earthUI

• Save and restore are **server-side and DB-authoritative**: configuration is written to / read from projects.sqlite directly from the inputs at the render and click, with no localStorage round-trip. This removes a class of restore races and a JSON double-encoding bug that could corrupt a saved row on the next save.
• The **Effective Date** field is shown in all purposes (general/CMA and market values have an effective date too, not just USPAP/IVS appraisals); it is still only used to compute sale_age in appraisal/market modes.

Database durability

• projects.sqlite opens in **WAL** journaling mode where supported, for crash-safe writes and non-blocking reads.
• The settings-table schema **migration runs in a single transaction** with crash recovery: an interrupted upgrade rolls back to the original table (no orphaned _old, no half-copied data), and a previously-stranded _old table is restored on the next connect. Legacy per-file rows are collapsed to one row per (project, purpose), keeping the most recent.

Bug fixes

• Predicting on raw data with a **Date column as a model predictor** no longer errors ("subscript out of bounds" / column-count mismatch). Newdata is now aligned to the trained frame (align_to_training_): date columns are coerced to the model's numeric encoding and factor levels are matched. Affects the Intermediate Output and RCA exports.
• fit_earth() drops a non-zero newvar.penalty (with a message) when case weights are present, since earth does not support that combination — previously it errored mid-fit (notably in market/appraisal mode where a skipped row is encoded as a zero weight).
• Character date columns are parsed with the same multi-format set as validate_types(), so any column that validates as a date also coerces (no more silent failures on MM/DD/YYYY and similar).

Project model (regProj)

• New first-class **project** concept replaces the old per-file workflow. A project is a stable, named entity at a fixed location in the regProj tree (<root>/<purpose>/<flat_segment>/<os>_in/... and <os>_out_<method>/...), holding all input files, outputs, and settings together. Files within a project travel as a unit.
• Top-of-sidebar Project picker with sort (Recent / A-Z) and a New Project wizard that captures purpose, country, and admin-level cascade with country-specific terminology.
• Path layout is **flat**: <purpose>/<country>_<state>_<county>_<city>_<project>/<os>_in/<file>. All admin levels concatenated into one segment via _. Reduces depth from 13 to 8 directories under the regProj root. Encoded by regproj_flat_segment(), decoded by regproj_parse_flat().
• Cross-OS multi-leaf scaffold: every project gets mac_in, mac_out_*, ubuntu_in, ubuntu_out_*, win11_in, win11_out_* siblings.
• Active-project model drives output paths automatically. The legacy Output Folder field and Project Location cascade are gone — output folder is set from the active project at session time.
• Section "Import Data" reads from the active project's <os>_in/; files added manually are picked up via Refresh.

Geo data + variable-depth admin schema

• New geo.sqlite (in regProj root) replaces the old .regproj-index.json. Two tables: countries and admin_entries with variable depth via (country, level, parent_codes) index. Supports any number of admin levels per country.
• Comprehensive shipped data:
• **US**: 51 states, 3,076 counties, ~21,000 incorporated places (Census).
• **GB / DE / IT / FR / SE**: full GeoNames-derived state / county / city data (~45,000 entries combined).
• **SG**: 1-level (planning area) data for Singapore.
• Total ~70,000 admin entries shipped; users can add more via the cascade's "create new" path. regproj_index_get/put API unchanged for callers, just SQLite-backed now.
• DE specifically uses admin3_code (Kreis) instead of GeoNames' admin2_code (Regierungsbezirk), matching how German appraisers think about administrative geography.

Per-project settings (projects.sqlite)

• New projects.sqlite in regProj root replaces the user-level settings.sqlite for project model settings. Keyed by (flat_segment, file_basename), separate JSON columns for earth, glmnet, and mgcv settings/variables/interactions.
• Public API: get_project_settings() and set_project_settings() — read/write project settings programmatically without launching the Shiny UI. Designed for ValEngr integration and batch automation.
• Settings now travel with the regProj tree (sync, backup, share with colleagues — all settings come too). User-level settings.sqlite retained only for locale defaults.

Quarto report split: Generate + Convert

• **Section 9 — "Generate Quarto Report"**: writes a self-contained .qmd bundle (source + plot assets + reference.docx) under <project>/mac_out_earth/<base>_qmd/. No Quarto rendering happens at this step.
• **Section 10 — "Convert Quarto Report"**: file picker + format checkboxes (HTML / Word / PDF). Renders any .qmd file (not just earthUI-generated) to selected formats. Useful for combining reports across projects via Quarto's {{< include >}}.
• New exports: generate_quarto_report(), convert_quarto_file().
• report_data.rds is now a lean ~hundreds-of-KB asset (down from hundreds of MB) — the bulky earth model object is no longer saved since the qmd reads pre-generated plots from disk.
• Quarto .qmd template now self-renders against a sibling report_data.rds when no params are passed, so quarto preview Appraisal_1.qmd works standalone.

Settings tab

• New "regProj Root Folder" config in the Settings dropdown — set the location of the regProj tree per machine. Persisted in the per-OS prefs file at R_user_dir("earthUI", "config")/prefs.json.
• Resolution order: REGPROJ_ROOT env var → prefs file → per-OS default (~/regProj Mac/Linux, C:/regProj Win).

Other

• Server startup extends PATH with /usr/local/bin, /opt/homebrew/bin, and TinyTeX bin paths so Quarto / pandoc / latex are findable from the Shiny session even when launched from a stripped-PATH context.
• Per-session DB connections (one geo, one projects) — eliminates per-render connection overhead.
• Sales Comparison Grid now exported as build_sales_grid() from the package (extracted from the old inst/app/sales_grid.R). Available to batch scripts and ValEngr.