Dosewise, a medication reminder

<!-- model: claude-opus; skill commit: 7559367 -->

Dosewise: design review (loupe)

1. Summary

Reviewed: the whole app (4 tabs: Today, Medicines, Add, History; 7 Swift files), from native SwiftUI source only, with no simulator. There was no .loupe.md, so I worked from an inferred profile (assumptions below) and did not write the file.

Verdict: The screens are built from correct stock SwiftUI parts, but the app doesn't yet do its one job. It asks for notification permission, then never schedules a reminder. Doses exist only for the day a medicine was added. Nothing survives a relaunch. Two status colours fail text contrast.

Most important: Rework 1, make the schedule a persistent first-class object that generates each day's doses and reminders. It clears findings 4, 5, 12 and 17. Separately, fix the two contrast Blockers (1, 2), which are small token changes.

Assumptions (owner unavailable, so I proceeded on defaults):

• purpose: "Remind a person to take their medicines on schedule and keep an honest record of what was taken or skipped." Inferred from the empty-state copy, TodayView.swift:12 and HistoryView.swift:12.
• feeling: calm and certain ("I know whether I took it"). Inferred.
• audience: occasional-public. A consumer who uses it daily, but not a professional.
• register: utilitarian. It's a frequent check-and-act task.
• motion_personality: serious, relaxed.
• platforms: ios (iPhone only: TARGETED_DEVICE_FAMILY = 1, portrait only in Info.plist).
• build_sdk: unknown. There's no LastUpgradeCheck, .xcode-version, CI config or Xcode-26-only API, and SDKROOT = iphoneos has no version.
• deployment_targets.ios: 17.0 (project.pbxproj). design_compatibility_opt_out: false (key absent).
• release_stage: development (as the request states).
• density: comfortable.
• brand: none. There's no asset catalog, AccentColor, custom font or app icon.
• voice:
  • Traits inferred as "plain but not clinical".
  • Casing follows Apple's native defaults (Title Case buttons). The app's current strings are mixed, see finding 22.
  • spelling: undetermined (no -our/-or or -ise/-ize words), so nothing is flagged.
  • humor: none. we_policy: avoid.
• category: medication reminders and adherence. peers: none given; the landscape step chose them.
• direction: empty, so section 5 proposes directions.
• waivers: none.

Questions I'd have asked in one batch (no answers, defaults above):

1. Purpose and feeling in one sentence each.
2. Who is the user: the patient, a caregiver, or both?
3. Which apps do you measure yourselves against?
4. Which Xcode/SDK do you build with?
5. Are persistence and notification scheduling planned but not yet written? This decides whether findings 4 to 6 are bugs or unfinished work.
6. Spelling preference.
7. Humor: none or light?
8. Which direction in section 5 do you prefer?
9. Saving .loupe.md: not done, per the instructions.

Run notes:

• Tools:
  • Read the source in dosewise/ and the skill references.
  • Ran scripts/contrast.py on source-known colour pairs. Apple's published system colour values were used for .green and the grouped-cell backgrounds (light #FFFFFF, dark #1C1C1E), and #4B5563 comes from Models.swift:68.
  • Computed OKLCH for the sketch values with an inline Python conversion.
  • Web research for the landscape (WebSearch and WebFetch, 2026-10-26). These requests went to third-party sites.
• Not used:
  • No simulator was available, so there are no screenshots and no appearance or content-size changes were made.
  • Playwright, measure-runner.js and type-eval.js were not run. The target is native, and no specimen page was built because the app ships only the system face.
• Side effects: none. No files were edited apart from this report, and no .loupe.md was written.
• Measured results:
  • "Taken" label: .green (#34C759) on a white cell is 2.22:1. With Increase Contrast (#248A3D) it's 4.40:1, which still fails for 17 pt text.
  • doseTakenText #4B5563 on a dark cell (#1C1C1E) is 2.25:1. On white it's 7.56:1.
• Whole-app split: none. The app has 4 screens and 7 view and model files, under the split threshold, so every pass ran inline.
• 26 gate: the SDK is unknown, so I gated on the iOS 17.0 deployment target and skipped the (26+) checks (see Unverified).

2. Findings

waived: 0

3. Trust

I found no deceptive patterns in the source: no confirmshaming, fake urgency, preselection or invented counts. The notification request at launch (finding 16) is a timing and rationale problem, not a deceptive one, because it uses the system alert with nothing pre-selected.

4. Landscape

Category: medication reminders and adherence (nearest seed playbooks: Health, and Consumer utilities). Checked 2026-10-26.

Conventions to keep:

• The reminder is a notification that carries the decision. In Apple Health you "tap the notification, then tap Skipped or Taken", and on Apple Watch "Remind Me in 10 Minutes" is also offered (Apple Support 105064).
• Both taken and skipped are recorded (Apple; MyTherapy: "documents every intake, whether taken or skipped").
• Doses recur daily from a schedule, and the record persists.

Frontier moves:

• Follow-up reminders when a dose isn't logged 30 minutes after it's due. Apple offers optional Critical Alerts per medication that sound through Focus and mute (Apple Support 105064, Macworld).
• Resending until resolved (MyTherapy, Healthify [S]).
• Escalation to a caregiver about 1 hour after an unlogged dose (Medisafe "Medfriend", help center), or 15 minutes after (Hero).
• Adherence reports shared with clinicians as PDF (Medisafe, [S]).

Convergence to avoid:

• A caregiver feature as the headline when the core reminder isn't reliable.
• Stock health teal or blue with a pill glyph as the whole identity (unverified: peers' identity combinations weren't checked visually).

