You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Feature]: Larger weight-selection font (2-3x) for low-vision / older-adult users #1367
The weight-selection text is too small for older adults and low-vision users. A training business using the machines with older trainees requests a larger weight-selection font — "twice to three times the size" — so weight changes can be made safely without leaning in or misreading values.
Desired Behavior
A weight-selection font that can be set (or defaults) at roughly 2–3× current size. Scope to be confirmed in feasibility: dedicated "large text / accessibility" mode vs. scaling only the weight-selection surfaces vs. following OS text-size settings.
Target users
Older adults in supervised strength training (rehab/senior fitness settings); safety-critical value reading.
Original Request (verbatim, relayed by owner)
Would it be possible to pay you to make 1 feature work for us to train older adults more safely?
The current stop button is great! But once clicked it moves you back to the general screen instead of just stopping the weight and staying in whatever lift mode you were in: Standard, Echo, Etc.
Could you make the stop button something that just turns off the weight?
Would it also be possible to make the weight selection font larger? Like twice to three times the size?
Please let me know how much would be appropriate for this kind of change. Again, immensely appreciate you work. The machines have helped us help so many regain strength and this kind of feature which used to work on android no longer does.
Note: this request bundles two asks. This issue tracks the weight-selection font; the stop-button behavior is tracked in the sibling issue linked below.
Feasibility assessment — weight-selection font (2–3×)
Recommendation: accept. Assessed against current main (mobile app, shared Android/iOS UI).
Product value: high for the target use — supervised senior/rehab strength training, where reading the weight value correctly before commanding load is safety-critical. Also improves at-a-glance readability for everyone during exertion.
Technical feasibility: high. This is a presentation-only change in the shared mobile UI — no machine protocol, data-model, backend, or portal impact. The app already has an accessibility seam (bold text / reduce motion) and already adapts layouts to the OS text size, which de-risks the work.
Surfaces covered: pre-set weight selection (set-ready screen), the in-lift weight readout (bottom bar, all lift modes), the weight stepper in exercise config/edit, and the Just Lift weight cards. Secondary weight displays to be confirmed in design.
Risks: 2–3× text must extend the existing compact-layout behaviour on small screens, and app-level scaling must be clamped against the OS text-size multiplier so they do not compound (e.g. 3× in-app on top of 2× OS).
Baseline reviewed: main @ 8ef4d59c55943291393142179194ccae31728ff8
Date: 2026-10-31
Scope decision
This issue covers larger, user-configurable weight values on safety-critical weight-selection and
weight-readout surfaces in the phone app. It does not cover the separate stop-button request in
issue #1366.
No reporter clarification is required. The request, feasibility evidence, current UI seams, and
existing app patterns are sufficient to choose the architecture. The assumptions below are
intentional and must be confirmed in the prototype stage.
Architecture decision
Adopt a device-global, app-level weight-text-size preference with three choices:
Standard — existing type styles (1×).
Large — approximately 1.5× the existing weight-value type size.
Extra large — approximately 2× the existing weight-value type size.
This is deliberately more conservative than the reporter's literal "twice to three times" request. sp units already respond to the operating system's text-size setting. A further 2–3× app
multiplier would compound with the OS multiplier and can make the safety-critical value illegible
or clipped (for example, 2× app setting with an OS scale of 2× can become an effective 4×). The
prototype stage should show Standard, Large, and Extra large at OS scales of 100%, 150%, and 200%;
if the review still considers 3× warranted, it may add a distinct 3× choice while retaining
the same composition and clamping contract.
The preference is global to the phone app, not profile-specific. Font size is a device/display
accessibility preference, not a training profile or cloud-synced training attribute. This keeps the
change out of SQLDelight, the profile-preference sync codec, portal APIs, and backup schema.
System shape
1. Preference domain model
Add a small, closed domain type rather than storing an arbitrary float:
The enum is the persistence contract. Invalid or unknown persisted values must normalize to STANDARD; this keeps forward/backward compatibility deterministic.
2. Persistence and state flow
Store the selected mode in the existing global SettingsPreferencesManager using a new key (for
example, weight_text_scale) and expose it through UserPreferences.weightTextScale.
State flow:
Settings screen
-> MainViewModel.setWeightTextScale(mode)
-> SettingsManager.setWeightTextScale(mode)
-> SettingsPreferencesManager persists key + emits UserPreferences.preferencesFlow
-> SettingsManager.userPreferences overlays global preferences
-> EnhancedMainScreen collects the value
-> CompositionLocalProvider supplies LocalWeightTextScale
-> weight-value composables resolve their final TextStyle
The existing UserPreferences facade is the right transport because it is already observed by the
app shell and already separates global preferences from profile-scoped training preferences. SettingsManager.overlayProfile() must not overwrite the new field, so it remains global.
Do not place this field in CoreProfilePreferences. That model is stored in the SQLDelight UserProfilePreferences table and encoded/decoded by the portal profile-preference sync codec;
adding a field there would create schema migration, portal payload compatibility, dirty/generation
handling, and cross-device sync work without user value.
3. Composition seam
Add a phone-level composition local, adjacent to the existing accessibility seam:
EnhancedMainScreen should provide the selected value once for the phone shell. Components can
also read it directly with a default of STANDARD, so isolated previews/tests and any screen
rendered outside the shell remain safe.
This is separate from PlatformAccessibilitySettings, whose current fields are platform-owned
(boldTextEnabled, reduceMotion). The new setting is app-owned and persisted, so mixing the two
would blur ownership and complicate the Android/iOS expect/actual implementations.
4. Style resolution helper
Create one pure/shared helper that maps a base TextStyle and the selected scale to the final
weight-value style. It must:
Preserve the base style's font family, weight, letter spacing, line height, and color.
Scale the base fontSize using the preference factor.
Apply a bounded effective scale so the OS font multiplier cannot produce unbounded growth.
Return a style suitable for weight values, not a global typography override.
Recommended contract:
fun TextStyle.scaledForWeightValue(
scale:WeightTextScale,
osFontScale:Float,
): TextStyle
with a capped effective factor. For example, cap the product of the app factor and OS font scale at
2.0–2.5. The exact cap is a prototype decision, but it must be deterministic and unit-tested. The
helper should prefer an explicit sp size derived from the base style rather than multiplying a
Compose sp value by the OS font scale again, because sp already incorporates OS scaling. In
other words: resolve the final physical size once, then let Compose apply the OS scale exactly
once.
5. Affected surfaces
The v1 contract covers the primary selection and live-readout surfaces:
Surface
Component/path
Current value style
Required v1 behavior
Pre-set weight selection
SetReadyScreen.kt -> SliderWithButtons
titleMedium
Scale the formatted weight value; keep the label readable but not scaled to the same magnitude.
Routine overview weight selection
RoutineOverviewScreen.kt -> SliderWithButtons
titleMedium
Same helper and contract as Set Ready.
Live workout readout
WorkoutHud.kt -> HudBottomBar
titleLarge
Scale the numeric weight. The Echo-mode word Adaptive and bodyweight No machine load use the same emphasis treatment because they occupy the safety-critical value slot.
Exercise config/edit stepper
components/WeightStepper.kt
headlineMedium
Scale the numeric value. The per-cable label and two-cable total remain supporting text.
Just Lift weight cards
JustLiftScreen.kt
mixed title styles
Scale the weight value within each card; do not scale the card's action label ("Finish Set").
Secondary displays (recommendations, bulk adjust, quick-history columns, set summaries) are not in v1. They are not selection surfaces, and enlarging them adds layout risk without
directly answering the request. They may be added later after the primary surfaces are validated.
6. Responsive-layout contract
The enlarged value must never be clipped or pushed into an unreadable layout. The current SliderWithButtons only stacks its label/value header when OS fontScale >= 1.15; the new app
scale must participate in that decision. Likewise, isCompactAccessibilityLayout() and the
Just Lift card stacking helper must consider the effective app scale.
Required behavior:
On compact widths, stack the slider label above the value when the effective weight-text
scale requires it, even if the OS font scale is 1.0.
The formatted value must remain fully visible (no ellipsis, clipping, or overlap).
WeightStepper must keep both +/- controls reachable; if the value cannot fit beside them at
Extra large, the value moves to its own row while the controls remain on one row.
The HUD bottom bar must allow its value block to consume the available width without colliding
with other bottom-bar content.
Just Lift cards must switch to their existing stacked layout when the enlarged value would
otherwise crowd the card.
Architecture should introduce a shared effectiveWeightTextScale / layout-pressure input rather
than duplicating fontScale >= 1.15 checks in each surface.
API, backend, and schema boundaries
Mobile shared UI: changes are required in shared/src/phoneMain and are compiled for
Android and iOS.
Persistence: one new global Settings key; no SQLDelight schema change.
Profile sync/portal: no changes.
BLE/machine protocol: no changes. This must not alter weight values, increments, units,
rounding, or commands.
Backups: if the app's global UserPreferences backup codec serializes the entire data class,
the new field must be backward-compatible (default STANDARD when absent). No backup version bump
is required if the existing codec already tolerates unknown/missing fields; the integration-model
stage must verify that exact codec path.
tvOS: no change. TvCoreComposition and TV-only UI remain untouched.
Compatibility and migration
Existing installs default to STANDARD, so the shipped UI is unchanged until a user opts in.
The new preference is absent from older backups; decoding must fall back to STANDARD.
Unknown enum names or corrupt stored values normalize to STANDARD rather than crashing.
No machine, routine, session, exercise, profile, or cloud data is migrated.
OS text-size changes remain supported. The app setting composes with OS scaling through the
bounded helper, not by silently replacing the OS behavior.
Rollout and flag strategy
This is a low-risk, user-initiated display preference, so it does not need a remote feature flag.
Recommended rollout:
Ship the preference and v1 surfaces with STANDARD as the default.
Expose the choice in the phone app's global Settings screen under an accessibility/display
section, using a segmented control or radio group with plain labels (Standard, Large, Extra
large) and a short description of the effect.
Apply changes immediately without restarting the app.
Do not pre-enable it for all users; the reporter can be invited to test the Extra-large option
after release.
If implementation discovers that Extra large cannot fit a required surface without changing
navigation or control layout, that surface should cap the factor at the largest size that remains
fully readable and visibly explain the cap in the settings description rather than silently
clipping.
Accessibility semantics
Enlarging text must not replace semantic labels. Weight controls and readouts must retain
meaningful accessibility text (value, unit, increment/decrement actions).
Do not announce a new spoken number for every composable separately; avoid duplicate semantics
between the visible value and any hidden copy.
