project creatortrack social meta ingest pipeline 20260715T172132
name: CreatorTrack social-meta-ingest project (voice profile pipeline) description: Active multi-phase CreatorTrack project on feat/social-meta-ingest (warm-slot-2): cross-platform video ingest -> cross-post matching -> voice profile -> content generation type: project
CreatorTrack is building a two-lane pipeline to synthesize Gus's cross-platform voice profile from his full video catalog, working in /home/justinwieb/forge-suite-wt/warm-slot-2 on branch feat/social-meta-ingest (DB ct_warm_slot_2, workspace 85).
Main lane (branch feat/social-data-layer-and-scripts): a numbered task list (Task 1 through ~18) covering transcript ingestion, per-video structure extraction (hook/beats/CTA/tone via Haiku), and culminating in Task 14, voice-profile synthesis, followed by generation and a holdout eval. As of 2026-07-15, extraction (Task 13) hit 515/515 real-transcript videos with zero failures; Task 14 is next.
Meta lane: phased M1-M5 build. M2 (Instagram/Facebook per-video ingest) and M5 (TikTok ingest, 887 videos backfilled) are both built and review-clean. M3 (deterministic cross-post matcher: duration + normalized title + date proximity, precision-first with an ambiguous-review file) landed with 190 auto-linked, 566 queued for human review, 1342 correctly unmatched. Only M4 (UI: crosspost chips + platform filters) remains before the Meta lane closes.
Catalog totals as of 2026-07-15: YouTube 548, Instagram 505, Facebook 699, TikTok 887 = 2639 videos.
Ship plan: phase 1 (main lane) ships to prod first, phase 2 (Meta lane) rebases onto it afterward, then whole-branch reviews before either merges.
Why: this is a multi-week, multi-agent build (worktree lanes running in parallel) with a lot of state; a future session picking this up cold needs the lane/task map rather than re-deriving it from commit history.
How to apply: before touching feat/social-meta-ingest or feat/social-data-layer-and-scripts, check current task/phase completion here first, then verify against actual branch state (this snapshot decays fast).
[auto-memory session 60d132de-4659-485c-8044-427af0ccce7f, confidence 0.72, mode staged]