Openings (unverified: no peer was confirmed to lack these):

• Show taken time against due time as a plain, judgement-free record ("taken 2 h late"). Dosewise already stores resolvedAt.
• A day modelled as the physical artefact people already use to organise doses, a weekly pill box or blister card (Direction A).

Core-task step count (log a due dose):

• Apple Health: 2 inputs (tap notification, tap Taken). Sourced, Apple Support 105064.
• Dosewise: 2 inputs (open app, tap Take), but only if the person remembers unprompted, because no reminder ever fires (finding 4).

The input counts are equal, so the peer-gap bar isn't met. The gap is reliability, not steps.

5. Distinctiveness verdict

default-assembled.

Channels (none authored):

• Authored channels: 0 of 9. Signature element: none. Signature moments: 0, visual or behavioural.
• Catalogued tells: 7, across layout idiom, iconography and sound and haptics (finding 19). That meets the slop-cluster threshold, which forces this verdict.
• Logo-off test (judged from source; no screenshots): nothing identifies the product.
• Competitor-swap test: with "Medisafe" or "MyTherapy" on these screens, nothing looks wrong.
• Gate: every flow has open Majors, so pleasure-tier work waits for the fixes. Directions are proposed now so the team can choose.
• Default check (the theme I'd generate for "medication reminder app"): a teal or health-blue accent, white rounded cards, a pill glyph, green checkmarks, and SF Rounded. Both directions below avoid it on colour and on layout idiom.

Inferred candidate principles (no brand doc; finding "No authored point of view" is folded into 19):

1. Certain over cheerful: state is always explicit.
2. The record is honest: late is shown as late, never hidden or scolded.
3. Quiet until due: nothing competes with the next dose.

Direction A: "Blister card"

Source sentence: many people organise doses in a pill box or blister card, where each pocket is one dose and an empty pocket is unmistakable.

Register: utilitarian, with one behavioural moment. Motion: serious, relaxed.

Color:

The three values derive from one input (hue around 70); the increased-contrast variant uses the light-text value on both fills.

Type: system face in every role, plus .monospacedDigit() on times. No new face, so nothing to license or measure.

Shape: each day row shows a strip of round "pockets", one per dose. Each pocket is filled while pending and emptied (outline) when taken. Skipped is marked with a slash and the text "Skipped". Radius is circular for pockets only; everything else stays system.

Motion:

• Productive: system defaults.
• Expressive: the pocket empties with a short scale-down and fade (about 0.2 s, ease-out) plus .sensoryFeedback(.success).
• Reduce Motion: fade only.

Iconography: custom pocket symbols drawn from an SF template, with accessibility labels.

Signature element: the pocket strip, which recurs on Today, the widget, notifications (as a thumbnail) and the icon. Signature moment: emptying the pocket on Take.

Voice samples:

• Button: "Take"
• Empty state: "No doses left today. Next: Metformin at 8:00 AM."
• Error: "Reminders are off. Turn them on in Settings to get a reminder at each dose time."

Landscape position:

• vs Apple Health: a physical-artefact day model instead of a list (Apple Health's look was not checked visually, unverified).
• vs Medisafe and MyTherapy: amber vial vernacular instead of their brand hues (unverified).
• vs Hero: borrows its "one dose, one slot" idea as UI, not hardware.

Direction B: "Administration record"

Source sentence: care homes and wards track this exact job on a medication administration record (MAR), a grid of times against days, initialled per dose, with exceptions marked in red.

Register: utilitarian, dense. Motion: serious, fast.

Color: monochrome ink, with one hue reserved for one meaning.

There's no accent for "taken": a tick plus the time in the label colour. Red never appears without text ("Missed", "Late").

Type: system face. Weight does the hierarchy, and tabular digits make the time column the spine.

Shape: square-cornered grid cells in History (days × dose times). Today remains a list.

Motion:

• Productive only. Taking a dose ticks instantly with .sensoryFeedback(.impact(weight: .light)).
• Reduce Motion needs nothing to change.

Signature element: the week grid in History, also in the PDF export (Enhancement 6). Signature moment (behavioural): every dose is ticked with its real time, the same way, every time.

Voice samples:

• Button: "Take"
• Empty state: "Nothing due today."
• Error: "Reminders are off. Turn them on in Settings."

Landscape position:

• vs Apple Health: a clinical-record grid instead of a consumer list (unverified).
• vs Medisafe: an honest record without a caregiver-first framing.
• vs MyTherapy: no brand hue at all.
• vs Hero: a record-first model, not device-first.

How A and B differ: colour (amber state fill versus red exceptions only), shape (pockets versus grid), motion (expressive moment versus productive only), and layout idiom.

6. Type

Face: SF Pro (system), stock weights via text styles. Roles: UI, data (times)
Legibility: PASS. x-height n/a (not measured); confusables not measured (native, no specimen run); weights Regular, Semibold (headline); Bold Text yes for text styles, the literal .system(size: 13) also follows Bold Text but not Dynamic Type
Figures and features: tabular available (.monospacedDigit()), not used on dose times; distinct zero n/a (no codes shown)
Coverage: system, all scripts; locale issues: hard-coded 24-hour "%02d:%02d" times (finding 8)
Scaling: Dynamic Type ok except TodayView.swift:36 (finding 13); 1.4.12 n/a native
Delivery (web): n/a
Distinctiveness: default tier yes; identity carried by nothing
Verdict: Keep (system face is right for a utilitarian task flow). Tune: .monospacedDigit() on times, a text style instead of size 13
Evidence: source only; type-eval.js not run (native target, system face only)