Touch targets remain at least 48dp; larger text must not shrink buttons.
The setting must be keyboard/screen-reader reachable and use the same Material semantics as the
rest of Settings.
Analytics and telemetry
No new telemetry is required for correctness. If the project already has a privacy-approved local
analytics/event mechanism, a privacy-minimal event such as weight_text_scale_changed with the
selected enum name (no user identifier, no workout data) can be considered. Do not add a remote
analytics dependency for this feature.
Whole-app large-text mode or replacing OS accessibility settings.
Profile-specific or cloud-synced font preferences.
Changes to weight units, increments, machine limits, BLE commands, or saved workout data.
Portal changes.
tvOS changes.
Any pricing or fee language; the owner develops Phoenix features voluntarily and will answer the
reporter's cost question separately.
Acceptance criteria
A user can select Standard, Large, and Extra large from phone Settings.
The preference survives app restart and is applied immediately after selection.
Standard reproduces the current weight-value typography on every v1 surface.
Large and Extra large enlarge the weight value on Set Ready, Routine Overview weight selection,
the workout HUD readout, WeightStepper, and Just Lift weight cards.
The value remains fully readable and unclipped at each option on a compact-width phone, a
short-height phone, and a tablet, including OS font scales of 100%, 150%, and 200%.
The app setting and OS font setting do not produce unbounded compounding; the documented cap is
enforced and unit-tested.
Weight controls remain operable, with no loss of +/- slider functionality or touch-target
accessibility.
Echo mode ("Adaptive"), bodyweight mode, and numeric mode retain a consistent value emphasis.
No change occurs to weight values, units, increments, BLE traffic, saved routines, sessions,
profile data, portal sync payloads, or tvOS behavior.
Missing or corrupt persisted preference data loads as Standard without a crash.
Automated coverage includes preference normalization, effective-scale clamping, surface wiring,
and compact-layout/stacking behavior; Android and iOS builds compile.
A visual prototype demonstrates the three options at representative OS scales before
implementation is finalized.
Downstream hand-off
Integration model must map
UserPreferences field and global preference key.
SettingsPreferencesManager load/set methods.
SettingsManager and MainViewModel delegation.
SettingsGlobalUiState and SettingsTab/NavGraph wiring.
EnhancedMainScreen composition provider.
The style-resolution helper and every v1 surface.
Backup codec compatibility for the new optional field.
Existing guard tests to extend and new unit/host tests to add.
Prototype must show
Settings control with Standard / Large / Extra large.
Set Ready slider at each option.
Workout HUD bottom bar at each option.
WeightStepper at each option.
Just Lift cards at each option.
At least one small/compact viewport and one OS-scale-200% comparison.
Explicit confirmation of the 1× / 1.5× / 2× mapping and the bounded effective-scale cap.
Implementation risk checkpoints
The repository's current main already contains the relevant accessibility and compact-layout
seams, but layout behavior is subtle on iPhone landscape/short-height devices. Keep the
implementation packet bounded to one shared helper, the preference wiring, the v1 surfaces, and
focused tests.
Do not use the existing SliderWithButtons component for rep sliders if the enlarged weight
behavior would affect those values. Add an explicit weight-value treatment or parameter rather
than globally changing every slider value.
Preserve the existing WeightDisplaySourceGuardTest and related guard tests: this feature changes
presentation only and must not alter the source of the displayed weight.
Integration Model — larger weight-selection font (stage 3 of 5)
Baseline: main @ 8ef4d59c (identical to origin/main at review time).
This stage maps the architecture onto the actual codebase at the current main. It confirms the
architecture's persistence design works as written, and it surfaces two corrections to the
surface list — both narrowing the work rather than widening it.
Corrections to the architecture's surface map
1. There are three weight sliders, not two. The architecture's table listed Set Ready and
Routine Overview. RestTimerCard.kt:577 has a third weight slider using the same formatWeightWithUnit helper as the other two. It is a weight selection surface (the user edits
the next set's load from the rest timer), so leaving it out would give the reporter large weight
text while planning a set and small weight text while resting. Recommend including it.
The full SliderWithButtons census is 8 call sites — 3 weight, 5 rep — all in phoneMain, none
shared with tvOS. The 5 rep sites and 3 weight sites are cleanly separable, and the component has
no value-text style parameter today, so adding one with a default that preserves the current titleMedium keeps the rep sliders untouched.
2. The Just Lift weight value is a native picker, not Compose text. The architecture says
"scale the weight value within each card". The Just Lift "Weight per Cable" card calls CompactNumberPicker, which is an expect/actual:
Android renders a native NumberPicker widget — its numeric text is sized by the OS widget, and NumberPicker exposes no text-size setter.
iOS renders a UIPickerView wheel whose Compose-side selected-value style is chosen inside CompactNumberPicker.ios.kt:96-127.
The expect signature has no typography parameter, so scaling that value is not implementable
through the existing API. Recommended v1 treatment: keep the picker as the input control and add an
enlarged Compose readout of the current per-cable weight in the same card. The prototype stage
can decide this (Option A) against deferring Just Lift (Option B).
A layout constraint the architecture's plan would have violated
The architecture asked that Just Lift cards "switch to their existing stacked layout when the
enlarged value would otherwise crowd the card". shouldStackWeightCards
(WindowSizeClass.kt:124-126) deliberately stacks only on short-height windows. Its KDoc
explains why: on iPhone portrait, stacking puts the iOS picker wheel directly above the ProgressionSlider, and the wheel's vertical LazyColumn was observed to claim drags intended for
the slider — the readout jumped to neighbour weights like 29/30/44-49 even though that slider's
range is -10..+10.
Integration model therefore does not change shouldStackWeightCards. Because the Just Lift
value is a native picker, the enlarged readout is Compose text that can take its own line inside
the existing card without changing stacking mode. If it still cannot fit at Extra large, the
fallback is to cap that surface's factor and say so in the settings description — the
architecture's own rollout rule — rather than restack and re-introduce the gesture conflict.
The backup question, resolved
The architecture asked the integration stage to verify whether the global UserPreferences backup
codec serializes the whole data class. It does not.
BackupData (BackupModels.kt:676) holds a BackupContent containing only profile/training
entities — userProfiles, profilePreferences, workoutSessions, routines, personalRecords, cycleProgress, gamificationStats, and so on. There is no global
app-preference field.
backupPrivacySummary() (:703-714) explicitly states the backup does not include preference
sync bookkeeping, voice phrase or calibration, or adult consent and prompt state.
A repo-wide search for UserPreferences across any backup/export/import/serialize path returns
no hits outside production UI/VM code and tests.
Global settings live in multiplatform-settings (SharedPreferences / NSUserDefaults) and are
device-local, outside backups entirely. Consequences: no backup codec change, no backup version
bump, no migration. The "absent from older backups → decode as STANDARD" concern is moot; the
real guarantee is "the key is absent from the settings store → STANDARD". The practical consequence
worth stating in the release note is that a device restore leaves this display preference alone.
Verified state seam
Every seam the architecture named exists, and one nearly identical precedent makes the design
low-risk:
Seam
Location
UserPreferences field
domain/model/UserPreferences.kt:8-73
Storage key + load
data/preferences/PreferencesManager.kt:111-165, 170-241 (SettingsPreferencesManager lives inside this file)
Normalization precedent
BleCompatibilitySetting.fromStorage(...) (:230-236) — unknown value → default, never throws
presentation/screen/SettingsTab.kt:279-309 — SingleChoiceSegmentedButtonRow enum pattern already used twice
Composition provider
presentation/screen/EnhancedMainScreen.kt:344-351
The field is deliberately not placed in CoreProfilePreferences, so there is no SQLDelight
schema change, no portal sync payload change, and no backup change. No BLE, machine-protocol,
tvOS, or portal impact.
Style helper
Two pure functions rather than one, so each is independently unit-testable:
TextStyle.scaledForWeightValue(scale: WeightTextScale): TextStyle — scales the base fontSize
in sp. Because Compose applies the OS fontScale to sp at render time, the OS scale is
applied exactly once, by Compose.
effectiveWeightTextScale(scale: WeightTextScale, osFontScale: Float): Float — the capped
product, used for layout decisions.
Recommended cap: 2.0 on appFactor × osFontScale. The largest base weight style is headlineMedium (28sp), so 2.0 → 56sp, which is already at the practical ceiling for a value on a
360dp phone; 2.5 → 70sp would force stacking everywhere. The prototype confirms the number; the
implementation keeps it as a single named constant.
The effective scale must make the slider header stack even when the OS font scale is 1.0 — that
is the architecture's explicit requirement.
Test plan highlights
New unit tests: enum normalization (unknown/corrupt → STANDARD, never throws); the scale
helper (Standard reproduces today's fontSize exactly; font family/weight/spacing/line-height/
color preserved; result in sp); the cap matrix (2.0 app × 2.0 OS → capped, not 4.0);
persistence (absent key, corrupt value, set/reload, flow emission); overlayProfile leaves the
field global across a profile switch.
New highest-value guard test: a source-text guard asserting the 3 weight SliderWithButtons call sites carry the treatment and the 5 rep sites do not. This is the
regression guard for "must not affect rep sliders".
Contract test to extend:ProfileSettingsSeparationContractTest asserts the Settings
destination collects exactly 3 flows and asserts each SettingsTab(...) assignment string.
The new preference must ride the existingglobalSettings flow — adding a fourth collectAsState() would break it. The two new wiring strings must be added in the same PR.
Confirm the 1× / 1.5× / 2× mapping and the effective-scale cap (recommended 2.0).
Choose Just Lift Option A (Compose readout — specified) or Option B (defer).
Confirm the cap at which the HUD bottom bar and WeightStepper still fit on a 360dp phone at
OS 200%.
Decide whether a 3× option is warranted — the reporter asked for "twice to three times".
Adding a 4th enum constant is source-compatible; the persisted contract is the enum name
string and unknown values still normalize to STANDARD.
Confirm settings placement (Appearance card) and plain-language labels.
Confirm the release-note line that the preference is device-local and not restored by a backup
restore.
Full integration model (file/module map, per-surface excerpts, test plan, risk table) is recorded
in the enhancement artifact directory.
Weight-text prototype — capture blocked, not approved
The design and integration inputs are complete. This stage requires screenshots of the proposed larger weight values inside the real phone app; no prototype screenshot or visual approval is being claimed yet.
