| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Third v5.0.0 roadmap item: define and implement a migration policy for the
on-device persisted store (tracked domains, history/snapshots, audits,
workflows, monitoring, settings), so data upgrades cleanly across app versions
instead of relying on a one-shot marker.
- DataMigrationService is reworked from a single boolean marker
(`data.migrations.v3_4_0`) into a versioned runner keyed by an integer store
schema version (`data.storeSchemaVersion`). It runs each step once in
ascending order up to `currentStoreSchemaVersion`, stamping the version as it
goes. Adding a future migration is now a `case N:` plus a version bump.
Policy guarantees, all covered by tests:
- Forward-only and idempotent; every step must be safe on an empty/older store.
- Never downgrades: a store written by a newer build (higher version) is left
byte-for-byte untouched.
- Pre-versioning installs are handled: a set legacy boolean marker reads as
"already at v1", so the v1 normalization never re-runs for them.
v1 is the existing normalization pass (dedup + drop the legacy `watchedDomains`
key + sanitize monitoring settings), now expressed as migration step 1.
- Docs/data-migration.md documents the persisted surface, the two independent
version lines (store vs. backup export), when to use lenient decoding vs. a
migration step, the runner contract, an "adding a migration" checklist, and
backup-import compatibility. Linked from the README.
- DataMigrationServiceTests: 6 tests over legacy fixtures — fresh-store stamping,
legacy `watchedDomains` migration + key drop, in-place dedup of the stored
blob, idempotence, legacy-marker-as-v1, and the no-downgrade guard. Full unit
suite: 58 passing.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Second v5.0.0 roadmap item: make the Local API's public JSON contract explicit,
documented, and regression-locked, so external consumers (Shortcuts, scripts,
integrations) have a stable surface with a defined compatibility promise.
- LocalAPIContract: new single source of truth for the wire-format version
("v1") and the canonical JSON encoder (ISO-8601 dates, sorted keys). Both the
success and error paths in LocalAPIService now route through it, so the format
can't drift between them, and the ad-hoc per-call-site encoders are gone.
- The response envelope and every payload struct are promoted from `private` to
internal so the contract is a first-class, testable part of the module. The
transport/handler internals (request parser, HTTP response, secret store)
stay private.
- Docs/local-api.md documents the base URL/auth, the envelope, the encoding
conventions (notably: absent optionals are omitted, not null), every endpoint
and its payload fields, the error codes, and the semantic-version-style
compatibility policy (additive changes keep v1; renames/removals/type changes
bump the version). Linked from the README.
- LocalAPIContractTests: 16 structure/"golden" tests pinning the envelope shape,
each payload's field names, the enum encodings, and the ISO-8601 date format.
They assert structure, not values, so ordinary behavior changes don't churn
them but a renamed or dropped field fails CI. Full unit suite: 52 passing.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
27.0 fix
Executes the Phase-6 manual runbook against the simulator, converting the
mechanically-checkable parts into permanent coverage and reporting the rest
honestly by tier.
- Fix an enforced `.dynamicType` failure surfaced by `audit-a11y.sh current`
on iOS 27.0: the Settings `Section("Services")` system header (app sets no
font; 18.6 floor and 26.x CI are clean). Narrow, proven `noiseReason`
carve-out scoped to dynamicType on the exact Settings header titles. Delta:
current FAIL -> SUCCEEDED, finding still prints as `[noise: …]`.
- AccessibilityMetadataTests: assert the icon-only control labels and the dense
Watchlist/Batch row label+value contracts (green on 18.6 and 27.0). These
were one-time manual VoiceOver checks; now they gate.
- Middle-band Dynamic Type sweep at AccessibilityL across the seeded screens.
Found no band-exclusive third bug (recorded), retained as regression
insurance for a band that historically shipped two.
- AccessibilityScreenshotTests: best-effort, non-gating capture utility used to
produce the cross-runtime Light/Dark/AXXXL screenshots (simctl appearance
does not propagate headlessly; driven through the in-app picker instead).