7. Reworks

Rework 1: The schedule as a first-class, persistent object

• What changes:
  • Before: Medicine.times is copied once into today's Dose instances at add time. Nothing recurs, nothing reminds, nothing can be edited without orphaning doses, and deletion takes history with it.
  • After: a persisted Schedule (per medicine: times and active days) is the single source. Today's doses and the repeating notification triggers are derived from it. Logged doses are an append-only record keyed to the medicine, which is archived rather than deleted.
• Bar: Cluster. Findings 4, 5, 12 and 17 share one cause: there's no schedule entity, only one-shot dose instances. Per-row fixes (for example regenerating doses at midnight) leave the cause in place, and editing a time would still orphan doses.
• Evidence: findings 4, 5, 12 and 17; Models.swift:43-57. Apple Health and MyTherapy both treat recurrence plus notification as baseline (section 4).
• Cost:
  • Models.swift (new Schedule and persistence), plus TodayView, HistoryView, MedicinesView and the add/edit form.
  • A notification scheduler.
  • No learnability risk (the UI shape barely changes). There's no user-data migration, because nothing persists today.
• Unlocks: reminders, a second day that works, editing, honest history, and the widget, intents and PDF enhancements, which all read from the schedule.
• Smallest experiment: build the Schedule plus derived-today in a branch. Run it on a device for 3 days with 2 medicines and confirm that Today, notifications and History agree each morning.

Rework 2: Medicines becomes the medicine's home (remove the Add tab)

• What changes:
  • Before: four tabs, one of which is an action. The list rows are inert, and saving silently resets a form.
  • After: three tabs (Today, Medicines, History). Medicines has a + toolbar button that opens an "Add Medicine" sheet (Cancel / Add), an empty state with "Add Medicine", and rows that open a detail screen with Edit.
• Bar: Cluster. Findings 3, 7 and 10 share one cause: creating a medicine was split into its own tab instead of being an action on the Medicines collection. So the collection has no create path or empty-state action, and the form has nowhere to return to.
• Evidence: findings 3, 7 and 10 (and 17, which this rework's detail screen also resolves). Apple tab bar rule (SHOULD).
• Cost: RootView (drop one tab), MedicinesView (toolbar, empty state, navigation), AddMedicineView (becomes a sheet with dismiss), plus one new detail view. Learnability risk is low, because this is the standard iOS pattern.
• Unlocks: save feedback for free (the sheet dismisses onto the new row), an edit path, and a tab bar that maps to the product's three nouns.
• Smallest experiment: a throwaway prototype of the sheet flow. Ask 5 people to "add your evening tablet, then check it's there" and count hesitations against the current build.

Rework 3: Identity from the subject (after Reworks 1 and 2)

• What changes:
  • Before: platform defaults throughout, with nothing owned.
  • After: one direction from section 5 applied to Today, History, the empty states, notifications and the icon. Task flows stay standard.
• Bar: Identity. The verdict is default-assembled with 7 tells, and fixing tells one by one would only swap in the next default.
• Evidence: finding 19; section 5's channel table; the logo-off and swap tests are both empty.
• Cost: Today's row component, History's layout, one colour set, custom symbols, a haptic, and the app icon. It's held behind the gate until the Majors are fixed.
• Unlocks: a reason to choose Dosewise over the free built-in Apple Health medications feature, and a defence against 4.3 "indistinguishable" risk in a saturated category.
• Smallest experiment: mock Today in Direction A and Direction B with the same 3 doses. Show both to 5 people who take daily medication, and ask which tells them faster what's left today.

8. Enhancements

Evidence was strong enough for 6 of the 8 allowed.

Platform capabilities

1. Act from the reminder (Agency). Give the dose notification "Taken", "Skip" and "Remind in 10 Min" actions, mirroring Apple Health's watch actions (section 4). Logging then takes two inputs without opening the app, which matches the best peer.
2. Follow up once if unlogged (Responsibility). Send one follow-up reminder 30 minutes after an unresolved dose, as Apple Health does. Offer it per medicine, honestly labelled, and never repeated beyond that. Critical alerts need an entitlement, so treat that as a later decision.
3. Next dose at a glance (Flexibility). Add a Lock Screen and Home Screen widget showing the next dose and how many are left today, plus a Control or App Intent "Log Next Dose". The Today list is exactly the glanceable, time-bound state these surfaces exist for.

Experience and flows 4. Take a whole time slot at once (Simplicity). People on several medicines take them together at 08:00. Group Today by dose time with one "Take All" per group (and individual overrides), which cuts N taps to 1. Apple Health has a comparable bulk log (unverified).

Motion and feedback 5. A felt "taken" (Craft). Add .sensoryFeedback(.success, trigger: dose.status) on Take, a distinct lighter haptic on Skip, and an animated status change that respects Reduce Motion. Today, Take is the core moment and it's silent. This is the seed of either direction's signature moment.

Information architecture 6. A record worth showing a clinician (Purpose). Turn History into a per-day record of due time versus taken time (from resolvedAt, already stored) with a "Share Report" PDF via the share sheet. Medisafe offers PDF adherence reports (section 4, [S]). Dosewise's honest late-or-taken record is the version no studied peer was confirmed to show (unverified).