The required Phantom launcher was executed and refused the checkout before a build or simulator launch. Its current contract rejects all uncommitted changes, whereas this prototype workflow explicitly requires temporary UI changes that are reverted, not committed. The shared harness also contains existing unrelated changes, which were left untouched. Current mobile main lacks the launcher's exact required simulator fixture, so switching to a clean current-main checkout is not a supported workaround.
Next step: repair/authorize a prototype-compatible capture path, then render the Settings size choice and weight surfaces across the required phone/tablet and text-size matrix. The proposed Standard / Large / Extra large sizes and effective cap remain unvalidated, not signed off. One implementation caution: capping only the layout scale does not cap rendered text; the text-style calculation must also account for the operating-system scale.
This affects only the font request. The separate stop-button request #1366 is unchanged. The issue remains open; implementation/signoff has not been unlocked by this stage.
The earlier temporary-change restriction is resolved: the coordinator approved a clean throwaway branch and the pinned simulator fixture. I exercised that approach against current mobile main (8b6ac6c1, v1.1.0), verified the exact fixture bytes, and committed the capture prerequisite with a clean worktree.
The required launcher still exits before build/simulator: it requires the old iosApp/VitruvianPhoenix Xcode project, while current main uses iosApp/PhoenixApp. This is a capture-tool compatibility blocker, not a question for the reporter. Shared harness tooling was left unchanged; no older host was substituted as current-app evidence.
No screenshot or visual approval is claimed. Standard / Large / Extra large, the scale cap and the compact/tablet comparisons remain pending a current-main-compatible capture path. Implementation/signoff remains gated on real app images. The separate stop-button request #1366 is unchanged.
The prototype capture blocker is root-caused and repaired on the throwaway prototype branch:
Launcher: current-main-compatible scoped launcher (PhoenixApp project/scheme/product paths) — builds, installs and launches the app in the Phantom simulator successfully.
Simulator target: the Xcode project refused simulator SDKs by design (device-framework-only Run Script); patched on the throwaway branch to map iphonesimulator* to the iosSimulatorArm64 framework/resources.
Fixture seam: the Phantom BLE + EULA fixture wiring (previously harness-branch-only) is ported; the simulator log confirms PHOENIX_IOS_FIXTURE_READY just-lift-connected.
Mockup implementation: the weight-text-size prototype (Standard/Large/Extra large at 1×/1.5×/2×, effective-scale cap 2.0, all scoped weight surfaces incl. both HUD branches and the Just Lift enlarged readout) is implemented and compiles for both iOS targets.
One remaining defect before screenshots: app startup currently fails during Koin construction of MainViewModel (InstanceCreationException, diagnostic STARTUP_INITIALIZATION_FAILED) on the simulator build. Diagnosis is one step from resolution (chained-cause logging is in place on the branch). No prototype screenshots are claimed yet — the gate stays incomplete until real captures land.
In-app prototype — weight-text-size preference (Standard / Large / Extra large)
Real screenshots of the real app (current main 8b6ac6c1 + throwaway prototype branch) running in the iOS simulator via the Phantom harness (just-lift-connected fixture). Visual mockup with hardcoded values; the final implementation wires to UserPreferences per the integration model.
Scale contract (confirmed):Standard/Large/Extra large = 1x/1.5x/2x; effective E = min(A x O, 2.0); rendered = base x M x O with M = E/O. Effective-scale cap = 2.0 — choices converge at the cap (see the two *_OS200 captures, badge reads CAPPED (physical 2.0x; choices converge at cap)). No 3x option in v1.
Preference levels across all in-scope surfaces (prototype screen: 3 weight sliders + WeightStepper + both WorkoutHud branches + Just Lift readout)
Standard (1x)
Large (1.5x)
Extra large (2x)
Cap convergence at OS 200% (both render weight values at 2.0x physical — identical to Extra large at OS 100%)
Extra large @ OS 200% (capped)
Large @ OS 200% (capped)
Real Just Lift screen (Option A per-cable readout) + compact header + short-height stacked branch
Just Lift Extra large
Just Lift @ OS 200% (compact stacked header)
Just Lift Extra large, short-height branch
Remaining captures (just_lift-STANDARD, just_lift-LARGE, just_lift-STANDARD-stacked, prototype-* full set) are on the same release asset list.
Verified
No clipping, overlap, or lost controls at any level, including the compact stacked-header branch (label-above-value, real SliderWithButtonsfontScale >= 1.15 path) and the real Issue [Bug] Weight Change Per Rep slider doesn't allow 1 lb increments in Just Lift mode (iOS) #571 short-height stacked branch (captured at 1x and 2x). The only ellipsis truncation (Stall Detection description) is identical at Standard — pre-existing, not scale-induced.
Secondary weight displays (e.g. "Total: …", rep sliders) are excluded from app scaling, as scoped.
Evidence index, responsive decisions, and the full 576-case coverage spec: phoenix-enhancements/issue-1367-github-issue-1367/prototype/capture-evidence-20261104.md and responsive-coverage-spec.md (639 assertions pass).
This is a design prototype: static mock data, no ViewModel wiring. Signoff gate remains for the product owner.
Decision: NOT APPROVED — capability hold for missing/insufficient visual evidence.
Reviewed 2026-11-04 by phoenix-rca, actual model gpt-6-astra (OpenAI Codex). No human approval is requested. This review supersedes the prototype completion claim for downstream eligibility, not the retained historical capture files. No production implementation is authorized.
Scope and review
Read live ticket, complete GitHub issue/comments, feasibility.md, architecture.md, integration-model.md, prototype/prototype-gate.md, prototype/prototype-consolidation.md, and prototype/capture-evidence-20261104.md. Issue is open, enhancement-only, not deferred; issue reactions are empty and comments contain no later human approval. No Discord thread exists.
Verified capture worktree HEAD 6e6506836eac97ba66ce3f20ba4ef9911f3caa1b, clean status. Enumerated captures: 11 named prototype images plus historical startup failure proto-standard.png. Directly inspected all five prototype-* PNGs, just_lift-XL_OS200.png, and just_lift-EXTRA_LARGE-stacked.png. Read the actual prototype source and searched production call-site wiring. This is a signoff audit, not a new build/runtime test.
What is supported
The displayed slider and stepper demonstrate meaningful enlargement at Standard/Large/Extra large. The two OS-200% screenshots show the slider value converging visually. Just Lift demonstrates a large additional readout in the existing card. These are genuine useful native images, not evidence of a shipped feature. The targeted global preference, three weight sliders, both HUD value branches, WeightStepper and Just Lift Option A remain a reasonable bounded design. Stop behavior #1366 remains excluded.
Retain consolidation §3–§5 as the proposed contract: 1x/1.5x/2x, no 3x, OS-aware rendered-size cap (not layout-only), scaled explicit sp line height, device-local preference and no picker/gesture/BLE/profile/backup/tvOS changes. The 2x option meets the lower end of the request; do not advertise 3x. Android nonlinear text conversion still requires platform validation. This is design direction, not final approval.
Blocking evidence findings (high confidence)
Settings choice has not been prototyped visibly. None of the enumerated images shows Settings → Appearance, its three-choice control, helper text or narrow/large-text fallback. WeightTextScalePrototype.kt:44–76 selects mock state from environment; :112–189 renders sample cards, not a Settings selector. The architecture explicitly requires the Settings control in the prototype.
The screenshots do not show all claimed surfaces at the claimed scales. At OS 200%, both prototype images show the heading/badge and first slider; the stepper and all HUD content are below the captured viewport. At Standard/Large, the bodyweight value is below the viewport; Extra large cuts through the Adaptive sample. These are scrollable-screen evidence omissions, not proof the app itself clips those values (:114–118 uses verticalScroll). Capture scrolled states. A badge claiming 2x is not layout verification of offscreen components.
HUD samples are not the HUD layout.WeightTextScalePrototype.kt:148–173 uses standalone Text/Column cards rather than HudBottomBar. That cannot prove fit with the actual HUD's neighboring context or multiline bodyweight value. Likewise a single shared slider sample is not evidence of three host-screen constraints.
No real compact/short-height/tablet geometry matrix. Capture evidence §53–58 explicitly says the short-height branch was forced with LocalWindowSizeClass while device geometry stayed unchanged. That tests branch selection, not short-height control reachability. OS 150% and tablet captures are absent. The 639 specification assertions are not 576 rendered test cases. Do not call this full viewport coverage.
Just Lift mock value contradicts its picker. The inspected screenshots show picker 22 lbs, readout L 22.5 kg R 20.0 kg, and total 44.1 lbs. JustLiftScreen.kt:537 hardcodes the added readout. Static mock data is allowed, but mutually inconsistent units/values and an invented asymmetric left/right presentation do not validate a safety-critical per-cable design. Show a coherent formatted per-cable value matching the picker (or explicitly demonstrate an existing asymmetric contract); do not carry this literal into implementation.
Required app-scale layout behavior is not present in the prototype.SliderWithButtons.kt:60 still stacks only on OS fontScale >= 1.15; app Extra large at OS 100% therefore does not trigger the required effective-scale branch. The Extra-large screenshot visibly keeps label/value in one row. Prototype the agreed effective-scale behavior before using it as responsive evidence.
Historical blocker text in prototype-gate.md and ticket enhancement_pr_blocker is stale: native capture tooling now works. The current hold is incomplete/mischaracterized evidence, not the old launcher failure.
Bounded remediation / evidence needed to resume
Prototype owner should use the retained throwaway capture branch; do not port simulator auth fallbacks or fixture plumbing into production.
Add/capture the Settings control with Standard/Large/Extra large, cap explanation and readable compact/large-text fallback.
Capture scrolled states for every required value, especially WeightStepper and No machine load at OS 200%; show actual HUD host layout and representative real weight-selection host constraints.
Apply the effective-scale layout trigger in the mockup and capture app XL + OS 100%, not only OS-driven stacking.
Make Just Lift mock picker/readout/total consistent using the existing units/formatter and approved per-cable semantics.
Produce genuine constrained compact-width and short-height geometry plus tablet comparisons at representative OS 100/150/200%; an explicitly bounded in-app preview viewport is acceptable if real measured constraints and chrome are documented. A size-class override alone is not geometry proof. Do not reset shared simulators without ownership checks.
Publish an honest coverage manifest separating captured/inspected cases from planned cases and source/arithmetic checks. Refresh prototype-gate.md and capture-evidence claims; preserve historical files.
Full production persistence, failure ordering, screen-reader behavior, Android adaptive conversion, touch bounds and final builds remain implementation/release obligations; the hold is not asking the prototype to implement the entire feature. The missing representative design visuals above are required now.
