Skip to content

Latest commit

 

History

History
96 lines (68 loc) · 10.3 KB

VALIDATION.md

File metadata and controls

96 lines (68 loc) · 10.3 KB

Release validation

Creative Director development verification, 2026-10-26

The development checkout adds the portable compound skill, pinned Elaya/Taste sources, selected-source preparation, agent installers and native Hypermotion adapter. It is not included in the archived v1.1.1 ZIP.

  • Local Python suite: 118 tests, 116 passed; two Playwright cases skipped under -S. Browser tests run separately in CI.
  • Full library: all 36 sources, 924 catalogue entries and 147 media hashes passed library_check.py. Windows long-path checkout failures were resolved by scoped Git configuration, not a global setting.
  • Portrait HyperFrames demo: actual 1080x1920 render passed technical and source-provenance checks.
  • Five synthetic FFmpeg assembly modes passed. Regression tests reject short source excerpts, audio-masked missing frames and re-probing stale renders against edited sources.
  • Native Hypermotion adapter: nine simulated process/validation/provenance/cleanup tests passed. No native desktop export, MCP connection or platform build is claimed verified.
  • Portable skill frontmatter and relative links validated. Installer tests cover complete copies, unchanged reinstallation, preservation of user edits, profile selection and failed staging. Manifest-bundle tests cover source pins, hashes and refusal to overwrite mismatches.
  • Independent behavioral evaluation produced two Polish planning responses: a workshop website retaining the user's gradient and Inter, and a 20-second .hype brief retaining its requested engine with no paid services. They preserved supplied answers and reported missing runtime honestly. This evaluated planning behavior, not a finished site or native video. The observed missing native export workflow was addressed with the adapter and documentation.

No paid media requests, publication or external account configuration were used for these checks. Live execution in Claude Code/Hermes, native Hypermotion and subjective listening review remain separate environment/production checks.

Full ZIP publication scan correction, 2026-10-24

The downloaded v1.1.0 ZIP exposed a false positive in the source-tree publication scanner: its generated bundle-manifest.json contains upstream fixture paths and exceeds the source-file size limit. v1.1.1 excludes only this generated root inventory from source scanning. verify_bundle.py remains responsible for the inventory and payload integrity; Git-index scanning still checks any staged inventory. Two regression tests cover this boundary. The film and its media are unchanged.

Premium intro release, 2026-10-24

Fresh GitHub clone of ba6d8e1 on Windows, without copying node_modules or upstream checkouts from the workshop:

  • npm ci --ignore-scripts: 74 packages, zero reported vulnerabilities.
  • bootstrap_library.py: all 34 pinned sources downloaded successfully.
  • library_check.py: 920 catalogue entries and 147 media files passed integrity checks.
  • unittest discover: 68 tests, 66 passed and two optional Playwright skips.
  • All five FFmpeg smoke scenarios passed; packaging without review was rejected.
  • doctor: all required tools present, no missing skill paths. External accounts were not probed.
  • premium_intro.py --render: complete 900-frame render from supplied media, without generation API calls. Technical probe passed: 1920x1080, 30 FPS, exactly 30 seconds, H.264 and stereo 48 kHz AAC.
  • Fresh render SHA-256 ffc69b5c8c5646536de3653d6c010255ab7d4140684b10cebe1c930bd629d6bd exactly matches the published master on this machine. Cross-platform byte identity is not promised.
  • Both the downloaded master and fresh render decoded fully without errors. A contact sheet from the fresh render was inspected: scenes, captions, generated backgrounds and genuine Codex skill screenshots are visible.
  • Publication scan: 270 source files, zero findings. Media hashes include the MP4 and audio; a test rejects a modified MP4.

CI now reproduces the full intro on Linux/Python 3.13 in addition to existing Python 3.11/3.13 checks. Release archive integrity and downloaded-archive verification results are recorded in the v1.1.0 release notes. The complete bundle is built from committed Git trees, including all pinned upstream files.

The test does not claim that every upstream runtime, desktop capture or paid provider connection is configured. The intro uses animated genuine screenshots, not continuous desktop capture. Technical and frame review do not constitute a human listening review. The owner explicitly requested publication of this film and its source media.

Fresh Windows download audit, 2026-10-24

Independently cloned GitHub commit d4cdd31f84492e6c79e99e3f2fe9b356932dd12b into a new directory, without reusing the workshop's node_modules or tool checkouts.