9. Unverified

• Rendering: no screens were rendered (no simulator). Light and dark appearance, layout, truncation and real tap-target sizes weren't seen. To verify, boot a simulator, run the app, and capture each tab in light, dark and AX5 with xcrun simctl.
• Contrast on iOS 26: the ratios use Apple's published system colour values for iOS 17/18 (#34C759, #248A3D, #1C1C1E). If the app is built with the iOS 26 SDK, re-measure on device screenshots with scripts/png_contrast.py.
• (26+) checks: Liquid Glass tab bar behaviour, tab label tint and icon appearances were skipped. The build SDK is unknown and the gate fell back to the iOS 17.0 deployment target. To verify, confirm the Xcode version or CI image.
• Possibly unfinished:
  • Findings 4, 5 and 6 (no reminders, no recurrence, no persistence) may be unfinished development rather than design decisions. They're rated Major under release_stage: development; data loss (finding 6) becomes a Blocker in a release build.
  • No app icon or asset catalog exists. That's a placeholder note in development, not a finding.
• Stress states:
  • The dose row at AX5, with long medicine names, under pseudo-localisation (+40%) and in RTL, was judged from source only.
  • Increased Contrast and Reduce Motion weren't seen. Reduce Motion is irrelevant today, because the app has no custom motion.
• Loading, error and offline: none exist for local data. Notification-permission denial has no UI path (finding 16). To verify, deny permission on a fresh install.
• VoiceOver: the actual name read for the forward.circle Skip button wasn't heard (finding 14). To verify, use VoiceOver on a device or Accessibility Inspector.
• Peer visuals: peer identity combinations (accent hue, typeface, layout) weren't checked visually. Section 5's landscape positions are (unverified).
• Third-party sources: the MyTherapy and Medisafe report claims come from third-party reviews [S], not from the products themselves.

10. Offer

I can fix these in code now, if you want (nothing has been edited):

• Contrast and colour tokens: 1, 2
• Dynamic Type and accessibility labels: 13, 14, 15, 24
• Copy and casing rewrites: 21, 22, 23
• Small structural fixes that don't need Rework 1: 3 (empty state), 8 (time picker), 9 (valid default time), 16 (move the permission request), 20 (prominence on due doses only), 25, 26

These need a design decision first:

• 11: undo versus editable status.
• 12: archive versus delete.
• 18: due/missed sections versus a subtitle.
• Rework 1 covers 4, 5, 6 and 17; Rework 2 covers 7 and 10; Rework 3 needs a direction pick (A or B).

Enhancements and reworks are your call. I can build an enhancement (Enhancement 1 or 5 are the cheapest), or a throwaway prototype of a rework's smallest experiment (Rework 2's sheet flow, or Rework 3's two Today mocks).

Optional input that would let me verify more:

• A booted simulator with the app running, to capture light, dark and AX5 and re-measure contrast from pixels.
• The Xcode version, to run or rule out the (26+) checks.
• Answers to the profile questions in the Summary.

Which would you like me to do?

<!-- model: claude-sonnet; skill commit: 7559367 -->

Loupe review: Dosewise (SwiftUI, iPhone)

Loupe is independent and not affiliated with Apple.

1. Summary

Reviewed: the whole app (4 screens, 7 Swift files, inline, no subagents), from source only (no simulator, no running build), against an inferred profile.

Verdict: the interface is tidy stock SwiftUI, but the product does not yet do its job. It is a medication reminder that sends no reminders, loses everything on relaunch, stops showing doses after the day they were added, and can silently erase a person's dose history.

Do first: the schedule-first model rework (Rework 1), then findings 1 to 4 (data loss and two measured contrast failures), then 5 to 9 (the Today screen cannot be trusted for a medicine you take twice a day).

Assumptions (no owner, so defaults)

• purpose: log and be reminded of daily medicine doses (inferred from the code and the empty-state copy). Audience: adults managing one or more regular medicines, glanced at several times a day, some older or low-vision (inferred from the category; unconfirmed).
• platforms: iOS, iPhone only (TARGETED_DEVICE_FAMILY = 1). release_stage: development (user said not yet released). So App Review 2.1/4.2 checks are not run, and dead controls are Major plus an Unverified line.
• register: utilitarian (frequent check-and-act task). Motion personality: quiet. Humor: none. Casing: sentence case in body strings, Apple title case on tab labels and titles (as inferred). Voice: plain, second person, no "we".
• Build SDK: Xcode 26 / iOS 26 SDK (project.pbxproj header comment and XCODE_VERSION_ACTUAL = 2600). UIDesignRequiresCompatibility is absent from Info.plist, so the (26+) gate passes and (26+) checks were run. Deployment target 17.0.
• direction: none, so section 5 proposes two. waived: 0 (no profile).
• .loupe.md was not written (per instruction). Questions I would have asked, recorded here, answered by the defaults above:
  1. Is the purpose "reminders plus logging", or logging only? 2. Who is the audience (self, caregiver, older adults)? 3. Is this going to App Review soon, and is a medical-device or clinical claim intended? 4. Should data sync (iCloud) or stay on device? 5. Any brand assets (icon, colors, typeface) that exist outside the repo? 6. Peers the team measures itself against? 7. iPad or Apple Watch intended?

Run notes

• Tools: file reads, scripts/contrast.py on source colors, WebSearch (2 queries) for the landscape. No Playwright (native source, no web). No simulator was booted or touched, so no global appearance or content-size changes were made. No files in the reviewed folder were edited. Only this report was written.
• Evidence: source files in dosewise/ (line numbers below). Contrast results: 4B5563 on FFFFFF 7.56:1; 4B5563 on 1C1C1E 2.25:1 (fail); 4B5563 on 000000 2.78:1 (fail); 34C759 on FFFFFF 2.22:1 (fail); 30D158 on 1C1C1E 8.42:1; secondary label rgba(60,60,67,.6) on white 3.44:1. System color values are from Apple's published iOS palette, not rendered; row backgrounds assumed to be the standard inset-grouped fills (white light, 1C1C1E dark).
• Landscape step used the web (third-party sites); see section 4.

2. Findings

Whole-app cap 40; 23 findings, none grouped; waived: 0.