State and routing
Record changes_requested / signoff_evidence_blocked, clear human-wait flags, set prototype_gate_complete=false while retaining real captures and their completion timestamp as historical production-of-artifacts facts. Do not set signoff_completed_at or approved. Route recommendation: prototype, for evidence repair. Bridge/coordinator owns routing; no new card is created by this worker. Resume signoff after corrected evidence is available.
The revised native mock now includes the three-choice Appearance control, dedicated weight-slider/stepper/HUD previews, coherent per-cable picker/readout values, and measured constrained-preview layouts. This is temporary design code, not a shipped feature.
The Kotlin iOS simulator framework compiled successfully. Building the current PhoenixApp host then stopped at its explicit resource-phase guard:
PhoenixApp links the iosArm64 device framework; unsupported SDK_NAME: iphonesimulator27.0
This is a different, directly observed build-configuration blocker—not the removed preview launcher. I did not bypass the guard or alter production authentication, storage, BLE, or stop-button behavior.
No new screenshots were produced, so the prototype and signoff gates remain incomplete. Earlier captures are retained but do not prove the missing Settings, visible HUD/stepper, consistent Just Lift, or constrained-layout cases identified by the signoff review. The next required step is a supported simulator host/resource configuration, followed by native capture and visual review. No additional reporter information or human approval is being requested.
The previously reported simulator SDK/resource build blocker is cleared under the sanctioned capture policy. The current PhoenixApp simulator build succeeds.
The capture app now stops during normal initialization: PortalTokenStorage.verifyStorageIntegrity attempts a Keychain write and receives -34018, “A required entitlement isn't present.” The earlier missing local build configuration was corrected first; this is a separately captured runtime failure, not another SDK-build failure.
The failure is at the later storage-integrity write, rather than the settings-constructor site covered by the narrow simulator fallback policy. I have not bypassed startup or storage checks, substituted runtime repositories, or changed production code. The concrete failure has been returned to the coordinator for a supported startup-ready capture environment or source-specific remedy.
The revised visual gate is not complete. No new accepted feature screenshots are claimed; the historical images do not yet satisfy the Settings, full-viewport, HUD/stepper/bodyweight, geometry and consistent Just Lift evidence requested in signoff. The capture install has been removed and its dedicated simulator shut down. This update is not design signoff or release approval; #1366 stop-button behavior remains separate.
Revised native weight-text prototype — ready for design review, not shipped
The larger-weight-text design has now been recaptured in the real Phoenix iOS app on phone and iPad simulators, including measured compact/short preview areas and scrolled checkpoints.
Proposed choices: Standard / Large / Extra large = nominal 1× / 1.5× / 2×. No 3× option. Weight-only scaling combines with device text size up to a2× base-size ceiling, so options can look identical at larger device text sizes.
V1 scope: Set Ready, Routine Overview, rest weight selection, exercise-setup stepper, and workout numeric / Adaptive / No machine load readouts. Just Lift stays unchanged in this first version: the coherent same-unit trial still crowded/truncated its native picker on the compact large-text preview, so this uses the previously authorized defer-Just-Lift fallback rather than silently shrinking it or changing picker gestures. The separate Stop-button request #1366 is not part of this change.
Settings — three choices at normal text scale
Standard
Large
Extra large
Representative weight readouts
App Extra large, OS100 preview
Stepper, OS200 preview (scrolled)
Actual HUD bodyweight value, OS200 preview (scrolled)
Constrained geometry and tablet comparisons
360×480dp Settings: Extra large selected
Same short Settings: scrolled preview
Short weight controls: scrolled
iPad800dp preview
Evidence and honest limits
Inventory: 236 retained successful PNGs: 200 curated design-evidence frames, 28 rejected Just Lift trial frames,8 excluded incorrectly selected Settings frames. The incorrect frames stay preserved, not relabeled; their corrected captures assert Extra large after native taps. A disk-full supplemental test and an earlier scroll-selection test remain recorded as failed runs. The corrected42-frame test passed; this is not a claim that the failed suites passed.
Phone111 and tablet30-frame core runs passed; successful frames from the failed supplemental run are individually validated, not counted as a green suite. Hashes, PNG decoding, AX-selected state and measured geometry were checked; a separate bounded integrity/spec audit passed32 checks. These totals are evidence inventory, not production test-case counts.
The preview really constrains360dp compact,360×480dp short, and800dp tablet geometry. “OS100/150/200” means scoped fontScale inside the preview, not a changed system Settings preference. Top/scrolled images must be read together; not everything fits simultaneously.
Native components run with local mock data. Settings/slider host cards are representative previews; HUD uses its real bottom-bar component with a mock workout body and inactive rack context. Orange Done/Finish preview controls are inert scale1 mock actions, not real workout controls. No persistence, app-restart preference, hardware/load, BLE or Phantom acceptance is claimed.
A standard simulator-signing configuration recovered ordinary startup through the real Keychain path; no storage fallback or integrity bypass was used. Temporary source changes are local-only, not pushed into a delivery branch.
Final implementation still needs real preference persistence, full-screen wiring, device-local/Just-Lift-exclusion copy, Android/real-OS scaling, longest localized values/units, accessibility/hit bounds, builds and final screenshots. The captured Settings helper shows the cap explanation; the longer final scope/device-local copy is specified for implementation, not falsely claimed to appear in these pictures.
Next: escalated-model design signoff reviews this packet and fallback choice before implementation. No human-approval wait. This is not release or merge approval. Phoenix features are developed voluntarily; there is no charge.
APPROVED for implementation specification and development, with the binding decisions below. Not release, merge, hardware-safety or finished-feature approval.
Reviewed 2026-11-14 by phoenix-rca, actual model gpt-6-astra, provider openai-codex; task t_529eaf05. This decision supersedes the 2026-11-04 evidence hold, archived as signoff-20261104-rejected.md. No human approval wait. Issue is OPEN, labeled enhancement, not deferred at review; issue reactions were empty and live comments supplied no subsequent human product-direction change. There is no Discord thread, so GitHub is the notification surface.
1. Scope and rationale
Approve a device-local, global phone-app Weight text size preference in Settings → Appearance, after Theme and before Dynamic color: Standard / Large / Extra large, nominal 1 / 1.5 / 2. Extra large meets the lower end of the requested 2–3× enlargement on supported surfaces at normal device scale. No 3× option and no promise of 2× relative to an already OS-enlarged value.
Included: Set Ready, Routine Overview and Rest Timer weight sliders; exercise setup/edit WeightStepper; workout numeric, Echo Adaptive, and bodyweight No machine load readouts. Approve the integration model's Option B: Just Lift selection cards/picker remain unchanged in v1, with no added readout. This is a deliberately partial first version, not fulfillment of larger text on every selection surface. The coherent Option A trial still shows truncated native-picker digits at compact OS200; rejecting it is preferable to silently shrinking text or changing gestures. A later Just Lift increment requires separate accepted design and picker-fit/gesture evidence; it is not silently included in this implementation.
Stop-button request #1366, rep app scaling, secondary displays, picker typography/gesture rules, load sources/units/rounding, BLE, profile/cloud/backup schemas, portal and tvOS remain excluded. No price is quoted; the feature is developed voluntarily with no charge.
2. Evidence reviewed and disposition of the old hold
Read feasibility, architecture, integration model, previous signoff, current prototype gate, revision-2003/design-review.md, final review contract, complete retained source diff, visual-inspection/independent-review records, live ticket and issue comments. Current capture SHA bb9eb58ffd05f756e2e1522b3f88ea4c76623295, base 4d4de558be1733c44e6ff9ead4bccfddbfd9b1dc.
Executed signoff-evidence-audit.py: all 236 PNG hashes/header dimensions and unique paths matched; classification recomputed as 200 curated design frames, 28 rejected Just Lift trials, 8 excluded wrong-selection frames; retained source-diff hash matched. This independent read-only check is not a new runtime suite or full pixel decode. Upstream full-decode/AX audits and passing core/corrected runs are retained evidence, not tests rerun here. Disk-full and failed scroll-selector runs remain failed.
Personally opened these eight native-frame contact sheets:
Findings: selected radios and increasing Settings preview are visible; corrected short Settings shows Extra large selected and scroll-reachable copy/preview. OS100 weight headers stack for Large/XL while rep values remain unchanged. Actual HUD numeric/Adaptive/bodyweight values are visible in scrolled samples, including two-line No machine load. Short-height stepper exposes value, +/- and total after scrolling. Tablet samples fit; OS200 stepper values visually converge across choices. Just Lift mock now has coherent kg values but visibly truncates picker digits and remains rejected.
These findings close the prior representative design evidence hold: missing selector, offscreen-only values, standalone HUD text, size-class-only geometry, contradictory mock values, and absent app-scale stacking were addressed. They do not prove every production host fits. Native source confirms measured BoxWithConstraints viewports and real component calls; local state, sample hosts, inactive rack context and inert scale1 preview actions remain limitations.
Public revised evidence: #1367 (comment) . Images attached to release v1.1.0 are prototype assets, not proof that this feature shipped in that release.
3. Binding corrections — precedence over upstream sketches
Rendered cap, not layout-only cap. For a valid linear OS scale O and app factor A, E=min(A×O,2), style multiplier M=E/O. Apply M to font size and explicit sp line height; preserve other style attributes and relative/unspecified line-height semantics. Use E for weight layout pressure. The integration model's app-only style helper plus separately capped layout helper is superseded: it would leave rendered text uncapped. Guard invalid/nonpositive/nonfinite scale inputs. On nonlinear Android conversion, derive the capped target and inverse conversion through the actual platform density converter, rather than assuming division by fontScale enforces the physical ceiling. Prove this in platform tests; no pixel/physical measurement claim from these iOS previews.
Standard exception is explicit. Standard preserves base typography through OS200 in the linear model; above the cap, weight values are intentionally bounded even for Standard. Consequently the old acceptance statement 'Standard reproduces current rendering at every OS size' is superseded. Non-weight text continues to honor OS sizing normally. Test above-cap boundaries and state this limitation in final Settings help. Do not silently introduce a different cap or make Large smaller than Standard at the same device setting.
Line height scales. Older instructions to preserve an explicit sp line height unchanged are superseded; multiline bodyweight text needs corresponding line-height growth.
Option B replaces all Option A work. Remove JustLiftScreen/native-picker edits and new per-cable readout from the delivery packet. Preserve shouldStackWeightCards and current picker behavior. The rejected trial remains archived, not implementation source.