- Docs/ACCESSIBILITY_VERIFICATION_RESULTS.md: full pass/fail/not-executable
matrix + 15 screenshots. Notable positive result: Differentiate Without Color
IS verifiable via the global com.apple.Accessibility defaults domain.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The point of the Phase 0 harness finally lands: named findings in
textClipped, dynamicType, hitRegion, elementDetection,
sufficientElementDescription, and trait now FAIL the build on the
empty-state suite. Regressions in five phases of accessibility work are
gated, not narrated.
Three carve-outs, each earned by evidence rather than convenience:
- contrast stays report-only. The two long-standing Settings findings are
rows scrolled under the translucent tab bar, and their attribution
flips between a row name and nil run-to-run — no suppression is narrow
enough to keep CI stable. The centralised palette is the real guard.
- The seeded dense-row tests run reportOnly. Bisection showed the audit
degrades on children-ignored content — the correct VoiceOver treatment
for dense rows — emitting unattributed contrast/dynamicType failures on
rows that measure 6-7:1 and render correctly.
- Characterised noise is suppressed narrowly and always logged as
[noise: reason]: disabled controls (WCAG 1.4.3 exempt), "nearly passed"
near-misses, system field placeholders (flagged at any length — proven
by shortening them to no effect), and unattributed clipped/dynamic-type
artifacts. noiseReason(for:) records each rule's provenance inline.
Validated in both directions. Positive: the full 11-test suite passes
with enforcement live, dark and light, on an erased simulator. Negative:
re-injecting the Phase 3 icon-exposure regression produced two named
[FAIL] findings and a failed suite — on both screens sharing the
component — then went green again on revert.
Docs and the CI workflow comment updated to describe the engaged state.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A device checklist for everything the automated audit cannot reach:
VoiceOver speech, the More Content rotor, custom-content ordering,
completion announcements, Voice Control label-in-name, the colour/motion/
transparency settings simctl cannot toggle, iPad keyboard focus order,
and both system design languages (classic vs Liquid Glass).
Grounded in the actual implementation rather than generic guidance: each
item quotes the real label/announcement/field it verifies (e.g. "Refresh
all tracked domains", "Lookup complete for <domain>", the More Content
field order) and carries a [Px] tag pointing at the phase that introduced
it. Organised by iOS setting so each is toggled once, with a seed-data
step up front because the dense rows and widget are otherwise untestable
— the same reason they are unverified by the audit today. Includes a
sign-off matrix, a defect-log template, and the known-deferred items so a
ViewThatFits overflow is filed correctly rather than mistaken for a
regression.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Audit unchanged at 11 dark / 14 light — expected, as none of these
settings are exercised by performAccessibilityAudit, and simctl can
toggle only Increase Contrast, not Differentiate Without Color, Reduce
Motion, or Reduce Transparency. Correct by construction and build-clean;
runtime behaviour is verified in the Phase 6 manual pass.
Color independence:
- Widget status is now an SF Symbol (checkmark.circle.fill /
exclamationmark.triangle.fill / exclamationmark.octagon.fill), the same
vocabulary as the in-app badges, replacing a silent colour-only dot on
both the domain rows and the small-view count pills. Status now survives
greyscale and reads consistently across surfaces.
- Under accessibilityDifferentiateWithoutColor: the Dashboard
summary-card dot becomes a per-filter symbol, the selected quick-filter
chip gains a checkmark and a border (selection was fill-colour only, and
also gains the .isSelected trait), and LabeledValueRow prefixes a
warning/failure symbol. All gated on the setting so the default UI stays
uncluttered.
Reduce motion: all five withAnimation/.animation sites now pass nil under
accessibilityReduceMotion — AppCopyButton's check cross-fade,
CollapsibleSectionView's expand/collapse, TimelineDiffView's scroll, and
WatchlistView's list reorder.
Reduce transparency: the single .thinMaterial capsule falls back to an
opaque AppSurfaceElevated fill under accessibilityReduceTransparency.
The SweepActivityController item from the plan is dropped: it is pure
ActivityKit lifecycle with no animation, confirmed back in the issue
triage.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
phase 4)
The audit count is unchanged at 11 dark, and that is the expected
result: performAccessibilityAudit validates descriptions, traits,
contrast, hit regions, and clipping, but exercises none of VoiceOver's
speech, the More Content rotor, custom-content ordering, or
announcements — which is the entire substance of this phase. It is
verified by construction and stays green with no regressions; the
manual VoiceOver pass is Phase 6.
