Adds a Layout Picker with persisted profiles (including 'Popup Layout'), moves layout persistence into the picker and initializes it first. Simplifies dream-task internals (separates visible vs generation counts), avoids loading every task by hydrating graveyard entries via getTask, adds scroll-to-current-hour and a Task Retirement 'Refresh', plus UI/bug fixes and tests. Key files: dream-task-internals.js, dream-task-service.js, graveyard-service.js. Impact: fewer unnecessary loads and improv...

Adds lib/graveyard-service.js: a Graveyard widget service that scans notes for open, non-excluded tasks, selects aged candidates using a multi-pass filter (6‑month, 3‑month, fallback), rehydrates cached UUIDs, and reads/creates a monthly archived note named like "Graveyard tasks of " containing a markdown table of candidate UUIDs. Includes error logging and note serialization to surface old tasks for monthly review.

Expose a throw_error flag across LLM metrics ExternalUsageClient and provider clients (ExternalAugmentCodeUsageClient, ExternalCursorAiUsageClient). Updated stat_dates_between and get_with_retry signatures to accept throw_error and re-raise on Unauthorized/TooManyRequests when requested while still recording failed token requests. Enables post-token v1 validation to fail fast and surface upstream auth/rate-limit errors for diagnostics and retry logic.

Introduce ai_defect_line_triaged_count to measure lines changed by AI that were later triaged as critical defects. Implemented by adding a calculation branch (CodeLineTriage.count scoped to AI-changed lines) and documenting the metric in defines/segment_purpose.rb. Deployed to production May 5, 2026. Enables tracking AI defect rate for ROI and code-quality monitoring.

Persist Recent Notes per-day selection, scan progress and visit history into a daily archived dashboard note (lib/recent-notes-service.js / lib/dashboard/recent-notes.js). Add deterministic date+reseed seeding with a simplified hash to avoid iOS Safari hash issues, cap discovery/scan and history sizes, hydrate persisted selections before render, and apply small DreamTask/mobile CSS tweaks — yielding predictable rotation, stable seeding, and reduced mobile scan/load.

Add new test file test/call-plugin-fallback.test.js that scaffolds calls to the Ample Agent used by the dashboard, including mock app and fetch helpers and scenarios exercising plugin fallback behavior. Provides coverage for AI-agent integration and helps catch regressions in dashboard plugin calls.

Group compare_stat_segments into pairs in the ai_roi_stats HAML view and render each segment with a CompareStat component; update CompareStat stylesheet selector and add responsive/flex rules, add a new _ai_roi_stats.scss to style pair containers and spacing, and a tiny report_tabs ordering tweak. Result: cleaner, responsive AI ROI comparison layout with consistent styling and spacing.

Replace the legacy imperative commit-value editor with a React-based CommitDiffDeltaEditor (app/javascript/apps/commit-diff-delta-editor.js plus SCSS). Updated controllers, models, views and tests to support inline and modal delta editing, client-side validation and AJAX saves, and to expose configurable human vs LLM authorship percentages — improving edit UX, real-time validation, and authorship attribution.

Updated day-sketch to reliably strip HTML and Markdown when converting task content to plain text (lib/dashboard/day-sketch.js) by using a new stripMarkdown utility in lib/util/utility.js. Added Shift+Up/Down handling to extend selection on the current-hour input. Tests (test/day-sketch.test.js) and AI_CONTRIBUTIONS.md updated. Result: cleaner prefills and improved keyboard selection UX.

Implemented keyboard handling for up/down arrow keys and reorganized DaySketchWidget to break a monolithic function into focused methods (entriesPrefilledFromTasks, entriesAreDirty, useDaySketchEntries, prefillTextFromTask) in lib/dashboard/day-sketch.js. Cleans up prefill/dirty logic and improves maintainability while enabling arrow-key navigation in the UI.

Fixes PostRatingSection so the app sends one randomly selected response message rather than the entire responses array. Updated components/mood-rating-screen/post-rating-section.js to pick a random entry when constructing the post-rating message. Deployed to production; users now receive a single concise feedback message after submitting mood ratings, removing prior confusing multi-message output.

Implements Dream Task lifecycle actions (preserve-for-days, mark-complete, remove) with updates to dream-task internals, service code and tests; refactors dev-mode prompt UX (mocked app.prompt, consolidated prompt/form helpers, split reseed labels, LLM default tweaks, prevent Peak Hours scroll); and adds platform-specific API-key link hints plus provider metadata in dashboard-settings-popup.js and constants/llm-providers.js so retrieval links open reliably from sandboxed frames. Improves task...

Expose selectable LLM options in the analyzeDreamTasks prompt/build flow (lib/dream-task-service.js) so when multiple models are available you can pick which model to use during a reseed. Also adjust the default widget wrapper style to allow scrolling. Impact: reseeds can target a specific LLM and the UI no longer clips overflowing content.

Remove legacy CommitValue flags (pending_review, rejected) and replace them with an inferred AI/CI assignment: commits with exceptional velocity are marked as LLM-authored CI runs. Changes live primarily in app/lib/commit_concerns/value_evaluation_concerns.rb (added deduce_ai_assignment_state and updates to state derivation). Result: simpler commit-state model and more accurate attribution of AI/CI-generated commits.

Update app/models/commit_value.rb to remove pending_review and rejected states and introduce pending_ai_assignment and llm_complete. Rename and add callbacks (detect_zero_pending_ai_assignment, sync_pending_ai_assignment_values, handle_pending_ai_assignment_complete), expand included_in_stats, and adjust approved_value, impacting? and derive_number_authority_em to treat exceptional-velocity commits as LLM/CI-authored. Result: LLM-generated CI runs are identified and correctly included in stat...

Relocate the Replace Markdown action so it can be invoked with one click from the selection menu. Changes in app/javascript/lib/ample-editor/components/selection-menu.js reorganize the ConvertFromMarkdown button and surrounding menu logic, removing the extra step. Result: quicker, more discoverable markdown replacement and improved selection-menu UX.

Detect markdown-like text in clipboard input and run it through the project's markdown parser (clipboard.js) instead of treating it as plain text HTML. Added logic to preserve blank lines as empty paragraphs and convert headings, bold and list markup into ProseMirror nodes, and added a between-apps test verifying OpenAI/Phind markdown is transformed correctly. Result: more faithful pastes from AI tools into the editor.

Detect full HTML clipboard copies and tag

elements so pastes from external pages insert empty paragraphs between adjacent paragraphs; parse pasted plain-text Markdown (e.g., from AI tools) into ProseMirror nodes. Primary changes in lib/clipboard.js with supporting tests (lib/clipboard.test.js, clipboard.between-apps.test.js). Improves paste fidelity and preserves paragraph separation.