Explicit weight opt-in. Use isWeightValue: Boolean = false (or the equivalent explicitly specified semantic parameter) in the shared slider, enabled only at weight sites. Do not transplant the prototype's ambient-for-all-sliders approach and rely on manually clearing it at rep sites. Re-census current main; the historical three-weight/five-rep census is not permission to miss new call sites.
Global device-local preference. Store the enum name under weight_text_scale; missing/unknown/corrupt → STANDARD. Route through UserPreferences → existing globalSettings mapper/flow → Settings UI and phone CompositionLocal. Preserve profile isolation and the existing Settings flow-count contract; no SQLDelight/cloud/backup migration or codec work. App backup restore does not restore this setting; it must not reset a device's existing choice merely because backup data lacks it.
Responsive design, not preview geometry transplant. Use native selected-state radio rows for compact/large text; horizontal selection only where labels fit. Whole row must be selectable with ≥48dp hit bounds, no duplicate announcements. Effective app scale drives compact weight-header stacking and stepper value-row separation. Revalidate HUD optional rack content and real host/action constraints. Do not ship local-density overrides, inert preview controls, sample workout data, early-return prototype entry, capture diagnostics, or simulator resource/signing adaptations.
4. Required user wording
Settings helper (localize, allow scrolling; do not cram into fixed height):
Enlarges weight values in Set Ready, Routine Overview, rest, exercise setup and workout readouts. Just Lift is unchanged in this first version. Saved on this device only.
Combines with your device text size, up to 2× the standard base size. Options may look the same at larger device text sizes. This limit also applies to Standard when your device text size exceeds it.
Release/user instructions must repeat the supported surfaces, Just Lift exclusion, no 3× option, device-local/not-in-app-backup behavior, and how to return to Standard. This expanded copy is a binding specification correction, not text already captured in the mock. Final Settings screenshots must show it without clipping.
5. Implementation and release acceptance still required
Revalidate seams on implementation HEAD; retain enum normalization, persistence/reload, immediate update, write-error and last-choice ordering tests, plus cold-start behavior that avoids a visible incorrect Standard flash for a saved larger choice.
Test cap/line-height/invalid-scale boundaries, actual Android nonlinear conversion and real OS scaling on both platforms, including above-200% where supported. Do not pass scalar arithmetic as platform visual acceptance.
Extend profile separation/global flow and weight-source guards; verify weight sliders opt in and all rep/secondary/Just Lift paths do not receive app scaling. Presentation changes must not change weights, units, increments, commands or saved data.
Native final-source screenshots on compact, short and tablet layouts at representative 100/150/200 settings: all included surfaces, kg/lb and valid maximum/long localized strings, bodyweight/Adaptive and optional rack context. Pair top/scrolled frames when needed; test real Stop/Finish and +/-/slider reachability rather than the inert scale1 mock controls. Existing behavior of [Feature]: Stop button should turn off the weight and stay in the current lift mode (Standard, Echo, etc.) #1366 is not changed.
Accessibility validation: screen-reader selected state/focus, unit/value announcements without duplicates, ≥48dp semantic hit bounds, meaningful disabled controls and supported bold-text behavior. Test expanded Settings help, not only short prototype copy.
Discover actual current KMP task names; use the current host test task (:shared:testAndroidHostTest where configured), not the stale testDebugUnitTest instructions. Run required Android/iOS builds/tests on the final delivery branch; preserve CI/review/merge gates.
Confidence: high that the revised representative evidence supports this bounded design decision; production behavior, platform-wide fit, persistence and release readiness remain unverified. No fresh build, simulator mutation or product code change occurred in this review.
6. Handoff
Set signoff approved, clear both awaiting/needs signoff flags and obsolete blocker, keep prototype evidence complete, and release bridge-managed implementation specification/user instructions. The bridge owns task creation. Approval applies only with this document's corrections; neither a merge nor issue closure is authorized here.
The signed-off weight-text design is now specified for implementation. Summary of the delivery plan (spec + user instructions are held in the enhancement artifacts; decisions below are binding):
What ships (v1)
A device-local Weight text size setting in Settings → Appearance (between Theme and Dynamic color): Standard / Large / Extra large (nominal 1 / 1.5 / 2×), applied immediately, persisted per device (weight_text_scale, unknown/corrupt → Standard). No 3× option.
Larger weight values on: Set Ready, Routine Overview and Rest Timer weight sliders, exercise setup/edit WeightStepper, and the workout HUD readout (numeric, Echo "Adaptive", bodyweight "No machine load").
Rendered 2× cap: effective size = min(app × OS, 2), applied to font size and explicit sp line height; Android's nonlinear conversion proven via the platform density converter. Standard is unchanged up to OS 200%; above the cap, weight values are intentionally bounded (stated in the Settings help text).
Responsive behavior: weight slider headers stack at compact widths when enlarged (rep sliders unchanged), stepper value moves to its own row at the largest sizes, HUD value wraps without collisions. Accessibility: ≥48dp targets, correct selected-state/announcements, no duplicate value semantics.
What does not ship (binding exclusions)
Just Lift stays unchanged in v1 (native picker unchanged, no added readout, no card restacking) — a later increment would need its own accepted design. Rep sliders and secondary weight displays stay as-is. No changes to weights, units, increments, BLE/machine protocol, saved data, profile/cloud/backup schemas, portal, or tvOS. The stop-button request ([Feature]: Stop button should turn off the weight and stay in the current lift mode (Standard, Echo, etc.) #1366) is tracked separately. This feature is developed voluntarily at no charge.
Verification
Bounded packets P0–P8: preflight/re-census → preference + persistence → style helper (rendered cap) → Settings UI + contract-test extensions → surfaces + responsive layout → accessibility → full discovered Gradle gates (:shared:testAndroidHostTest, :shared:iosSimulatorArm64Test, compiles) → mandatory in-app visual evidence (all surfaces × Standard/Large/Extra large × OS 100/150/200% on compact/short/tablet, real OS text-size settings) → PR with Fixes #1367, CI green, escalated merge gate makes the merge call.
Guard tests ensure rep sliders, weight display sources, and the Just Lift stacking behavior are untouched, and that backup restore never resets the device's saved choice.
Release note will state the supported surfaces, the Just Lift exclusion, the device-local / not-in-backup behavior, and how to return to Standard.
The escalated merge gate merged PR #1399 into main as squash commit 2734310. This closes #1367.
What ships: a new Weight text size setting under Settings → Appearance, with Standard, Large and Extra large options. It enlarges the weight number on these screens:
Set Ready weight selector
Routine Overview weight selector
Rest-timer weight selector
Exercise weight stepper
Workout HUD weight readout
The text grows up to about 2×. When the phone's own text size is also large, the effective size is capped so the two settings don't multiply each other. The setting is saved on each device. Just Lift is unchanged, and the stop-button request (#1366) is tracked separately. This feature is developed voluntarily at no charge.
Gate audit:
Signoff is approved.
The diff matches the signed-off scope. It is presentation-only, with no changes to BLE, sync, schema, backup or tvOS.
Activity
9thLevelSoftware commentedon Oct 31, 2026
Sibling request: #1366 (stop button stays in lift mode). Both asks came in one user request; tracked separately.
9thLevelSoftware commentedon Oct 31, 2026
Feasibility assessment — weight-selection font (2–3×)
Recommendation: accept. Assessed against current
main(mobile app, shared Android/iOS UI).Next: architecture + integration modelling, then a visual prototype for signoff.
9thLevelSoftware commentedon Oct 31, 2026
Architecture — Larger weight-selection font for low-vision / older-adult users
github-issue-13679thLevelSoftware/Project-Phoenix-MPphoenix-enhancement-pipeline)main@8ef4d59c55943291393142179194ccae31728ff8Scope decision
This issue covers larger, user-configurable weight values on safety-critical weight-selection and
weight-readout surfaces in the phone app. It does not cover the separate stop-button request in
issue #1366.
No reporter clarification is required. The request, feasibility evidence, current UI seams, and
existing app patterns are sufficient to choose the architecture. The assumptions below are
intentional and must be confirmed in the prototype stage.
Architecture decision
Adopt a device-global, app-level weight-text-size preference with three choices:
This is deliberately more conservative than the reporter's literal "twice to three times" request.
spunits already respond to the operating system's text-size setting. A further 2–3× appmultiplier would compound with the OS multiplier and can make the safety-critical value illegible
or clipped (for example, 2× app setting with an OS scale of 2× can become an effective 4×). The
prototype stage should show Standard, Large, and Extra large at OS scales of 100%, 150%, and 200%;
if the review still considers 3× warranted, it may add a distinct 3× choice while retaining
the same composition and clamping contract.
The preference is global to the phone app, not profile-specific. Font size is a device/display
accessibility preference, not a training profile or cloud-synced training attribute. This keeps the
change out of SQLDelight, the profile-preference sync codec, portal APIs, and backup schema.
System shape
1. Preference domain model
Add a small, closed domain type rather than storing an arbitrary float:
The enum is the persistence contract. Invalid or unknown persisted values must normalize to
STANDARD; this keeps forward/backward compatibility deterministic.2. Persistence and state flow
Store the selected mode in the existing global
SettingsPreferencesManagerusing a new key (forexample,
weight_text_scale) and expose it throughUserPreferences.weightTextScale.State flow:
The existing
UserPreferencesfacade is the right transport because it is already observed by theapp shell and already separates global preferences from profile-scoped training preferences.
SettingsManager.overlayProfile()must not overwrite the new field, so it remains global.Do not place this field in
CoreProfilePreferences. That model is stored in the SQLDelightUserProfilePreferencestable and encoded/decoded by the portal profile-preference sync codec;adding a field there would create schema migration, portal payload compatibility, dirty/generation
handling, and cross-device sync work without user value.
3. Composition seam
Add a phone-level composition local, adjacent to the existing accessibility seam:
EnhancedMainScreenshould provide the selected value once for the phone shell. Components canalso read it directly with a default of
STANDARD, so isolated previews/tests and any screenrendered outside the shell remain safe.
This is separate from
PlatformAccessibilitySettings, whose current fields are platform-owned(
boldTextEnabled,reduceMotion). The new setting is app-owned and persisted, so mixing the twowould blur ownership and complicate the Android/iOS expect/actual implementations.
4. Style resolution helper
Create one pure/shared helper that maps a base
TextStyleand the selected scale to the finalweight-value style. It must:
fontSizeusing the preference factor.Recommended contract:
with a capped effective factor. For example, cap the product of the app factor and OS font scale at
2.0–2.5. The exact cap is a prototype decision, but it must be deterministic and unit-tested. The
helper should prefer an explicit
spsize derived from the base style rather than multiplying aCompose
spvalue by the OS font scale again, becausespalready incorporates OS scaling. Inother words: resolve the final physical size once, then let Compose apply the OS scale exactly
once.
