Version bump, the README line about units, and two new manual checks in the release checklist (export then New invoice
through the real save dialog, and units).
Claude-Session: https://claude.ai/code/session_01PZypiWDfMkDTeEPeXjRhW5
Quantities can now be time, count, distance, area, weight or volume, or a custom unit, instead of only
second/minute/hour/session/unit.
- Registry in src/lib/units.ts (mirrored in src-tauri/src/units.rs): grouped built-in units plus a custom unit of 1-12
characters. A rate line's unit is validated when an invoice or a preset is saved; migration M12 drops the old unit
allow-list from item_presets and keeps every row.
- Time is typed as h:mm: "4:30" hours counts as 4.5, "4:20" as 4.333... (kept exact, so rate x quantity rounds once, the
same in TypeScript and Rust). Other units take plain decimals up to 3 places. A bad quantity blocks issuing.
- The PDF quantity column shows the unit ("4.5 hr", "12.75 km"); the rate keeps "per hour" / "/hr". RenderItem carries
priceText, perText and rateShort, so no template parses the rate text with a regex over five hard-coded units.
- Quantity columns are sized from the widest word once a quantity with its unit is wider than 72 pt, so one long
custom unit cannot squeeze the description (the serenity and citrus-split tables overflowed in the new fixture).
Classic's quantity column is now content-sized.
- ERPNext: the default UOM map comes from the registry, with fractional-capable UOMs for time, distance, area, weight
and volume. A custom unit is sent as Nos. A quantity that is not exact to 3 decimals (4:20 hours) is still refused for
push, as before.
- Goldens and template thumbnails regenerated; the template harness has a "units" fixture with the longest texts.
Claude-Session: https://claude.ai/code/session_01PZypiWDfMkDTeEPeXjRhW5
The first-run seed no longer inserts the original vendor's name, address, contact
details, PAN or bank account; the wizard now starts empty and requires a name and
state. Settings > Data gets a Run setup again button that clears the onboarded flag.
Re-running the wizard keeps the active invoice series unless the prefix or padding
changes. The sample logo moves out of public/ so it no longer ships in the app.
- Bump the version to 1.0.0 in package.json, Cargo.toml, tauri.conf.json and the Void template. Fingerprint goldens are unchanged.
- Rewrite the README: features, data paths on Linux and Windows, backup and restore (backups contain the ERPNext API secret), bundled fonts and licences, font overrides, the ERPNext recipe, build and packaging commands, gates and known limitations.
- Add docs/RELEASE.md with the release steps and a manual first-run checklist.
- Void template: use a plain npm ci (the optional-dependency omit breaks vite and tsc), and note vendoring and that no OpenSSL is needed.
Verified: build-xbps.sh builds voiced-1.0.0_1; it installs and removes cleanly in a clean void-glibc container with fonts, licences and fc-cache handling in place. The cross-compiled NSIS installer is 11.45 MiB and has not been run on Windows.
- Settings gets an Integrations tab: connection fields with a write-only API secret, a CA PEM option, Test connection with versions and warnings, live mapping lists loaded after a test, naming mode, toggles and its own Save.
- Pushing for a GST-registered vendor is refused until a connection test result is stored. The UI shows the reason and disables Send.
- History shows an ERPNext status column from one bulk call, a row Send and Open in ERPNext, and a bulk Send selected with a draft-or-submit confirm step and per-row results that never abort the batch. The detail view gets an ERPNext panel and per-payment send. All of it is hidden or disabled when the integration is not configured.
- Optional auto-push after issue runs in the background and never blocks or fails the issue.
- Generic export: a deterministic voiced.invoice.v1 JSON (and CSV line items) for selected invoices, documented in docs/voiced-invoice-v1.md with a golden test. Money is integer paise plus a decimal string.
The new screens have not been run in a webview yet. The ERPNext Payment Entry deduction fields still need a live check.
When only the closing block moved to page 2, the page-1 deficit was shared across every gap, including the closing block's own gaps, which cannot help page 1. Page 1 then came out a few points short and the totals spilled over with the closing block. Monolith and Cobalt Stripe now squeeze only the page-1 gaps in that case, as Marble and Highlighter already do.
Monolith's flow notes block could start in the last few points of a page and be cut into zero-height lines. It now holds 70 pt of room under its label first, as Marble does.
Adds tests for both cases (they failed before the fix) and notes in the template specs. Fingerprint goldens and thumbnails are unchanged.
Marble uses a flat-to-gradient band in place of the marble photo and bundles Lustria (OFL).
Both logo slots are opt-in. Quirk 13 documents a fontkit glyph-cache issue that the
tests work around with warm-up renders; the engine fix follows separately.
Standalone layouts for reference templates 3 and 8, with the user's logo sized to the
placeholder box, a CapRule decor primitive, optional TableRows rule/heading options,
and a separate 1..45-row sweep test file.
Purple Pop (cream page, purple pills, asterisk and dome decor, opt-in
logo) and Citrus Split (grey split panel, halftone chevrons, icon
contacts, logo in the measured slot) are rebuilt from the reference PDFs on
the engine with A4/Letter support, GST slots and zero audit errors. Adds
static DM Sans, Jost and Poppins 800 as look-alikes for the commercial
fonts. The picker now lists ten templates.
Crimson Grid, Tangerine Ledger, Slate Band and Teal Swoosh are rebuilt from
the reference PDFs with grid and band tables that survive page breaks,
full-bleed fixed bands, pennant tabs, gradient corner swooshes with
keep-outs, Carbon icon circles, and the user's logo in the measured slot
(knockout or white chip on the dark Slate band). Adds static Poppins
500-700 and Inter fonts. Thumbnails regenerated for all eight templates;
the item-count sweeps run in their own test file.
Linea (the vendor's original format.pdf design), Monogram and Serenity are
rebuilt from the reference PDFs as one parametric family on the engine, with
the user's logo in the measured slot, a home for every GST field and
A4/Letter support. Adds static Montserrat, Open Sans and Poppins fonts and
a generated Ms Madi outline for Serenity's vertical word. Text-only logo
placeholders (Monolith, Tangerine) use a doc-derived allowance box. A
Carbon TemplatePicker with committed thumbnails opens from the page
setup controls.
xavierk