.web.js` 10.0.1, its attribute synchronisation step) keeps the existing `<details>` element
but removes every attribute that is not in the newly server-rendered HTML. That includes
`data-zbnav-initialized`, which the server never renders. Then `enhancedload` calls `apply()`, which
finds no marker and calls `wire()` again, adding another `toggle` listener to the same element. After N
navigations the element carries N+1 listeners.

The transient-reveal protection assumes a single listener. The first listener consumes
`data-zbnav-transient` and returns. Every later listener sees no flag and writes
`localStorage[zbnav:<key>]`.

**Failure scenario**

1. A user collapses the "Admin" rail section; `zbnav:admin` is saved as `"0"`.
2. They move around the app by enhanced (static-SSR) navigation.
3. They reach a page whose active link lives inside "Admin", for example from a link in the page body
   or a redirect.
4. `apply()` reveals the section and flags it transient.

The duplicate listeners then overwrite the saved preference with `"1"`, so the section stays expanded on
every later page. This is the issue #2 bug the transient flag exists to prevent.

A jsdom simulation reproduced it, running the shipped `nav-state.js`, stripping the attribute the way
DOM sync does, and firing `enhancedload` 3 times:

- one collapse produced 4 `setItem` writes;
- the reveal produced 3 writes of `"1"`, leaving `saved=1`;
- the control run, with the attribute not stripped, produced 1 collapse write and 0 reveal writes,
  leaving `saved=0`.

**Recommendation**

Track wiring in a JS-side structure that DOM sync cannot touch, for example a module-level
`WeakSet` of wired elements or an expando property (`el.__zbnavWired = true`), instead of a DOM
attribute. Add a test that strips attributes, re-runs `apply()`, and asserts one write per toggle.

**Resolution**

_Unresolved._