5. Affected surfaces
The v1 contract covers the primary selection and live-readout surfaces:
SetReadyScreen.kt->SliderWithButtonstitleMediumRoutineOverviewScreen.kt->SliderWithButtonstitleMediumWorkoutHud.kt->HudBottomBartitleLargecomponents/WeightStepper.ktheadlineMediumJustLiftScreen.ktSecondary displays (recommendations, bulk adjust, quick-history columns, set summaries) are
not in v1. They are not selection surfaces, and enlarging them adds layout risk without
directly answering the request. They may be added later after the primary surfaces are validated.
6. Responsive-layout contract
The enlarged value must never be clipped or pushed into an unreadable layout. The current
SliderWithButtonsonly stacks its label/value header when OSfontScale >= 1.15; the new appscale must participate in that decision. Likewise,
isCompactAccessibilityLayout()and theJust Lift card stacking helper must consider the effective app scale.
Required behavior:
scale requires it, even if the OS font scale is 1.0.
WeightSteppermust keep both +/- controls reachable; if the value cannot fit beside them atExtra large, the value moves to its own row while the controls remain on one row.
with other bottom-bar content.
otherwise crowd the card.
Architecture should introduce a shared
effectiveWeightTextScale/ layout-pressure input ratherthan duplicating
fontScale >= 1.15checks in each surface.API, backend, and schema boundaries
shared/src/phoneMainand are compiled forAndroid and iOS.
Settingskey; no SQLDelight schema change.rounding, or commands.
UserPreferencesbackup codec serializes the entire data class,the new field must be backward-compatible (default
STANDARDwhen absent). No backup version bumpis required if the existing codec already tolerates unknown/missing fields; the integration-model
stage must verify that exact codec path.
TvCoreCompositionand TV-only UI remain untouched.Compatibility and migration
STANDARD, so the shipped UI is unchanged until a user opts in.STANDARD.STANDARDrather than crashing.bounded helper, not by silently replacing the OS behavior.
Rollout and flag strategy
This is a low-risk, user-initiated display preference, so it does not need a remote feature flag.
Recommended rollout:
STANDARDas the default.section, using a segmented control or radio group with plain labels (Standard, Large, Extra
large) and a short description of the effect.
after release.
If implementation discovers that Extra large cannot fit a required surface without changing
navigation or control layout, that surface should cap the factor at the largest size that remains
fully readable and visibly explain the cap in the settings description rather than silently
clipping.
Accessibility semantics
meaningful accessibility text (value, unit, increment/decrement actions).
between the visible value and any hidden copy.
rest of Settings.
Analytics and telemetry
No new telemetry is required for correctness. If the project already has a privacy-approved local
analytics/event mechanism, a privacy-minimal event such as
weight_text_scale_changedwith theselected enum name (no user identifier, no workout data) can be considered. Do not add a remote
analytics dependency for this feature.
Non-goals
reporter's cost question separately.
Acceptance criteria
the workout HUD readout,
WeightStepper, and Just Lift weight cards.short-height phone, and a tablet, including OS font scales of 100%, 150%, and 200%.
enforced and unit-tested.
accessibility.
profile data, portal sync payloads, or tvOS behavior.
and compact-layout/stacking behavior; Android and iOS builds compile.
implementation is finalized.
Downstream hand-off
Integration model must map
UserPreferencesfield and global preference key.SettingsPreferencesManagerload/set methods.SettingsManagerandMainViewModeldelegation.SettingsGlobalUiStateandSettingsTab/NavGraphwiring.EnhancedMainScreencomposition provider.Prototype must show
Implementation risk checkpoints
mainalready contains the relevant accessibility and compact-layoutseams, but layout behavior is subtle on iPhone landscape/short-height devices. Keep the
implementation packet bounded to one shared helper, the preference wiring, the v1 surfaces, and
focused tests.
SliderWithButtonscomponent for rep sliders if the enlarged weightbehavior would affect those values. Add an explicit weight-value treatment or parameter rather
than globally changing every slider value.
WeightDisplaySourceGuardTestand related guard tests: this feature changespresentation only and must not alter the source of the displayed weight.
9thLevelSoftware commentedon Oct 31, 2026
Integration Model — larger weight-selection font (stage 3 of 5)
Baseline:
main@8ef4d59c(identical toorigin/mainat review time).This stage maps the architecture onto the actual codebase at the current
main. It confirms thearchitecture's persistence design works as written, and it surfaces two corrections to the
surface list — both narrowing the work rather than widening it.
Corrections to the architecture's surface map
1. There are three weight sliders, not two. The architecture's table listed Set Ready and
Routine Overview.
RestTimerCard.kt:577has a third weight slider using the sameformatWeightWithUnithelper as the other two. It is a weight selection surface (the user editsthe next set's load from the rest timer), so leaving it out would give the reporter large weight
text while planning a set and small weight text while resting. Recommend including it.
The full
SliderWithButtonscensus is 8 call sites — 3 weight, 5 rep — all inphoneMain, noneshared with tvOS. The 5 rep sites and 3 weight sites are cleanly separable, and the component has
no value-text style parameter today, so adding one with a default that preserves the current
titleMediumkeeps the rep sliders untouched.2. The Just Lift weight value is a native picker, not Compose text. The architecture says
"scale the weight value within each card". The Just Lift "Weight per Cable" card calls
CompactNumberPicker, which is anexpect/actual:NumberPickerwidget — its numeric text is sized by the OS widget, andNumberPickerexposes no text-size setter.UIPickerViewwheel whose Compose-side selected-value style is chosen insideCompactNumberPicker.ios.kt:96-127.The
expectsignature has no typography parameter, so scaling that value is not implementablethrough the existing API. Recommended v1 treatment: keep the picker as the input control and add an
enlarged Compose readout of the current per-cable weight in the same card. The prototype stage
can decide this (Option A) against deferring Just Lift (Option B).
A layout constraint the architecture's plan would have violated
The architecture asked that Just Lift cards "switch to their existing stacked layout when the
enlarged value would otherwise crowd the card".
shouldStackWeightCards(
WindowSizeClass.kt:124-126) deliberately stacks only on short-height windows. Its KDocexplains why: on iPhone portrait, stacking puts the iOS picker wheel directly above the
ProgressionSlider, and the wheel's verticalLazyColumnwas observed to claim drags intended forthe slider — the readout jumped to neighbour weights like 29/30/44-49 even though that slider's
range is
-10..+10.Integration model therefore does not change
shouldStackWeightCards. Because the Just Liftvalue is a native picker, the enlarged readout is Compose text that can take its own line inside
the existing card without changing stacking mode. If it still cannot fit at Extra large, the
fallback is to cap that surface's factor and say so in the settings description — the
architecture's own rollout rule — rather than restack and re-introduce the gesture conflict.
The backup question, resolved
The architecture asked the integration stage to verify whether the global
UserPreferencesbackupcodec serializes the whole data class. It does not.
BackupData(BackupModels.kt:676) holds aBackupContentcontaining only profile/trainingentities —
userProfiles,profilePreferences,workoutSessions,routines,personalRecords,cycleProgress,gamificationStats, and so on. There is no globalapp-preference field.
backupPrivacySummary()(:703-714) explicitly states the backup does not include preferencesync bookkeeping, voice phrase or calibration, or adult consent and prompt state.
UserPreferencesacross any backup/export/import/serialize path returnsno hits outside production UI/VM code and tests.
Global settings live in
multiplatform-settings(SharedPreferences / NSUserDefaults) and aredevice-local, outside backups entirely. Consequences: no backup codec change, no backup version
bump, no migration. The "absent from older backups → decode as STANDARD" concern is moot; the
real guarantee is "the key is absent from the settings store → STANDARD". The practical consequence
worth stating in the release note is that a device restore leaves this display preference alone.
Verified state seam
Every seam the architecture named exists, and one nearly identical precedent makes the design
low-risk:
UserPreferencesfielddomain/model/UserPreferences.kt:8-73data/preferences/PreferencesManager.kt:111-165,170-241(SettingsPreferencesManagerlives inside this file)BleCompatibilitySetting.fromStorage(...)(:230-236) — unknown value → default, never throwssettings.putString(KEY, mode.name)+updateAndEmit { copy(...) }(:243-251)presentation/manager/SettingsManager.kt:44-88— a field not listed inoverlayProfilestays global automaticallypresentation/viewmodel/MainViewModel.kt:128-135,137-144navigation/NavGraph.kt:448-490(globalSettingsflow)presentation/screen/SettingsTab.kt:279-309—SingleChoiceSegmentedButtonRowenum pattern already used twicepresentation/screen/EnhancedMainScreen.kt:344-351The field is deliberately not placed in
CoreProfilePreferences, so there is no SQLDelightschema change, no portal sync payload change, and no backup change. No BLE, machine-protocol,
tvOS, or portal impact.
Style helper
Two pure functions rather than one, so each is independently unit-testable:
TextStyle.scaledForWeightValue(scale: WeightTextScale): TextStyle— scales the basefontSizein
sp. Because Compose applies the OSfontScaletospat render time, the OS scale isapplied exactly once, by Compose.
effectiveWeightTextScale(scale: WeightTextScale, osFontScale: Float): Float— the cappedproduct, used for layout decisions.
Recommended cap: 2.0 on
appFactor × osFontScale. The largest base weight style isheadlineMedium(28sp), so 2.0 → 56sp, which is already at the practical ceiling for a value on a360dp phone; 2.5 → 70sp would force stacking everywhere. The prototype confirms the number; the
implementation keeps it as a single named constant.
Layout decision points to extend
SliderWithButtonsheader stackingSliderWithButtons.kt:56-57fontScale >= 1.15fshouldUseCompactAccessibilityLayoutWindowSizeClass.kt:79-101fontScale >= 1.15|| bold)shouldStackWeightCardsWindowSizeClass.kt:124-126WeightSteppercontrol rowWeightStepper.kt:78-144weight(1f)between two buttons in oneRowRoutineOverviewScreenviewportRoutineOverviewScreen.kt:444-451height < 700.dp||fontScale >= 1.15fThe effective scale must make the slider header stack even when the OS font scale is 1.0 — that
is the architecture's explicit requirement.
Test plan highlights
STANDARD, never throws); the scalehelper (Standard reproduces today's
fontSizeexactly; font family/weight/spacing/line-height/color preserved; result in
sp); the cap matrix (2.0 app × 2.0 OS → capped, not 4.0);persistence (absent key, corrupt value, set/reload, flow emission);
overlayProfileleaves thefield global across a profile switch.