(26+) checks run: standard TabView, List and NavigationStack adopt the system look; no custom glass, no colored tab labels, no custom bar backgrounds. No hits.

3. Trust

No deceptive design patterns found in what was seen (no confirmshaming, urgency, pre-selection, hidden costs or fake social proof). The unsupported reminder promise is reported as finding 5 (an accuracy defect, not a dark pattern).

4. Landscape

Category: personal medication reminder and dose log. Web used; sources are partly secondhand, marked [S].

• Apple Health, Medications. Why: the platform baseline for this job. Logging: Health, Browse, Medications, "+", Taken or Skipped, Done; forgotten doses can be added later; optional follow-up reminder if not logged after 30 minutes; optional Critical Alerts. Source: https://support.apple.com/en-gb/HT213351 and https://tidbits.com/?p=59847 (search results, 2026-10-26; page text not fully fetched).
• Medisafe. Why: leading dedicated app. Reminder lock-screen actions "Take All" and "Snooze All" with the medicine name withheld for privacy; wrong choice can be undone from the pill box. [S] https://www.techenhancedlife.com/research_content/how-use-medisafe-app/ (page updated 2023, checked 2026-10-26).
• MyTherapy. Why: challenger. Pop-up to take, skip or snooze 30 minutes; re-sends until answered. [S] https://healthify.nz/apps/m/mytherapy-meds-pill-reminder-app (not iOS-specific, 2026-10-26).
• Adjacent exemplar: not researched. Seed playbook only (Health): "rest treated as valid", AVOID guilt-based streak notifications (unverified).

Conventions to keep: Taken / Skipped vocabulary (Apple's own); a way to log late; reminder actions that do not need opening the app (Medisafe, MyTherapy [S]); undo of a wrong log (Medisafe [S]; Apple allows editing). Convergence to avoid: a pill-icon health-blue app with a progress ring and streak score; guilt copy (unverified). Frontier moves: follow-up reminder 30 minutes later (Apple); critical alerts for must-not-miss doses (Apple); snooze (MyTherapy [S]). Openings (from the subject): medicine is physically organised by day-part in a weekly pill organiser (dosette); no studied app (from what I read) structures Today that way. (unverified: only three products, none opened). Best peer step count for the core task "log a dose": Apple Health from Health open: Browse, Medications, "+", Taken, Done (5 steps). Dosewise: 1 tap on Today once the app is open (it has no reminder to open it). Peer gap: none; the rework bar's peer-gap condition is not met.

5. Distinctiveness verdict

Verdict: default-assembled. Platform default on every one of the nine channels (type: SF Pro text styles plus one fixed 13 pt size; color: system blue, system green and one stray hex; shape, motion, iconography, imagery, sound and haptics, layout idiom: stock inset-grouped lists and tab bar; voice: generic "Add a medicine to start getting reminders"). Authored: 0 of 9. Signature element: none (the one-tap Take is a candidate behavioural element but is not carried through). Signature moments: 0. Slop-catalog tells: below the cluster threshold (platform conventions are never counted). Logo-off: nothing identifies the product. Competitor swap: renaming the screens "Medisafe" looks no more wrong. For a utilitarian product, platform default controls are correct; the verdict flags only that identity is not carried by any channel yet. Gate: Blockers and Majors are open on every flow, so the directions below are held until after the fixes.

direction is empty, so two directions. Default check: the generic theme for this category is a teal or blue gradient, pill icon, progress ring. Both avoid it.

Direction A: Dosette (day-part compartments).

• Trace: the weekly pill organiser's morning, noon, evening, night compartments; amber pharmacy dispensing glass.
• Color roles: accent amber #9A4D00 (oklch .51 .126 55) light, #FFB547 (oklch .82 .149 73) dark: 6.11:1 on white and 9.68:1 on #1C1C1E. Taken: #1D6B4F (oklch .47 .088 165) light 6.43:1, #4FD1A0 (oklch .78 .135 165) dark 8.88:1; state also carried by word and glyph. Overdue: #B3261E (oklch .50 .178 29) 6.54:1 on white. Neutrals: system backgrounds. Derives from two inputs: amber hue and green hue.
• Type: system SF Pro for all text; candidate display face for the next-dose time only: Atkinson Hyperlegible Next (open license; trace: legibility for older or low-vision users). Not yet measured with type-eval.js; measure before adoption. Dynamic Type kept.
• Layout idiom (signature element): Today as four compartments; a taken dose fills its slot. Moment: Take fills the compartment (brief fade under Reduce Motion).
• Shape and motion: continuous-corner compartments, productive motion only.
• Voice samples: button "Take Metformin"; empty state "Nothing scheduled today. Add a medicine to see its doses here."; error "Reminders are off. Turn them on in Settings so Dosewise can alert you."
• Landscape position: differs from Apple Health's per-medicine log and from Medisafe and MyTherapy's reminder-list model by structuring the day.

Direction B: Label (prescription-label vernacular).

• Trace: the dispensing label on the box: name, strength, instructions.
• Color: warm paper #F6EFE2 (oklch .95 .019 83) in light with the same amber accent #9A4D00 (5.35:1 on paper); dark mode derived from the system dark set with #FFB547. Differs from A on color use and layout.
• Type: system text; tabular figures for times and strengths (.monospacedDigit()), no brand face.
• Layout idiom (signature element): every row prints "name, strength, time" as one label block; strength is never omitted (also fixes finding 8).
• Moment: a stamped "Taken 08:12" line appears in place of the Take button.
• Voice samples: button "Take"; empty "No medicines yet. Add one to start your list."; error as above.
• Landscape position: the most plainly documentary of the studied peers. Pick one, or keep platform defaults and adopt only the behavioural signature (strength always shown, one-tap Take).

