Tide — Project Status

Tide is a macOS + iOS SwiftUI net worth tracker: an editable list of accounts, balances logged by hand, and a total tracked over time with expressive-but-honest visuals. No bank connections, no forecasting.

Status at a glance

A theming and overview-visualisation branch is complete and has passed a full whole-branch review, with all review findings fixed. It is not yet merged — the owner's merge gate requires visual confirmation against real data, which is outstanding. 187 automated tests pass across four pure packages and both platforms build.

Current phase

Phase: Theming & overview charts — feature branch complete, post-review, awaiting the owner's visual sign-off before merge to main.

Latest verified checkpoint: Whole-branch review returned READY WITH FIXES; every Critical and Important finding was fixed and re-verified (commit fix: address final whole-branch review).

Recently completed

• New pure package TideTheme: colour maths, a 21-theme catalog, and account-type band assignment. [verified]
  • 35 tests pass via swift test. Contains no SwiftData and no SwiftUI, which is what makes it testable on the current toolchain.
• Accessibility guarantees run as executable tests, so an inaccessible theme fails the build rather than reaching a user. [verified]
  • Gates cover WCAG AA text contrast, colourblind separation (Machado-Oliveira-Fernandes simulation, gated on min(protan, deutan)), diverging-pair separation, and band uniqueness — parameterised across all 21 themes.
• Overview accounts are sorted; previously they rendered in undefined, rebuild-unstable order. [verified]
  • Pure comparator in the domain package with 9 tests. Assets group before liabilities in every order, unobserved balances sort last rather than being treated as zero, and the sort is stable.
• Theme picker with a System option that follows the OS appearance, synced across devices. [verified]
  • System is the default, so the app looks native until a theme is chosen. Sync uses iCloud key-value storage rather than the app database — no schema change and therefore no migration risk to account data.
• Account-type colours drive both the account bars and the composition view. [verified]
  • 12 of 21 themes carry bands derived from their own accents; the remaining 9 use a shared validated palette because their palettes are too harmonious to separate 8 categories.
• Overview shows a net worth trend line plus a composition strip broken down by account type. [verified]
  • Replaced an earlier stacked-area chart. Recorded reversal: a stacked area needs a value for every band at every date, and the project's honesty rule forbids inventing unobserved values — so an account added part-way through drew a cliff. The strip reads a single snapshot, so adding an account only changes proportions.
• Carried-forward balances are visually distinguished from observed ones in the composition strip. [verified]
  • Carried segments render dimmed AND carry a text label, because dimming alone would be a colour-only signal.
• In-app licence acknowledgements for the upstream open-source colour palettes. [verified]
  • All upstream palettes used are MIT-licensed; reproducing the copyright lines and permission notice is a licence obligation before release.

In progress

• Merge of the theming/charts branch into the main line. [observed]
  • Branch is complete, reviewed, and fix-verified. Held at the owner's explicit gate pending visual confirmation.

Remaining work / roadmap

• Owner visual confirmation of the composition strip against real account data. [planned]
  • Raised to a gate because sample data has already missed one real defect this cycle — see blockers.
• One upstream theme remains excluded: its accent colours cannot reach the required contrast on its own near-white surface. [observed]
  • Can be added once a darkened colour is chosen that passes the existing gates.
• Asset-band separation is gated on slot order only, a weaker guarantee than the liability gate. [observed]
  • Five categories all-pairs is a bar no palette in the catalog clears. Accepted because every surface labels each segment, so identity never rests on colour alone. Documented in the test itself.
• Deferred minor findings from review remain open (test-coverage gaps and cosmetic items). [observed]
  • Triaged as non-blocking during the whole-branch review; tracked in the branch's execution ledger.
• Mac App Store deploy steps remain outstanding: production push entitlement and CloudKit schema deployment to production. [planned]

Blockers, decisions, human gates

• HUMAN GATE — no merge to the main line until the owner confirms the overview visually against real data. [user-reported]
  • Set by the owner, who confirmed the visual check counts as one of the required checks. Automated tests and green builds alone do not satisfy it.
• Sample-data verification is necessary but demonstrably not sufficient. [verified]
  • The seeded sample data has every account present from the first snapshot, so it structurally cannot exercise accounts added later. A real-data case surfaced a band that rendered nothing at all; the sample data had shown the same code as correct throughout.
• Automated tests cannot cover the database layer or UI on the current toolchain. [verified]
  • Any database-container operation aborts the whole test run, so a passing line with zero executed tests is a false green. Pure packages carry the real coverage; database writers and UI are verified by inspection and screenshots.

Verification evidence

• 187 automated tests pass across four pure packages. [verified]
  • Domain 108, formatting 37, theming 35, store 7 — each run fresh this session via swift test.
• Both platform builds succeed from the command line. [verified]
  • macOS and iOS Simulator destinations, after project regeneration.
• Data safety confirmed by independent review of the full branch diff. [verified]
  • No database-layer files touched, no schema change, no migration. The pinned cloud container entitlement survived three project regenerations and was re-checked afterwards.
• Overview, theme picker, and composition strip confirmed by screenshots of a running signed build. [verified]
• A correction was made to this project's own analysis during the cycle and is recorded rather than quietly fixed. [verified]
  • An earlier root-cause attribution for two rejected themes was wrong: the deciding factor was a threshold definition, not the colour-vision model. Both the reference document and the execution ledger were corrected.

Project documents

• Project guidance: CLAUDE.md
• Colour theme reference (palettes, licences, validation): docs/reference/color-themes.md
• Implementation plan (with superseded sections marked): docs/superpowers/plans/2026-07-19-theming-and-overview-charts.md
• Execution ledger (per-task reviews, deviations, reversals): .superpowers/sdd/progress.md

Freshness & provenance

• Last updated: 2026-07-19 19:07 UTC.
• Evidence gathered from a working tree with no uncommitted changes, on the feature branch, immediately after the whole-branch review fixes were committed.
• Test counts and build results are from commands run in this session, not from documentation.
• The implementation plan carries explicit SUPERSEDED markers where the shipped code deliberately diverges from it; the plan is a historical record, not a description of current behaviour.
• Nothing here is merged to the main line yet.
• Evidence labels: [verified] fresh commands/reviewed commits · [observed] in code/git, not runtime-validated · [user-reported] confirmed by owner, not independently verified · [planned] spec'd, not built · [stale] docs conflict with current state.