What's new.

Two related fixes for dual-system documentary workflows:

• Compose Timeline exports are now frame-accurate. Paper Edit → Compose was producing Timelines without multicam-angle data, which made Resolve / Premiere / FCPXML exports drift — audio extended past video on the timeline. Compose now runs the same per-file timecode assignment that normal mines use. (Existing Compose Timelines need to be Recomposed to v2 for the fix to apply; new Compose runs land correct.)
• Resolve-synced Sources now retain video file paths. Sources ingested via Resolve sync were persisting with videoPaths: nil, leaving the Hi-Ho video player black and falling through to audio-only playback. Future syncs land with full video linkage; existing broken Sources can be re-synced to self-heal.

Both fixes carry over the Tahoe + M1 bundled Python from 1.0.156 — no Python regression here.

Critical fix for users on macOS Tahoe (M1, M2, M3, M4). The bundled Python runtime that powers Resolve integration was unintentionally dropped from the build script several releases ago — the regression was hidden because older macOS shipped with /usr/bin/python3 as a fallback. Tahoe removed that fallback, exposing the missing dependency and causing launch crashes on affected machines.

• Python 3.12.13 bundled inside the .app (via astral-sh/python-build-standalone) — no more dependency on system Python
• Resolve integration works again on Tahoe
• Launch crash on Tahoe resolved
• Universal binary (arm64 + x86_64) — Intel Mac support restored alongside Apple Silicon

If you experienced a crash on Tahoe with 1.0.155 — this is the fix. Auto-update will apply on next launch.

Paste a paper edit, get an assembled Timeline. Hi-Ho cross-references quotes against every Source in the Project, lets Claude assemble the order, and produces a versioned Timeline you can re-prompt — edit the paper edit on the loaded Timeline, hit Recompose, and v2/v3/v4 chain in the same lineage.

• New "From Paper Edit" tile on every ProjectView
• Full-screen paste sheet with live section-count detection
• Inline paper-edit editor on every Compose-derived Timeline
• Timeline version chip (v1/v2/v3…) + lineage menu on tiles
• B-roll/non-dialogue directions become subtitle placeholders so paper-edit structure stays 1:1 in the export
• Silent schema bump v2 → v3 on first launch; legacy projects unaffected
• Pre-mining Paper Edit field retired (StylePage cleanup)

• Security: API keys moved to server-side proxies — no more secrets compiled into the app binary. (Previous versions are now broken until updated; old keys have been revoked.)
• Multi-cam sync via Resolve timeline — audio ↔ video pairing survives free-run timecode mismatch between cameras and audio recorders.
• Network retries — Anthropic / Groq 5xx and 429 are now retried automatically with backoff (3 attempts).
• Audit + speaker results persisted — reloading a project no longer re-calls Claude. User edits to speaker names survive.
• Auto-relaunch on update fixed — proper detached-shell handoff so the new version takes over without manual reopen.
• Multi-instance Resolve error visible — toast banner instead of silent failure.
• Disk hygiene — auto-cleans stale /tmp media copies and stale chunks on app launch.
• Mining robustness: filter_complex audio merge handles mixed codecs (no more 32GB looping garbage), fast-seek channel probes (50× faster on long .MOV), parallel PCM extraction, output sanity check (rejects truncated/corrupt files), Otsu threshold fallback (no crash on missing PCM).
• Order canonicality: mediaPaths order flows through every step — addable catalog, energy profile, Resolve build script — so chat-added clips and the Resolve timeline always reference the right source file.
• Stale-cache invalidation: transcript cache key includes file size + mtime — replacing a source file no longer reuses an old transcript silently.
• Telemetry: categorical events (mine start/complete, sign-in, Resolve send, errors) without content — helps us catch silent failures.
• UI: sparkle logo replaces the pixel-axe wordmark on sign-in / opening; Settings, Account, Update, Feedback panels are now click-outside-to-dismiss with a unified rounded panel chrome; "Include diagnostics" toggle in Send Feedback (categorical state only — no transcripts or paths).

• Multi-Cam Sync from Resolve Timeline — Audio and video clips paired by their position on your Resolve timeline, surviving free-run TC mismatch between cameras and audio recorders. Mined edits land in Resolve with the right video on V1 alongside ZOOM mic audio on A1.
• Correct Source-Clip Alignment — Whisper segments now map to the right source file (was using TC-sorted order against a mediaPaths-ordered merged audio).
• Sticky Sign-In — App and browser sessions no longer fight over the same refresh token. You stay signed in indefinitely. Cached identity stays visible if a session ever does expire.
• Parallel PCM extraction — Energy profiling for 4× source files runs concurrently (was sequential).
• Fast seek for channel detection — volumedetect probes use ffmpeg fast-seek (-ss before -i + -vn). 50× faster on long .MOV files from external drives.
• Robust filter — Energy filter no longer nukes all segments when PCM extraction is incomplete.
• filter_complex concat — Audio merge handles mixed codecs (WAV + MOV) cleanly. No more 32GB garbage files.
• No auto-build to Resolve — Mining produces the segments and you trigger the Resolve build when you're ready.
• Speaker dedup — Sidebar shows one entry per unique speaker, not one per source clip.

• Audio Extraction Timeout — Bumped from 2 min to 15 min. 4-track interview merging from external drives now completes
• All 1.0.146 Fixes — Resolve bridge entitlement, auto-relaunch on update, auth refresh, mine-once catalog

• Resolve Bridge Fixed — Bundled Python now signed with library-validation entitlement so it can load Resolve's API
• Auto-Relaunch on Update — App now properly closes and reopens on the new version (was foregrounding old instance)
• All 1.0.144 Improvements — Auth refresh, Resolve auto-connect, mine-once catalog, story-order playback, waveforms

• No More Hourly Logouts — Refresh tokens keep you signed in across sessions
• Resolve Auto-Connect — Picker now connects automatically when opened
• Mine-Once Catalog — AddableSegment catalog built during mining for instant clip adds
• Story-Order Playback — Auto-advance follows your edit order, not source time
• Stable Captions — Captions on chat-added clips stay locked to the right words
• Scrubber Accuracy — Clicking timeline jumps to the actual clip you clicked
• Waveforms on Timeline — Main timeline shows the audio shape of your edit
• Performance — Faster transcript merge, smoother playback at 30Hz

• Resolve Export Fixed — Clips land on timeline with correct durations + subtitles on subtitle track
• Premiere Pro Export — Full clip import via CEP extension with chapter markers + SRT captions
• Multi-Instance Detection — Warns when multiple Resolve instances could cause wrong-project connections
• Proportional Caption Timing — Subtitles distribute evenly by character count (no more drift)
• Transcript Click Accuracy — Click any word in the transcript and it selects correctly
• AI Chat: Add to Timeline — Clips added from Story Chat now have working karaoke + playback
• Transcript/Caption Sync — Video captions and transcript column always show the same text
• Chapter Pills Wrap — Chapter overview wraps into rows instead of horizontal scroll
• Folder Drop Support — Drop folders into New Mine dialog, recursively finds media files
• Sync Companion Panel — Multicam sync details in a dedicated side panel
• New App Icon — Asterisk/starburst gradient design
• Search Shows 30 Results — Up from 10 in Story Chat