v0.3.0

binquery v0.3.0

v0.3.0 adds one reproducible command for trying the complete local workflow without
providing footage.

Highlights

• Added binquery demo --out <directory>.
• The demo generates a 30-second ffmpeg lavfi test video locally, then runs
  split -> index -> doctor -> query and writes the clips, index, and query.json
  under the requested directory.
• The output directory must be new or empty. Existing files are never overwritten.
• The GitHub Actions real-model smoke test now invokes the same public demo command
  instead of maintaining a separate shell-only workflow.
• Added unit coverage for orchestration order, output writing, CLI error handling,
  and refusal to use a non-empty output directory.

Try it

python3 -m venv .venv
.venv/bin/pip install binquery
.venv/bin/binquery demo --out /tmp/binquery-demo

The first run may download the OpenCLIP ViT-B-32 weights. Generated files remain
local. The synthetic test pattern verifies that the pipeline runs end to end; it is
not a semantic-quality benchmark for real editing footage.

Verification

• 13 unit tests pass on the release candidate.
• GitHub Actions passes installation, unit tests, and the real-model no-footage demo.
• The demo produces three time-grid clips, nine indexed stills, a readable index,
  and a non-empty ranked query.json on the GitHub runner.

v0.2.0

binquery v0.2.0

v0.2.0 focuses on packaging, automated verification, and clearer documentation. It
does not change the CLI commands, ranking behavior, gates, JSON output, or model.

Highlights

• Moved the Python modules into the dedicated binquery package namespace.
  The installed binquery command, python -m binquery, and the repository wrapper
  now use the same package entry point.
• Added 10 deterministic query unit tests covering routing, ranking, output shape,
  limit clamping, duration bonus behavior, and missing-index errors.
• Extended GitHub Actions from the existing local indexing check to run the unit
  tests and a real-model lavfi smoke path through
  split -> index -> doctor -> query.
• Migrated package licensing metadata to the PEP 639 SPDX form (MIT) and declared
  the LICENSE file in built distributions.
• Improved the README's scope, installation guidance, index-path semantics, output
  field descriptions, and rendering portability outside GitHub.
• Added a local 10-clip query capture to show the current shortlist output without
  presenting it as a large-project or quality benchmark.

Upgrade note

The old top-level Python modules such as cli and query are no longer installed.
They were not a documented library API, but code importing them directly must move
to the binquery namespace. For a source checkout, reinstall the editable package
so the console entry point is regenerated:

python3 -m venv .venv
.venv/bin/pip install -e .

The command-line interface remains the supported interface.

Verification

• 10 query unit tests pass on the release commit.
• GitHub Actions passes installation, unit tests, and the real-model synthetic-media
  smoke workflow.
• The wheel and source distribution build successfully with metadata version 2.4.
• Both distributions pass twine check.
• A clean Python 3.12 environment can install the wheel with all declared
  dependencies satisfied and run both help entry points: the binquery command and
  python -m binquery.

The lavfi workflow proves that the pipeline runs end to end; it is not a semantic
quality benchmark for real editing footage. See the README for the full scope and
known limitations.

v0.1.0

Local CLI: index your own footage folder, then one intent sentence → 8–15 clip shortlist (path, score, gate, reasons).

• Query encodes only the sentence
• Not Premiere, not an auto-editor, not a 500-clip demo
• CI uses 3 lavfi clips (index → doctor)
• Skill install is instructions only, not a PyPI package

`npx -y skills add jennifferweslowski-design/binquery`