6. Type

Face: SF Pro (system) via text styles. Roles: UI, reading, data (times)
Legibility: PASS (system face); fixed 13 pt time is a CONCERN for scaling, not shape
Figures and features: tabular available (.monospacedDigit); distinct zero n/a
Coverage: system; locale issues none seen; times hard-coded %02d:%02d in Add (finding 13)
Scaling: Dynamic Type issue: TodayView.swift:36 fixed size; Bold Text not seen (no simulator)
Distinctiveness: default tier yes; identity carried by nothing
Verdict: Tune (replace .system(size: 13) with .subheadline; use monospacedDigit for times)
Evidence: source only; type-eval.js not run (no web target)

Candidate face Atkinson Hyperlegible Next: not measured (offline from fonts); verdict deferred.

7. Reworks

Rework 1: Schedule-first model (persisted schedule generates doses and reminders).

• What changes: before, add writes one day's Dose objects into memory and nothing else. After, a persisted Medicine schedule generates each day's doses on demand (and their reminders), and logged doses are the only stored history, kept when a medicine is archived.
• Bar: Cluster. Findings 1, 2, 5, 6 and 7 share one cause: dose records are one-day, in-memory snapshots rather than the product of a stored schedule.
• Evidence: findings 1, 2, 5, 6, 7; Models.swift:43-52; no scheduling or storage code in the project.
• Cost: Models.swift, TodayView, MedicinesView, HistoryView, DosewiseApp; no new screen. Risk: store schema decision (SwiftData vs file) before any user data exists, which is the cheapest moment.
• Unlocks: Today is correct on day 2; reminders; missed state; history survives deletes; later widgets and Siri can read one source of truth.
• Smallest experiment: persist medicines plus a computed dosesFor(day:) and run it for a simulated week.

No other rework clears the bar. Moving Add off the tab bar (finding 12) is a fix. Peer gap not met (section 4).

8. Enhancements

Fewer than eight; evidence is thin without a running build.

Experience and flows

1. Take and Skip from the reminder itself (notification actions). Principle: Agency. Mechanism: a notification category with "Take" and "Skip" actions that call resolve; the name can be withheld on the lock screen like Medisafe [S]. Unlocks: the core job without opening the app; builds on finding 5 and Rework 1.
2. Today ordered around "what is due next". Principle: Simplicity. Mechanism: Overdue, Next, Later, Done sections; the next dose carries the one filled button. Traced to the 2-to-4-times-a-day pattern of the data model (times: [DateComponents]).

Platform capabilities 3. A Lock Screen or Control Center control that logs the next dose. Principle: Flexibility. Mechanism: App Intent "Take next dose" (the noun is Dose, the verb is Take). Source: seed platform capabilities in landscape.md (unverified for this app's OS target). 4. Offer a late log: "I took this earlier" with a time picker. Principle: Responsibility. Mechanism: Apple's Medications supports adding forgotten doses (source above). Unlocks an honest log, because resolvedAt already exists.

Motion and feedback 5. Success haptic plus a short fill on Take. Principle: Craft. Mechanism: sensoryFeedback; fade only under Reduce Motion.

Brand and identity 6. Carry one behavioural signature: strength and time always visible, one tap to log. Principle: Purpose. Traced to the label and dosette directions in section 5. Held until the Blockers and Majors above are closed.

9. Unverified

• Everything rendered: layout, Dynamic Type up to AX5 (finding 11 is from source), dark mode and Increased Contrast, Reduce Motion, tab bar and large-title behavior. Verify by booting the app in a simulator.
• Contrast for findings 3 and 4 is computed from known system and literal colors on assumed standard row fills; confirm with a screenshot and scripts/png_contrast.py.
• Notification delivery, denial and re-prompt behavior: nothing is scheduled (finding 5, possibly unfinished given development); verify after scheduling is built.
• The "Add time" button inside the form row (tap behavior next to a text field), the swipe-delete gesture, Voice Control and VoiceOver reading order: need a device or simulator.
• Pseudo-localization, long names and RTL: not run. The code has no String(localized:) and no localization resources. Long medicine names are only limited by default wrapping.
• Midnight rollover, time-zone and daylight-saving changes for scheduled doses.
• Pending doses and the Today empty/loading states over time (no loading states exist; storage is in memory).
• Landscape: step counts for Medisafe and MyTherapy [S], and the adjacent exemplar, were not confirmed from primary sources. iPad and Apple Watch were not reviewed (not targeted).
• App Review (2.1, 4.2, and medical-app expectations) not assessed (development).
• Fonts in section 5 not measured.

10. Offer

Fixable in code now, grouped:

• Data safety: 1 (needs the storage choice), 2, 9, 17.
• Contrast and accessibility tokens: 3, 4, 10, 11, 16.
• Today row content and layout: 7, 8, 15, 22 (7 needs a decision on the grace window for "missed").
• Add flow: 12, 13, 18, 20 (12 needs a design call: sheet from Medicines).
• Empty states and copy rewrites: 5 (copy only), 14, 19.
• Larger: 5, 6 (scheduling) need Rework 1 first. 21, 23 are trivial.

Enhancements and reworks need your call: say if you want Rework 1's smallest experiment prototyped, or any enhancement built. Which should I do?

Optional extra input that would let me verify more: a booted simulator with the app (for screenshots, AX5, dark mode, Increased Contrast), the app icon and any brand assets, and answers to the seven questions in section 1.