Icon-only controls (~14) get accessibilityLabel, obeying label-in-name:
where a control has visible text the label keeps it, so Voice Control
still works. The pin and bookmark toggles gain accessibilityValue and
.isSelected; the audit and workflow checkboxes gain .isSelected and a
hint. Decorative icons split out of Labels are hidden.
AppStatusBadgeView now reads as one word ("Critical"), not "icon,
Critical", via children: .ignore + label. SectionTitleView and
CollapsibleSectionView headers get the .isHeader trait for rotor
navigation; the collapsible header also exposes expanded/collapsed as a
value with a hint. The header deliberately does NOT use children:
.combine — its trailing() closure can hold Track/Pin controls, and
combining would swallow them.
Dense rows use combine-for-summary, custom-content-for-detail.
BatchResultRowView (8 elements) and WatchlistRowView (up to 9) become a
single element — domain as label, status as value — with risk, IP,
timestamp, source, certificate, and monitoring on the More Content
rotor, risk and certificate at .high importance. Reading all of it
inline would make a long sweep unnavigable. The custom-content chains
live in ViewModifiers because inlining six of them plus the layout broke
the type-checker. The shorter 3-4 element portfolio rows are left to
NavigationLink's automatic combine, per WWDC21-10121.
Technical strings get a speechStyle field on InfoRowViewData:
.technical applies speechAlwaysIncludesPunctuation and
accessibilityTextContentType(.sourceCode), set on DNS record values and
cipher suites so load-bearing punctuation is not swallowed.
Completion announcements: the sweep posts from the view model; the
single lookup posts from an onChange in the view, since resultsLoaded is
derived from many loading flags and has no single view-model moment.
Widget: each domain row was a silent 8pt status dot plus a bare "12d"
countdown. Rows now read as one phrase ("example.com, critical,
certificate expires in 12 days"); the count pills are labelled.
Not verifiable by the suite: the dense rows and the widget never render
in the audit (no tracked domains or batch results in the test
simulator), same limit as the deferred Phase 3 row reflow. Documented in
Docs/ACCESSIBILITY.md.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Takes the audit from 18 findings to 11 in dark mode. Everything that
remains is system-rendered or placeholder noise, characterised below.
The largest win was not where the plan expected. Every empty-state
heading reported as clipped text, and the cause was `Label`: it
constrains its own title, and `.fixedSize` applied to the Label does not
reach the `Text` inside. Splitting into `HStack { Image; Text }` and
putting the modifier on the Text cleared all four empty states at both
default and accessibility sizes.
That fix then caused a regression the audit caught immediately. `Label`
folds its image into the title's accessibility element; an HStack does
not, so the icon began announcing its raw SF Symbol name
("checklist.unchecked") to VoiceOver. Decorative icons split out of a
Label now carry .accessibilityHidden(true).
Tap targets:
- AppCopyButton was a literal 30x30 on nearly every data row. Now
@ScaledMetric from 44, floored at AppLayout.minimumTapTarget —
@ScaledMetric scales down below the default text size as well as up,
so the floor is load-bearing.
- controlMinHeight was 42 in compact density, putting every collapsible
section header and both Run buttons under the minimum.
Reflow:
- CardView's allowsHorizontalScroll defaulted to true, so nine call sites
hid content behind a horizontal gesture instead of wrapping — a WCAG
1.4.10 failure and the mechanism behind clipped rows at large text
sizes. The default is now false, and the remaining opt-in is suppressed
at accessibility sizes.
- Fixed .system(size:) point sizes replaced with text styles in the app
and the widget.
- The widget is clamped at accessibility1, the one place clamping is
correct: a widget canvas is a fixed size and WidgetKit truncates
overflow with no scroll affordance.
Two hypotheses were tested and discarded rather than left in. Monospaced
fonts looked like the clipping culprit — the app is 82% monospaced and
hyphenates mid-word at accessibility sizes — but switching the empty
state to proportional changed nothing, and prose typography is a design
decision rather than an accessibility fix. Shortening search prompts and
the domain placeholder also changed nothing: placeholder text is reported
clipped regardless of length, so "Search" is flagged exactly as "Search
portfolio" was.