  • npm ci --ignore-scripts: 74 packages installed, audit reported zero vulnerabilities.
  • bootstrap_library.py: all 34 pinned public upstream sources fetched successfully.
  • library_check.py: all 920 catalogue entries and 138 media hashes passed.
  • Existing suite: 63 tests, 61 passed, two optional Playwright skips.
  • All five FFmpeg assembly smoke scenarios passed, including original audio and ducking. Packaging without review remained blocked.
  • Fresh-clone portrait demo: 1080 × 1920, 30 FPS, eight seconds; technical probe passed.
  • GitHub CI for that exact commit completed successfully: https://github.com/aievolutionpl/social-motion-agent/actions/runs/40110001381.
  • Release v1.0.0 has the full 545,361,718-byte ZIP and checksum attached; GitHub reports SHA-256 e24ac90c450939878c6f72499be2d3f37a34bff7b39584d2821aa0b03897ca3c. This is the old release, not the new showcase.
  • The retained local ZIP matched that remote digest. All 23,200 files in its extracted copy passed verify_bundle.py with all 34 sources. The archive was not downloaded a second time.

On this machine, the first sandboxed render failed with EACCES on Chromium's loopback connection. Retrying outside the restricted Windows sandbox succeeded. This is an execution-permission requirement, not evidence that every Windows installation is identical. Paid accounts, private voice IDs, live desktop capture and optional upstream runtimes were not tested.

New showcase changes: 65 tests, 63 passed, two optional skips. Tests additionally verify a complete reproducible job, preservation of supplied intake answers, local assets, absence of unresolved template markers and refusal to overwrite an existing composition or accept the wrong format. Visual and audio approval are separate from these checks.

All seven portrait scene layouts passed studio.py polish, including the installed upstream taste/slop checks, at 2.5, 7.5, 12.5, 16, 23.5, 27.5 and 33 seconds. Screenshots were inspected; a contact sheet from the rendered draft was also inspected. The README montage is captured from the composition, not a generated interface mockup. A human listening review is still required before final delivery approval.

Checked 2026-10-23 on Windows, Python 3.13, HyperFrames 0.8.188, GSAP 3.15.0 and system FFmpeg.

  • Initial core suite: 60 tests, 46 passed, 14 skipped before the extended library was included. The current complete-library result is recorded below.
  • FFmpeg smoke: five successful modes (voice, music, ducked mix, original audio, original audio with music and a still). Unreviewed packaging rejected.
  • HyperFrames landscape: 1920 × 1080, 30 FPS, 8 seconds; technical probe passed.
  • HyperFrames portrait: 1080 × 1920, 30 FPS, 8 seconds; technical probe passed.
  • Actual frames and a portrait contact sheet inspected. Demo is intentionally silent; no narration quality is claimed.
  • README GIF is taken from the landscape render, not a generated product mockup.
  • Offline tests cover narration dry-run, missing credentials and sanitized HTTP failures. No paid provider requests were made.
  • Source audit excludes personal media, local account state, runtime caches, private upstream modifications and production outputs. Public upstream sources are now included as pinned submodules.

HyperFrames lint reports zero errors and three advisory warnings about grouping nested content into sub-compositions. The compact teaching example deliberately keeps its three scenes in one file; renders were checked. For larger films use separate sub-compositions.

CI tests Python 3.11 and 3.13 on Linux, performs FFmpeg smoke tests and renders the bundled demo. See the Actions badge for the current remote result. OpenScreen live capture, optional video-studio, paid services are not part of this release's end-to-end verification.

Complete-library expansion

All 34 exact commits were fetched successfully from their public upstream sources. The original 126 audio files, nine backgrounds and three fonts were copied without modification; SHA-256 values are recorded in assets/media-manifest.json. Seven local skills and 913 upstream catalogue entries are included. Media rights are documented separately.

Complete-library verification: 63 tests, 61 passed and 2 optional Playwright skips. All 920 catalogue files and 138 media hashes validated; all 34 source commits matched. Original audio index check and motion catalogue search passed.

Kinetic Film integration validation

The kinetic-launch integration was exercised locally on Windows with the pinned external package, Node 26.12.1, Python 3.13, NumPy 2.2.3, OpenCV 5.0.0 and Playwright 1.59.1. A full 30-second reproduction through studio.py render-kinetic produced 1280×720, 60 FPS, 1800 frames, H.264/AAC and -14.0 LUFS. Motion gates passed: 15.9% still frames, median flow about 4.8 px/frame, longest hold 0.55 s before the actual Logo block. Full FFmpeg decode, source receipt and probe passed. Sampled render frames were inspected.

15 focused unit tests cover source integrity, allowlisted copying, preservation, path/lock guards, optional audio, motion rejection, source changes during rendering and memory limits. The complete local suite passed 142 tests with seven optional browser tests skipped in that environment. Synthetic FFmpeg smoke scenarios passed. Windows CI includes test_kinetic; Linux suites discover it automatically.

The external engine is not downloaded by CI or included under MIT. Actual native exports on Linux/macOS, complete 1080p/4K exports, destination-device playback and independent full listening review are not claimed. The README film has owner publication authorization, recorded separately from review and asset terms.