SliderWithButtonscall sites carry the treatment and the 5 rep sites do not. This is theregression guard for "must not affect rep sliders".
ProfileSettingsSeparationContractTestasserts the Settingsdestination collects exactly 3 flows and asserts each
SettingsTab(...)assignment string.The new preference must ride the existing
globalSettingsflow — adding a fourthcollectAsState()would break it. The two new wiring strings must be added in the same PR.WeightDisplaySourceGuardTest,WeightFeatureGuardTest,PreferencesIsolationTest,SetReadyScreenScrollWiringTest.phoneMainfeedsboth), full
:shared:testDebugUnitTest.Open items for the prototype stage
WeightStepperstill fit on a 360dp phone atOS 200%.
Adding a 4th enum constant is source-compatible; the persisted contract is the enum name
string and unknown values still normalize to
STANDARD.restore.
Full integration model (file/module map, per-surface excerpts, test plan, risk table) is recorded
in the enhancement artifact directory.
9thLevelSoftware commentedon Oct 31, 2026
Weight-text prototype — capture blocked, not approved
The design and integration inputs are complete. This stage requires screenshots of the proposed larger weight values inside the real phone app; no prototype screenshot or visual approval is being claimed yet.
The required Phantom launcher was executed and refused the checkout before a build or simulator launch. Its current contract rejects all uncommitted changes, whereas this prototype workflow explicitly requires temporary UI changes that are reverted, not committed. The shared harness also contains existing unrelated changes, which were left untouched. Current mobile
mainlacks the launcher's exact required simulator fixture, so switching to a clean current-main checkout is not a supported workaround.Next step: repair/authorize a prototype-compatible capture path, then render the Settings size choice and weight surfaces across the required phone/tablet and text-size matrix. The proposed Standard / Large / Extra large sizes and effective cap remain unvalidated, not signed off. One implementation caution: capping only the layout scale does not cap rendered text; the text-style calculation must also account for the operating-system scale.
This affects only the font request. The separate stop-button request #1366 is unchanged. The issue remains open; implementation/signoff has not been unlocked by this stage.
9thLevelSoftware commentedon Nov 4, 2026
Weight-text prototype — capture retry update
The earlier temporary-change restriction is resolved: the coordinator approved a clean throwaway branch and the pinned simulator fixture. I exercised that approach against current mobile main (
8b6ac6c1, v1.1.0), verified the exact fixture bytes, and committed the capture prerequisite with a clean worktree.The required launcher still exits before build/simulator: it requires the old
iosApp/VitruvianPhoenixXcode project, while current main usesiosApp/PhoenixApp. This is a capture-tool compatibility blocker, not a question for the reporter. Shared harness tooling was left unchanged; no older host was substituted as current-app evidence.No screenshot or visual approval is claimed. Standard / Large / Extra large, the scale cap and the compact/tablet comparisons remain pending a current-main-compatible capture path. Implementation/signoff remains gated on real app images. The separate stop-button request #1366 is unchanged.
9thLevelSoftware commentedon Nov 4, 2026
Prototype stage progress — capture tooling repaired (2026-11-04)
The prototype capture blocker is root-caused and repaired on the throwaway prototype branch:
iphonesimulator*to theiosSimulatorArm64framework/resources.PHOENIX_IOS_FIXTURE_READY just-lift-connected.One remaining defect before screenshots: app startup currently fails during Koin construction of
MainViewModel(InstanceCreationException, diagnosticSTARTUP_INITIALIZATION_FAILED) on the simulator build. Diagnosis is one step from resolution (chained-cause logging is in place on the branch). No prototype screenshots are claimed yet — the gate stays incomplete until real captures land.9thLevelSoftware commentedon Nov 4, 2026
In-app prototype — weight-text-size preference (Standard / Large / Extra large)
Real screenshots of the real app (current main
8b6ac6c1+ throwaway prototype branch) running in the iOS simulator via the Phantom harness (just-lift-connectedfixture). Visual mockup with hardcoded values; the final implementation wires toUserPreferencesper the integration model.Scale contract (confirmed):
Standard/Large/Extra large = 1x/1.5x/2x; effectiveE = min(A x O, 2.0); rendered= base x M x OwithM = E/O. Effective-scale cap = 2.0 — choices converge at the cap (see the two*_OS200captures, badge readsCAPPED (physical 2.0x; choices converge at cap)). No 3x option in v1.Preference levels across all in-scope surfaces (prototype screen: 3 weight sliders + WeightStepper + both WorkoutHud branches + Just Lift readout)
Cap convergence at OS 200% (both render weight values at 2.0x physical — identical to Extra large at OS 100%)
Real Just Lift screen (Option A per-cable readout) + compact header + short-height stacked branch
Remaining captures (
just_lift-STANDARD,just_lift-LARGE,just_lift-STANDARD-stacked,prototype-*full set) are on the same release asset list.Verified
SliderWithButtonsfontScale >= 1.15path) and the real Issue [Bug] Weight Change Per Rep slider doesn't allow 1 lb increments in Just Lift mode (iOS) #571 short-height stacked branch (captured at 1x and 2x). The only ellipsis truncation (Stall Detection description) is identical at Standard — pre-existing, not scale-induced.Evidence index, responsive decisions, and the full 576-case coverage spec:
phoenix-enhancements/issue-1367-github-issue-1367/prototype/capture-evidence-20261104.mdandresponsive-coverage-spec.md(639 assertions pass).This is a design prototype: static mock data, no ViewModel wiring. Signoff gate remains for the product owner.
9thLevelSoftware commentedon Nov 4, 2026
Issue #1367 — escalated-model signoff review
Decision: NOT APPROVED — capability hold for missing/insufficient visual evidence.
Reviewed 2026-11-04 by
phoenix-rca, actual modelgpt-6-astra(OpenAI Codex). No human approval is requested. This review supersedes the prototype completion claim for downstream eligibility, not the retained historical capture files. No production implementation is authorized.Scope and review
Read live ticket, complete GitHub issue/comments, feasibility.md, architecture.md, integration-model.md, prototype/prototype-gate.md, prototype/prototype-consolidation.md, and prototype/capture-evidence-20261104.md. Issue is open, enhancement-only, not deferred; issue reactions are empty and comments contain no later human approval. No Discord thread exists.
Verified capture worktree HEAD
6e6506836eac97ba66ce3f20ba4ef9911f3caa1b, clean status. Enumerated captures: 11 named prototype images plus historical startup failureproto-standard.png. Directly inspected all fiveprototype-*PNGs,just_lift-XL_OS200.png, andjust_lift-EXTRA_LARGE-stacked.png. Read the actual prototype source and searched production call-site wiring. This is a signoff audit, not a new build/runtime test.What is supported
The displayed slider and stepper demonstrate meaningful enlargement at Standard/Large/Extra large. The two OS-200% screenshots show the slider value converging visually. Just Lift demonstrates a large additional readout in the existing card. These are genuine useful native images, not evidence of a shipped feature. The targeted global preference, three weight sliders, both HUD value branches, WeightStepper and Just Lift Option A remain a reasonable bounded design. Stop behavior #1366 remains excluded.
Retain consolidation §3–§5 as the proposed contract: 1x/1.5x/2x, no 3x, OS-aware rendered-size cap (not layout-only), scaled explicit sp line height, device-local preference and no picker/gesture/BLE/profile/backup/tvOS changes. The 2x option meets the lower end of the request; do not advertise 3x. Android nonlinear text conversion still requires platform validation. This is design direction, not final approval.
Blocking evidence findings (high confidence)
WeightTextScalePrototype.kt:44–76selects mock state from environment;:112–189renders sample cards, not a Settings selector. The architecture explicitly requires the Settings control in the prototype.:114–118uses verticalScroll). Capture scrolled states. A badge claiming 2x is not layout verification of offscreen components.WeightTextScalePrototype.kt:148–173uses standalone Text/Column cards rather than HudBottomBar. That cannot prove fit with the actual HUD's neighboring context or multiline bodyweight value. Likewise a single shared slider sample is not evidence of three host-screen constraints.22 lbs, readoutL 22.5 kg R 20.0 kg, and total44.1 lbs.JustLiftScreen.kt:537hardcodes the added readout. Static mock data is allowed, but mutually inconsistent units/values and an invented asymmetric left/right presentation do not validate a safety-critical per-cable design. Show a coherent formatted per-cable value matching the picker (or explicitly demonstrate an existing asymmetric contract); do not carry this literal into implementation.SliderWithButtons.kt:60still stacks only on OS fontScale >= 1.15; app Extra large at OS 100% therefore does not trigger the required effective-scale branch. The Extra-large screenshot visibly keeps label/value in one row. Prototype the agreed effective-scale behavior before using it as responsive evidence.Historical blocker text in prototype-gate.md and ticket enhancement_pr_blocker is stale: native capture tooling now works. The current hold is incomplete/mischaracterized evidence, not the old launcher failure.
Bounded remediation / evidence needed to resume
Prototype owner should use the retained throwaway capture branch; do not port simulator auth fallbacks or fixture plumbing into production.
No machine loadat OS 200%; show actual HUD host layout and representative real weight-selection host constraints.Full production persistence, failure ordering, screen-reader behavior, Android adaptive conversion, touch bounds and final builds remain implementation/release obligations; the hold is not asking the prototype to implement the entire feature. The missing representative design visuals above are required now.
State and routing
Record changes_requested / signoff_evidence_blocked, clear human-wait flags, set prototype_gate_complete=false while retaining real captures and their completion timestamp as historical production-of-artifacts facts. Do not set signoff_completed_at or approved. Route recommendation: prototype, for evidence repair. Bridge/coordinator owns routing; no new card is created by this worker. Resume signoff after corrected evidence is available.
9thLevelSoftware commentedon Nov 13, 2026
Weight-text prototype revision — capture remains blocked
The revised native mock now includes the three-choice Appearance control, dedicated weight-slider/stepper/HUD previews, coherent per-cable picker/readout values, and measured constrained-preview layouts. This is temporary design code, not a shipped feature.
The Kotlin iOS simulator framework compiled successfully. Building the current PhoenixApp host then stopped at its explicit resource-phase guard:
This is a different, directly observed build-configuration blocker—not the removed preview launcher. I did not bypass the guard or alter production authentication, storage, BLE, or stop-button behavior.