Not done: ViewThatFits reflow for BatchResultRowView and
WatchlistRowView. Those rows never render in the audit because the test
simulator has no tracked domains or batch results, so any change there
would be unverifiable. Absence of findings is absence of data.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reported as "colors seem muted and hard to see on light mode", and
correct. The light palette optimised contrast and produced mud: #7A5600
reads olive rather than amber, #146C2E bottle-dark rather than green.
Contrast passed while the UI got harder to read, because hue identity is
what distinguishes warning from critical at a glance.
Two causes, both fixed.
Every foreground was required to clear 4.5:1 against its own 16% badge
tint — the harshest surface it ever sits on — which pushed each colour
about 20% darker than the common case needed. Most of what is actually on
screen is plain text on a card, with far more headroom. The fill is now
decoupled from the foreground: AppStatusTone carries a foreground and a
surface authored independently, with matching …Surface colorsets, so a
foreground no longer has to survive a wash of itself. Every status
foreground is now fully saturated.
And warning was yellow. Yellow cannot stay yellow at a lightness low
enough to pass 4.5:1 on white — it becomes olive. That is colorimetric,
not a tuning problem. Warning is now orange: #AD5100 light, #FF9F0A dark.
New light values: positive #008035, warning #AD5100, critical #CC0700.
Worst-case ratios 4.54–6.76 across page, card, and surface in both
schemes. Audit findings are unchanged — light 21, dark 18 — so the
vividness costs nothing.
Also picks up a literal .blue missed in phase 1: DomainDiffItem's
low-severity change colour, which the phase 1 sweep did not cover because
its pattern listed only cyan/yellow/green/red/orange/pink.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Removes the 16 scattered .preferredColorScheme(.dark) calls and the one
.toolbarColorScheme, and applies appearance in exactly one place — the
WindowGroup in DomainDigApp. Re-applying per view is what let the lock
spread across eight files unnoticed until light mode was unreachable.
Adds AppAppearance (System / Light / Dark) in @AppStorage, exposed under
Settings > Display next to Density. Honouring the system setting and
offering an override is one key, and it keeps the deliberate dark
aesthetic reachable for anyone who wants it.
Also replaces .secondary with AppTextSecondary across 191 sites. iOS's
own secondaryLabel is 3.29:1 on a light card — below AA — which never
showed while the app was locked to dark, where the same colour reads
6.32:1. Unlocking light mode is precisely what exposed it, so it belongs
here rather than in a later phase: without it, light mode would ship with
body text under 4.5:1 app-wide.
Dark mode reports 18 findings, unchanged from phase 1 — no regression
from unlocking. Light mode reports 21. The three extra are iOS-rendered
Section headers (TIER, PREFERENCES, SERVICES) using the system's grey;
overriding system header styling across every section to gain ~0.3:1 on
decorative labels is a poor trade and is left alone.
Two long-standing Settings contrast findings are now explained. They are
the last rows of a section sitting under the translucent tab bar, so the
audit measures text against a blended background — confirmed by
screenshot, present in dark mode since phase 0, and standard iOS
scroll-under behaviour rather than a defect.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
WCAG 1.4.3 exempts inactive components from contrast requirements, so
Inspect's Run button — disabled until a domain is typed — was reporting a
contrast failure that was never a real defect.
The first attempt typed a domain to enable the button. That worked for
the single-screen test but raised the keyboard, which then followed the
audit onto every later screen in the Dynamic Type sweep and reported nine
phantom hit-region findings per screen against the system emoji picker's
category buttons. Suppressing on the rule instead — drop contrast
findings whose element reports isEnabled == false — fixes it everywhere
with no UI manipulation.
Also documents that simulator keyboard state persists across runs, so a
dirty simulator inflates the burndown with system-UI findings. Erase and
re-run before believing anything that names system UI.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Replaces every hard-coded colour with semantic asset colours that adapt to
light, dark, and Increase Contrast. Dark mode stays locked, so this is a
pure refactor: the audit reports the same findings before and after.
The accent is now blue rather than cyan, per the tech/DNS theme.
