Changelog

All notable changes to Telltake. Dates are the day the work landed in the repository; the version is the app's CFBundleShortVersionString.

1.0.1 — 2026-09-23

The first downloadable build, with a redesigned workspace. 1.0.0 was the source freeze and was not distributed.

• New look. A shared navy and teal theme with a sidebar using the app icon and system symbols, applied to the project and review windows.
• Review window layout. Preview on the left, transcript on the right, equal by default with an adjustable divider that is remembered.
• Grouped transcript. Contiguous lines are grouped under the name of the clip they come from; the headers are presentation only, and a line that spans two clips stays one edit target. Inline Restore on removed lines.
• Reorganized controls. Filters and actions moved into the toolbar and popovers; AI Rough Cut, Line ends and Text & rough cut are disclosure sections.
• No change to editing rules, saved projects, cloud requests, approval or export. 429 tests pass, including new tests for the grouped transcript.
• Distributed as an ad-hoc signed DMG; see the first-launch note on telltake.com.

1.0.0 — source freeze, 2026-09-22 (not distributed)

The first release: a macOS app that turns a multi-clip talking-head recording into a reviewable rough cut by editing its transcript. Earlier history (before the first commit on 2026-09-18) lived outside version control; the decisions it produced are ADR-001 to ADR-105 in docs/decisions/.

Editing workflow

• Import clips, order them by capture time, leave clips out without deleting anything, and confirm the sequence through a media compatibility gate (dimensions, aspect ratio, frame timing, audio, orientation, colour format).
• Transcribe with Doubao Turbo, clip by clip, with durable per-clip checkpoints, up to two clips at a time, saved progress with an estimate, and a Stop that keeps finished clips (ADR-106, ADR-110; 2026-09-19).
• AI Rough Cut: the transcript text goes to GPT-5.6 Sol, which proposes the continuous passages to keep; every quotation is validated against the saved transcript and anything that cannot be placed exactly is kept and marked Needs review with its reason (ADR-107; 2026-09-19).
• Gaps enclosed by removed speech are removed by default with a warning and a restore, including gaps under two seconds; Remove enclosed gaps applies the rule to a saved cut without another request (ADR-109; 2026-09-19).
• Transcript review: keep or remove lines and gaps, take groups, line ends snapped to real frames, AI cut edge adjustment, text corrections with revert, manual pause shortening, previous/next join playback, persistent undo and session-scoped redo, search and filters, playback highlight that follows the cut (ADR-078 to ADR-088, ADR-108).
• Full-cut preview with coverage tracking, explicit approval of the exact plan, and export of that captured plan to an MP4 rendered from the original clips (ADR-089, ADR-091).
• Project backup of canonical state, recovery status, and relinking of moved clips by fingerprint (ADR-092, ADR-075).

Release hardening (2026-09-20 to 2026-09-22)

• Cloud: response waits sized from clip duration; key validation bound to the shipped model; model-access denial keeps the key and names the model; the GPT-5.6 Luna analysis client removed and Story Cut generation hidden; experiment-only provider clients moved out of the app.
• Reliability: all project database work serialized on one queue; damaged database columns and unmigrated projects reported instead of trapping; a current cut that fails validation reported instead of hidden; audio extraction failures recorded per clip; every Creator edit tied to an open editing session.
• Desktop: standard menu bar with Undo and Redo; confirmation before closing or quitting during transcription or export; the review window reused while open; sanitized diagnostic codes with Save Diagnostics…; the review window split into focused controllers; export ownership moved to the project window.
• Copy and localization: every user-visible string in plain English through a string catalog, with Core carrying codes instead of sentences; errors from every module explained in plain language; Chinese planned after v1.
• Media: HDR and 10-bit sources exported as 10-bit HEVC with the source's colour tags, and HDR never mixed with SDR in one sequence; verified on generated HLG footage and one real iPhone HDR recording (schema migration 7).
• Packaging: renamed to Telltake with an app icon; the .easycut document type registered so projects open from Finder; universal release build, inside-out signing, smoke test and notarization script; bundle identifier com.telltake.app, version 1.0.0, build number from the commit count.
• Documents: CONTEXT, product brief, glossary, ADR index, ticket file with named verification, provider and schema statements, these terms, this changelog and the privacy document brought in line with the shipped product.

Known limits at release

See "Current scope and limitations" in the README. In short: one speaker, one camera, Mandarin-first; English interface; no Story Cut; quality figures are targets for a future held-out evaluation, not measured results.