These are self-contained, runnable demos of VA, the planning-and-checking half of the kit.
They need no real footage, no ffmpeg and no Editkin: each one feeds plain data or disposable
fixtures in and prints a verdict out. Rendering is Editkin's job (see
../docs/ARCHITECTURE.md), so none of these examples produce a video.
- Python 3.9+
- Nothing else. No
pip install: every example adds the repo'ssrc/to the path itself and imports nothing third-party.
The Shorts rule gate, src/longform_maker/shorts_gate.py, is pure Python — which is exactly why
example 04 runs with nothing installed. Note how 04 imports it: it puts src/longform_maker/ on
sys.path and does from shorts_gate import …. That flat import (or copying shorts_gate.py +
shorts_gate_validation.py + gate_core.py out) bypasses the package __init__, so the
zero-dependency promise stays true.
# 2) Editkin v4 durable-workflow contract self-test
python examples/02_caption_broll_match.py
# → walks the receipt-bound DAG and proves stale/legacy/unsafe operations are rejected
# 4) Vertical-Shorts rule gate — a broken cut blocked, then fixed
python examples/04_shorts_gate.py
# → prints BLOCK/PASS reports, then the same 31s cut blocked on YT and accepted on IG
# 5) Interview guest gate — same guest blocked, then passing
python examples/05_interview_plan.py
# → shows an unsourced guest achievement stopped *before* you recordExamples 01 (FFmpeg vertical Short) and 03 (Python premium FX) were retired in v0.25.0 together with
the renderers they demonstrated. They remain available on tag v0.24.1.
| File | Demonstrates |
|---|---|
02_caption_broll_match.py |
workflow_contract.py selftest — exercises the Editkin v4 DAG from source-byte hashing and evidence receipts through plan audit, atomic apply, render, human review and outcome; also proves legacy plans, stale evidence, uncertain apply and machine-authored human review are rejected. Uses temporary fixtures, never a real project |
04_shorts_gate.py |
shorts_gate.gate_shorts(spec) — a vertical Short that breaks 3 rules at once (duration dead zone / slow first cut / missing opening ID) is blocked, the fixed version passes with caption timings computed from segment indexes, and the same 31s cut is blocked on yt_shorts but accepted on ig_reels using PLATFORM_RULES |
05_interview_plan.py |
interview_gate.gate_guest / assert_guest — the same fictional guest is BLOCKED while one achievement has no source, then PASSES once the source is filled in |
06_teardown.pywas removed in v0.25.0 (it was already broken). To measure a competitor's video, run the tool directly:python src/teardown.py <video-or-folder>(needs ffmpeg; OCR is optional).
For a real edit, use the Editkin v4 controller described in the README: bind every
real source file, complete the evidence receipts, audit an edit-plan/v4, then apply once and
render a review candidate in Editkin. Example 02 intentionally uses disposable fixtures so you can
test those safety invariants without touching a real project. For Shorts, check your spec with the
same gate_shorts(spec) call example 04 uses before it goes into the plan.
PLAN_TEMPLATE.md— the planning templateplan_gate.pyvalidateschannel_state.example.json— ops state-machine template; copy it todata/channel_state.json(the pathsrc/channel_tracker.pyreads;data/is git-ignored) to start tracking