<!-- model: claude-haiku; skill commit: 7559367 -->

Loupe review: Dosewise (SwiftUI, iPhone, development build)

1. Summary

• Reviewed: whole app, inline (5 screens, 7 Swift files). Input: native source only (dosewise/*.swift, Info.plist, project.pbxproj). No simulator, no running build, so every rendered check is Unverified or judged from code.
• Verdict: a functional prototype built almost entirely from stock SwiftUI parts. Two Blockers: dark-mode text contrast on taken doses, and swipe-delete that silently erases a medicine's dose history. The main gaps are missing features the screens imply (reminders, editing, persistence) and an Add flow with no feedback or proper time entry.
• Most important: fix the two Blockers, then decide whether reminders are in scope for this build, since the Today empty state promises them.
• Rework: No rework warranted. The defects are fixable per row, and the missing pieces are features (enhancements), not a structural cluster or a wrong core flow.

Assumptions (no owner answers; no .loupe.md, not written):

• release_stage: development (default). App Review checks (4.2, 2.1) are not gated; placeholder-like copy is noted in Unverified.
• Build SDK: Xcode 26 / iOS 26 SDK (project.pbxproj line 2 comment Toolchain: Xcode 26 (iOS 26 SDK)). UIDesignRequiresCompatibility absent from Info.plist, so the (26+) gate passes and (26+) checks were considered. The app uses stock components, so there is no custom Liquid Glass to flag.
• Deployment target iOS 17.0; TARGETED_DEVICE_FAMILY = 1 (iPhone only); portrait only.
• purpose: personal medication dose tracker (inferred from MedicinesView, TodayView, HistoryView).
• register: utilitarian (frequent check-and-act health task).
• humor: none. voice: plain, sentence case, second person. direction: empty, so distinctiveness gives a verdict and offers directions.
• density: comfortable (iOS 17 pt body).
• audience: an individual managing their own medicines, several times a day.
• platforms: iPhone.

Run notes

• Tools used: Bash (file listing, reading, contrast arithmetic in Python with -I), WebSearch (1 query). No Playwright or simulator (native source only). No scripts/measure*.js or scripts/png_contrast.py runs, since there is no rendered page or screenshot.
• Contrast figures were computed from the hex values in source (WCAG relative luminance formula) against assumed backgrounds. They are not measured on a device.
• Landscape web research was done. Skill files read: SKILL.md, references/report-template.md, workflow-intake.md, workflow-passes.md, augment/red-team.md, augment/landscape.md, plus targeted greps of references/apple/*.md.
• Side effects: none on the user's files. Nothing edited. The only file written is this report.

2. Findings

Trust: No deceptive patterns found in what was seen. Finding 5 is a false capability claim, not a dark pattern. It is listed as Major, not as a Trust-gate Blocker.

waived: 0 (no profile waivers).

3. Trust

No deceptive patterns found in what was seen. The reminders copy in finding 5 is an inaccurate description of a feature that does not exist yet. Rewrite: "Add a medicine to see today's doses."

4. Landscape

Category: personal medication reminder and dose tracking, iOS.

Studied (sources checked 2026-10-26; from search snippets, not full page fetches, so landscape claims are not fully verified):

• Medisafe Pill Reminder (medication reminders, drug-interaction database). Source: https://apps.apple.com/th/app/id573916946 (store page, checked 2026-10-26). A 2026 roundup calls it the largest name in pure medication reminders: https://digitalhealth.folio3.com/blog/?p=15182 (checked 2026-10-26; secondary, third-party roundup).
• MyTherapy Medication Reminder (reminders plus a health diary to share with a doctor). Source: https://apple.co/2FB72Zp (App Store, checked 2026-10-26). Reviews praise its reminders; a recurring complaint is that family members cannot see which medicines were taken or missed (secondary roundup and reviews, (unverified) detail).
• Apple Health (adjacent exemplar: no-cost health log). Source: https://digitalhealth.folio3.com/blog/?p=15182 (same roundup; secondary).
• Round Health: not found in search. Not studied.
• Caveat: https://bearable.app/?p=19925 is a comparison by Bearable's own team, disclosed as such; not used for ranking.

Four lists (landscape step, (unverified) where marked):

• Conventions to keep: a dose list ordered by time with Take and Skip actions ((unverified) from roundups); a history log of taken and skipped doses.
• Convergence to avoid: none measured. The product is already the stock template, which is itself the risk in the Distinctiveness verdict.
• Frontier moves: reminders delivered as notifications with Take and Skip actions (Apple notifications allow up to four actions, components-presentation.md:419); an adherence summary.
• Openings: sharing a schedule and taken status with a caregiver. MyTherapy's reviews name the caregiver gap as a complaint ((unverified)); this fits the product's audience of people managing other people's medicines. Do not add it without a product decision.
• Core-task step count (peer): not sourced. The rework bar's peer-gap condition is not met.

5. Distinctiveness verdict

default-assembled. Evidence:

• Channels: platform default (stock TabView, NavigationStack, List with .insetGrouped, ContentUnavailableView, .borderedProminent, SF Symbols throughout). Library default: none. Trend default: none. Authored: only the hard-coded #4B5563 (Models.swift), which is the one custom color and breaks dark mode (finding 1).
• The slop-catalog was not counted tell by tell (no rendering), so no numeric tell count is given. The verdict rests on the channel review: one authored choice, and it is a defect.
• Logo-off and competitor-swap tests need rendered screens: Unverified.
• No direction is set in the profile, so directions are offered below.

Directions (2, for the user to pick; none adopted):

• A. Calm clinical: neutral system palette, one accent (a teal or blue) used only for the due dose and the Take action, system text styles, minimal chrome. Differs from default on: accent role, density of the Today screen (grouped Due / Done sections).
• B. Quiet routine: schedule-first layout where the day reads as a timeline with dose ticks; accent reserved for taken. Differs on: layout idiom (timeline vs list) and motion (one restrained tick animation on Take).
• Neither is a look to copy from a peer. Both must be checked against the product's audience and the Blocker fixes first.