Why custom values rather than the system palette: every system colour
fails WCAG AA in light mode. Measured on white — systemYellow 1.51:1,
systemOrange 2.20:1, systemGreen 2.22:1, systemCyan 2.54:1, systemRed
3.55:1. All of them pass in dark mode, which is why the dark-locked app
looked fine, and why unlocking light mode was never a matter of deleting
.preferredColorScheme(.dark). Every new value clears 4.5:1 as text on its
page, its card, and its own 16% badge tint — the way AppStatusBadgeView
actually draws it.
The accent needed splitting in two. As text on a dark background it must
be light; as a fill behind a white label it must be dark. #4DA3FF reads
8.00:1 as text on black but 2.63:1 behind white text, so StatusInfo /
AccentColor cover the foreground role and AccentFill covers
.borderedProminent. AppOnAccent is the label colour for a solid fill and
flips by scheme.
Colours live in Shared/Colors.xcassets rather than the app catalog: the
Shared folder is already a synchronized group in all three targets, so
the widget and share extension pick the palette up with no project-file
surgery. AccentColor stays in the app catalog as the global tint — and is
now actually defined, having been an empty colorset that silently left
system controls rendering in stock blue while custom chrome used cyan.
Two deliberate visual changes: orange folds into StatusWarning and pink
into StatusCritical. They encoded the same severity as the colours they
now share, and both sites also carry a text label.
Audit findings drop 15 to 14, and one of the originals turned out to be
a phantom: the Inspect contrast failure was the Run button in its
disabled state, which WCAG 1.4.3 exempts. testInspectScreen now types a
domain first so the audit measures an enabled control. Findings also
carry the offending element now, so the remaining clipped-text items name
themselves ("No Portfolio Yet", "Search domains") instead of being
anonymous.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The two-job CI matrix was buying two near-identical iOS 26.x runs at
double the macOS minutes. GitHub's macos-26 image ships only 26.x
simulator runtimes, so it cannot test the 17.6 floor at all, and floor
coverage was the entire justification for the second job.
Split the work by what each side can uniquely do instead:
CI keeps one job on the newest runtime. Its real value is not the
runtime — it is building a clean checkout of the merge result, which
catches a file that was never committed. A local run cannot, and that
failure mode is live here: DomainDig.xcodeproj is hand-edited and uses
file-system-synchronized groups, where an entire missing folder still
builds locally. sr.ht cannot run macOS, so this is the only place that
check exists. Collapsing the matrix also removed the deployment-target
math, since "newest" is always above the floor.
Scripts/audit-a11y.sh runs the audit against real runtimes, defaulting
to floor + current. It reads the deployment target from the project
rather than hard-coding it, selects the oldest runtime at or above it
(one below is useless — the app cannot install), and says so plainly
when the nearest installed runtime is a major version above the target
rather than implying floor coverage it does not have.
.githooks/pre-push runs the floor tier, and only when Swift, asset, or
project files changed. Pre-push rather than pre-commit because the suite
takes ~85s: at pre-commit that blocks every commit, and a hook routinely
bypassed with --no-verify is worse than none. Opt in per clone with
'git config core.hooksPath .githooks'.
Docs/ACCESSIBILITY.md records the split, the measured non-nested
coverage that motivates it, and the enforcement ratchet.
|
| |
|
|
|
|
|
|
|
|
|
| |
- Make DomainDig/DomainDig/Audit* the single active Audit Mode implementation
(models, views, exporter) with an Audit tab and session/export UI
- Include audit sessions in backup/restore lifecycle counts, summaries, and
merge behavior via DomainDataPortabilityService
- Remove the DomainDigCLI target, source file, scheme, and all project
references; keep the shared inspection/report pipeline for the app
- Align AppVersion.current to 4.4.1 and refresh README/architecture docs
- Add RELEASE_ROADMAP.md
|
|
|
* consolidate DomainReport as canonical data model
* add feature tier system (free/pro/data+ scaffolding)
* apply clean feature gating across workflows and tracking
* refactor app structure for maintainability
* standardize navigation and settings
* add data lifecycle controls
* ensure consistency across UI, export, and CLI
* stabilize internal inspection API
|
