Phases
Architecture and foundation
Constitution with hashing, per-agent policy packs, audit engine, publication gate, workflow state machine, database schema, API, dashboard. 34 tests.
Channel integration
OAuth, analytics ingestion, historical snapshots, live analytics dashboard.
Research and editorial engine
Director loop, topic scoring, biblical research wired to a corpus, hermeneutics attestations, script generation.
Video production
Storyboards, image and video generation, narration, FFmpeg assembly, technical QC.
Packaging and distribution
Thumbnails, SEO, Shorts, uploads, scheduling, duplicate prevention, synthetic-content disclosure.
Optimisation
Weekly reports, pillar performance, mission-fit metrics, cost accounting, feedback into scoring.
Autonomous operation
Scheduled workflows, automatic publishing after all gates, monitoring, kill switch, rehearsed rollback.
Resolved
Decisions that were blocking, and what closed them.
Scripture corpus Resolved
2026-11-01 · BSB installed
The Berean Standard Bible is dedicated to the public domain. It is fetched from the publisher, hash-pinned, and refused if the edition changes. 66 books, 1,189 chapters, 31,086 verses. Verified end to end: an exact BSB quotation passes, the same verse in KJV wording submitted as BSB fails, and a reference to a verse this edition omits says so.
Definition of the oneness of God Resolved
2026-11-01 · constitution 1.0.1
Rather than write a Christology the channel owner has not stated, the constitution now defines what is settled — God is one in being and purpose, revealed definitively in Christ — and leaves the internal relations undefined. Any assertion of a Trinitarian, modalist or unipersonal formulation is flagged, and every Godhead artifact routes to human review.
Open decisions
These are the real blockers. None of them is a coding problem.
Bible translation licence
Publication is impossible without a verifiable corpus. Until one is installed, every script fails SCRIPTURE_UNVERIFIED — by design.
Provider stack and budget
Text, voice, image and video providers are unchosen, so no credible cost-per-video exists yet.
Narrator identity
One consistent voice across a hundred videos is a production decision, not a technical one.
Testimony sourcing policy
Fabricated testimony is a hard blocker. A documented sourcing process is required before conversion-story videos can be produced.
YouTube API quota
A default quota cannot carry 24 uploads a month. Either request an extension or change the cadence — before Phase 5, not after.
Dashboard authentication
Acceptable while local; unacceptable once OAuth credentials exist.
Known risks
- Volume against originality. Twenty-four uploads a month with autonomous publishing is exactly where platform policy on inauthentic content becomes a live risk. The gate protects theology, not originality; an originality check across the channel's own catalogue is required.
- Detectors catch phrasing, not meaning. They are review triggers. Exegesis stays with the hermeneutics agent and recorded attestations — which is why a missing attestation blocks.
- Single-vendor dependency. Provider adapters are designed but not implemented. Commit to none until they exist.
- Cost is unknown. Orchestration is estimable in cents. Video generation dominates the real number, and no figure should be promised until one end-to-end render has been measured.