No new screenshots were produced, so the prototype and signoff gates remain incomplete. Earlier captures are retained but do not prove the missing Settings, visible HUD/stepper, consistent Just Lift, or constrained-layout cases identified by the signoff review. The next required step is a supported simulator host/resource configuration, followed by native capture and visual review. No additional reporter information or human approval is being requested.
9thLevelSoftware commentedon Nov 14, 2026
Weight-text prototype progress — native startup blocker isolated
The previously reported simulator SDK/resource build blocker is cleared under the sanctioned capture policy. The current PhoenixApp simulator build succeeds.
The capture app now stops during normal initialization:
PortalTokenStorage.verifyStorageIntegrityattempts a Keychain write and receives -34018, “A required entitlement isn't present.” The earlier missing local build configuration was corrected first; this is a separately captured runtime failure, not another SDK-build failure.The failure is at the later storage-integrity write, rather than the settings-constructor site covered by the narrow simulator fallback policy. I have not bypassed startup or storage checks, substituted runtime repositories, or changed production code. The concrete failure has been returned to the coordinator for a supported startup-ready capture environment or source-specific remedy.
The revised visual gate is not complete. No new accepted feature screenshots are claimed; the historical images do not yet satisfy the Settings, full-viewport, HUD/stepper/bodyweight, geometry and consistent Just Lift evidence requested in signoff. The capture install has been removed and its dedicated simulator shut down. This update is not design signoff or release approval; #1366 stop-button behavior remains separate.
9thLevelSoftware commentedon Nov 14, 2026
Revised native weight-text prototype — ready for design review, not shipped
The larger-weight-text design has now been recaptured in the real Phoenix iOS app on phone and iPad simulators, including measured compact/short preview areas and scrolled checkpoints.
Proposed choices: Standard / Large / Extra large = nominal 1× / 1.5× / 2×. No 3× option. Weight-only scaling combines with device text size up to a2× base-size ceiling, so options can look identical at larger device text sizes.
V1 scope: Set Ready, Routine Overview, rest weight selection, exercise-setup stepper, and workout numeric / Adaptive / No machine load readouts. Just Lift stays unchanged in this first version: the coherent same-unit trial still crowded/truncated its native picker on the compact large-text preview, so this uses the previously authorized defer-Just-Lift fallback rather than silently shrinking it or changing picker gestures. The separate Stop-button request #1366 is not part of this change.
Settings — three choices at normal text scale
Representative weight readouts
Constrained geometry and tablet comparisons
Evidence and honest limits
Next: escalated-model design signoff reviews this packet and fallback choice before implementation. No human-approval wait. This is not release or merge approval. Phoenix features are developed voluntarily; there is no charge.
9thLevelSoftware commentedon Nov 14, 2026
#1367 — escalated design signoff
APPROVED for implementation specification and development, with the binding decisions below. Not release, merge, hardware-safety or finished-feature approval.
Reviewed 2026-11-14 by
phoenix-rca, actual modelgpt-6-astra, provideropenai-codex; taskt_529eaf05. This decision supersedes the 2026-11-04 evidence hold, archived assignoff-20261104-rejected.md. No human approval wait. Issue is OPEN, labeled enhancement, not deferred at review; issue reactions were empty and live comments supplied no subsequent human product-direction change. There is no Discord thread, so GitHub is the notification surface.1. Scope and rationale
Approve a device-local, global phone-app Weight text size preference in Settings → Appearance, after Theme and before Dynamic color: Standard / Large / Extra large, nominal 1 / 1.5 / 2. Extra large meets the lower end of the requested 2–3× enlargement on supported surfaces at normal device scale. No 3× option and no promise of 2× relative to an already OS-enlarged value.
Included: Set Ready, Routine Overview and Rest Timer weight sliders; exercise setup/edit WeightStepper; workout numeric, Echo
Adaptive, and bodyweightNo machine loadreadouts. Approve the integration model's Option B: Just Lift selection cards/picker remain unchanged in v1, with no added readout. This is a deliberately partial first version, not fulfillment of larger text on every selection surface. The coherent Option A trial still shows truncated native-picker digits at compact OS200; rejecting it is preferable to silently shrinking text or changing gestures. A later Just Lift increment requires separate accepted design and picker-fit/gesture evidence; it is not silently included in this implementation.Stop-button request #1366, rep app scaling, secondary displays, picker typography/gesture rules, load sources/units/rounding, BLE, profile/cloud/backup schemas, portal and tvOS remain excluded. No price is quoted; the feature is developed voluntarily with no charge.
2. Evidence reviewed and disposition of the old hold
Read feasibility, architecture, integration model, previous signoff, current prototype gate, revision-2003/design-review.md, final review contract, complete retained source diff, visual-inspection/independent-review records, live ticket and issue comments. Current capture SHA
bb9eb58ffd05f756e2e1522b3f88ea4c76623295, base4d4de558be1733c44e6ff9ead4bccfddbfd9b1dc.Executed
signoff-evidence-audit.py: all 236 PNG hashes/header dimensions and unique paths matched; classification recomputed as 200 curated design frames, 28 rejected Just Lift trials, 8 excluded wrong-selection frames; retained source-diff hash matched. This independent read-only check is not a new runtime suite or full pixel decode. Upstream full-decode/AX audits and passing core/corrected runs are retained evidence, not tests rerun here. Disk-full and failed scroll-selector runs remain failed.Personally opened these eight native-frame contact sheets:
prototype/revision-2003/resume-inspection/settings.jpgprototype/revision-2003/resume-inspection/app-stack.jpgprototype/revision-2003/resume-inspection/justlift-trial.jpgprototype/revision-2003/resume-inspection/hud200.jpgprototype/revision-2003/resume-inspection/short-stepper.jpgprototype/revision-2003/inspection/tablet-os150-2.jpgprototype/revision-final-2017/inspection/corrected-short-settings.pngprototype/revision-final-2017/inspection/convergence-stepper.pngFindings: selected radios and increasing Settings preview are visible; corrected short Settings shows Extra large selected and scroll-reachable copy/preview. OS100 weight headers stack for Large/XL while rep values remain unchanged. Actual HUD numeric/Adaptive/bodyweight values are visible in scrolled samples, including two-line No machine load. Short-height stepper exposes value, +/- and total after scrolling. Tablet samples fit; OS200 stepper values visually converge across choices. Just Lift mock now has coherent kg values but visibly truncates picker digits and remains rejected.
These findings close the prior representative design evidence hold: missing selector, offscreen-only values, standalone HUD text, size-class-only geometry, contradictory mock values, and absent app-scale stacking were addressed. They do not prove every production host fits. Native source confirms measured BoxWithConstraints viewports and real component calls; local state, sample hosts, inactive rack context and inert scale1 preview actions remain limitations.
Public revised evidence: #1367 (comment) . Images attached to release v1.1.0 are prototype assets, not proof that this feature shipped in that release.
3. Binding corrections — precedence over upstream sketches
isWeightValue: Boolean = false(or the equivalent explicitly specified semantic parameter) in the shared slider, enabled only at weight sites. Do not transplant the prototype's ambient-for-all-sliders approach and rely on manually clearing it at rep sites. Re-census current main; the historical three-weight/five-rep census is not permission to miss new call sites.weight_text_scale; missing/unknown/corrupt → STANDARD. Route through UserPreferences → existing globalSettings mapper/flow → Settings UI and phone CompositionLocal. Preserve profile isolation and the existing Settings flow-count contract; no SQLDelight/cloud/backup migration or codec work. App backup restore does not restore this setting; it must not reset a device's existing choice merely because backup data lacks it.4. Required user wording
Settings helper (localize, allow scrolling; do not cram into fixed height):
Release/user instructions must repeat the supported surfaces, Just Lift exclusion, no 3× option, device-local/not-in-app-backup behavior, and how to return to Standard. This expanded copy is a binding specification correction, not text already captured in the mock. Final Settings screenshots must show it without clipping.
5. Implementation and release acceptance still required
:shared:testAndroidHostTestwhere configured), not the staletestDebugUnitTestinstructions. Run required Android/iOS builds/tests on the final delivery branch; preserve CI/review/merge gates.Confidence: high that the revised representative evidence supports this bounded design decision; production behavior, platform-wide fit, persistence and release readiness remain unverified. No fresh build, simulator mutation or product code change occurred in this review.
6. Handoff
Set signoff approved, clear both awaiting/needs signoff flags and obsolete blocker, keep prototype evidence complete, and release bridge-managed implementation specification/user instructions. The bridge owns task creation. Approval applies only with this document's corrections; neither a merge nor issue closure is authorized here.
9thLevelSoftware commentedon Nov 14, 2026
Implementation plan — approved for development
The signed-off weight-text design is now specified for implementation. Summary of the delivery plan (spec + user instructions are held in the enhancement artifacts; decisions below are binding):
What ships (v1)
weight_text_scale, unknown/corrupt → Standard). No 3× option.What does not ship (binding exclusions)
Verification
:shared:testAndroidHostTest,:shared:iosSimulatorArm64Test, compiles) → mandatory in-app visual evidence (all surfaces × Standard/Large/Extra large × OS 100/150/200% on compact/short/tablet, real OS text-size settings) → PR withFixes #1367, CI green, escalated merge gate makes the merge call.Release note will state the supported surfaces, the Just Lift exclusion, the device-local / not-in-backup behavior, and how to return to Standard.
fix(phone): relocate WeightTextScaling to phoneMain to fix tvOS compi…
merge: origin/main into feat/1367-weight-text-size (#1367)
merge: origin/main into feat/1367-weight-text-size (#1367)
test: bump ExerciseMediaWiringTest migration ratchet 63 -> 64 (#1367)
9thLevelSoftware commentedon Nov 18, 2026
Merge gate: merged ✅
The escalated merge gate merged PR #1399 into
mainas squash commit 2734310. This closes #1367.What ships: a new Weight text size setting under Settings → Appearance, with Standard, Large and Extra large options. It enlarges the weight number on these screens:
The text grows up to about 2×. When the phone's own text size is also large, the effective size is capped so the two settings don't multiply each other. The setting is saved on each device. Just Lift is unchanged, and the stop-button request (#1366) is tracked separately. This feature is developed voluntarily at no charge.
Gate audit:
maingained two more commits (fix: keep every rest-timer control tappable at narrow widths (refs #1360) #1411, fix(workout): publish Idle on manual summary continue so SetReady can navigate (#1413) #1414). A local merge of this PR onto thatmainsucceeded with no conflicts, and the targeted rest-timer, weight-text and settings tests passed: 16 suites, 0 failures.The change will be in the next rolling release (v1.1.0 notes).