6. Type

System font only (no UIAppFonts in Info.plist). One fixed size: .system(size: 13) in TodayView.swift (finding 8). Other text uses built-in styles (.headline, .subheadline, .caption, default). The type-eval script was not run (no rendered page). Unverified.

7. Reworks

No rework warranted. Reason: the defects are fixable per row (findings 1 to 16), and the largest gaps (reminders, persistence, editing) are missing features, which are enhancements under the rework bar's premise rule. No root cause is shared by three or more findings that per-row fixes cannot remove.

8. Enhancements

Up to 8. Fewer than that, because the evidence is thin without a rendered build.

Experience and flows

1. Reminders as notifications with Take and Skip actions. Principle: Responsibility. Product hook: the Today empty state already promises reminders; the Take and Skip actions match the existing row actions. Mechanism: one UNNotificationRequest per dose with two actions; request permission in context after the first medicine is saved. Unlocks: the app does what its copy says, and doses can be taken without opening the app. Peers: frontier move (roundup sources above, (unverified)).
2. Due and Done grouping on Today. Principle: Simplicity. Product hook: today's list mixes pending, taken and skipped doses sorted only by time (TodayView.swift). Mechanism: two sections, "Due" (pending, sorted by time) and "Done" (taken or skipped). Unlocks: the next dose is the first thing seen.
3. Haptic on Take. Principle: Craft. Product hook: the Take button gives no feedback beyond the row changing. Mechanism: .sensoryFeedback(.success, trigger:) on the taken state. Unlocks: confirms the action without another screen.

Information architecture 4. Adherence summary at the top of History. Principle: Purpose. Product hook: History lists rows with no totals. Mechanism: a single line such as "This week: 12 taken, 2 skipped" above the list. Unlocks: the question "am I on track" is answered without scanning rows.

Content and voice 5. Dose time shown as a single, locale-aware string everywhere (see finding 12). Principle: Familiarity. Mechanism: .time style in Add and Medicines as well as Today. Unlocks: consistent time format with the user's system setting.

Platform capabilities 6. Home Screen or Lock Screen widget for today's next dose. Principle: Flexibility. Only once finding 5 is done and persistence exists. Unlocks: a glance at the next dose without opening the app. Landscape: not checked beyond the roundups. 7. Editable medicines (tap a row to edit name, strength, times, notes). Principle: Agency. Product hook: the only way to correct a medicine is to delete it, which loses history (finding 2). Unlocks: corrections without data loss. 8. Caregiver view of today's doses. Principle: Responsibility. Opening only; needs a product decision and a privacy review before any design. Not proposed as a build item.

9. Unverified

• Dark-mode contrast of finding 1: the background (#1C1C1E) is assumed. Verify with Accessibility Inspector on a dark-mode device or simulator; ratios would change if the cell background differs.
• Light-mode contrast of finding 3: system green RGB recalled from memory. Verify on a device; if it fails, upgrade finding 3 to Blocker.
• Skip and Take hit area: the .borderless icon buttons have no explicit frame, so they may be smaller than the 44x44 pt control size (content-a11y.md:31). Verify with Accessibility Inspector frames.
• Dynamic Type at AX5: TodayView DoseRow uses an HStack with two buttons and a long medicine name. Truncation or overlap is unseen. Verify by setting the largest accessibility text size on a device.
• Empty, error, loading and offline states: no loading or error state exists in source (synchronous in-memory store). Empty states for Today and History were read from code and not seen rendered. Verify by launching with no medicines.
• Dark mode, increased contrast, reduced motion, RTL, pseudo-localized text, narrow and very wide widths: not seen (no simulator). Verify by running the app.
• Transient feedback: none in source for Save, Take or Skip (no toast, haptic or animation). Verify by taking a dose on a device and watching for a visible change.
• Notification delivery: no scheduling code exists, so delivery cannot be verified from source. Verify once finding 5 is implemented.
• Past-due behavior (finding 13): read from add(_:) and the Today filter; not exercised. Verify by adding a medicine with a time earlier than now.
• App Review (4.2, 2.1): not gated (release_stage: development). The reminders copy would be a placeholder-like issue at release. Re-check before submission.
• Landscape: based on search snippets and roundups (some from a vendor's own comparison, disclosed). Step counts for peers were not sourced, so the peer-gap rework condition is not met.
• Localization: no .strings or .xcstrings in the project. Localization not reviewed.

10. Offer

Fixable in code now, if the user picks findings:

• Contrast tokens: findings 1 and 3.
• Copy rewrites: findings 5 (empty-state line), 15 (validation footer text), 16 (status text).
• Accessibility labels and text styles: findings 6 and 8.
• Add flow and navigation: findings 9, 10 (add button, empty state, schedule summary), 11 (confirmation after save), 12 (time picker), 14 (remove times).
• Data loss fix: finding 2 (choice needed: keep history on delete, or confirm the loss).

Design decisions needed first:

• Finding 2: keep history after deleting a medicine, or confirm the loss? (Affects the data model.)
• Finding 4 and enhancement 1: persistence and notifications are scope decisions, not bug fixes.
• Finding 13: Missed state wording and visual treatment.
• The two directions in section 5.

Enhancements and reworks: the user decides. Enhancements 1 to 8 are offered as options. No reworks.

Optional input that would let the skill verify more: a simulator build or a booted device (for dark mode, Dynamic Type and hit areas), and the Xcode project's .xcassets color sets if any exist (none were in the source provided).

Not written: no .loupe.md. No files in the user's project were edited.