Artefact 08 · Step 2 / 7 · Provenance Chain
UUID chain that maps every artefact in this research package back to its source intel + forward to every downstream creative deliverable. Fibonacci calendar shows when each step compound-checks the previous. Agent-fleet roles assigned per node.
Every research artefact has a UUID + parent. Every downstream creative deliverable (ad, post, email) must reference an upstream artefact UUID in its asset-frontmatter. Breaks the "I can't tell where this hook came from" question forever.
Niche: korean fried chicken malaysia. Geo: MY. Operator: Tricia.
13 brands + 9 pain buckets + 7 whitespace gaps. → 01-market-map.html
4 named personas + channel mix matrix. → 03-icp-catalog.html
24 scored variants as alpha-particle set. 8 structural archetypes. → 10-semantic-particles.html
8 monthly patterns May-Dec 2026. Each = skeleton + ICP cheat-sheet. → 11-pattern-decoder.html
Fast synth verdict (no real OASIS — MiroFish backend down). → 15-mirofish-synth.html
Steal-map across 13 brands. → 23-competitive-landscape-map.html
The pitfall in 12-month brand campaigns: you ship Month 1, then never re-check the assumptions. Fibonacci scheduling forces compound-check cycles at expanding intervals (1, 1, 2, 3, 5, 8) so the operator's attention compounds with the campaign's effects.
Brief delivered to Tricia / Jenn.
ICP review + creative-director G0 gate.
10 hero alphas live in Meta + TikTok test cells.
First A/B/C/D variant readout · prune low-perf.
MiroFish OASIS sim (backend restored) · cross-check vs paid signal.
Month-1 retro · alpha refresh · template iteration.
Month-2 plan reset · whitespace #2 activation.
Mid-campaign deep-revisit · refresh particle set with new alphas.
Six roles. Each node in the chain has a named human owner AND an agent owner. The human owner approves go/no-go; the agent owner ships the artefact.
| Node | Artefact | Human owner | Agent owner | Gate |
|---|---|---|---|---|
| N0 | Brief | Jenn | — | BMAD G0 (problem framing) |
| N1 | Market Map · 01 | Tricia | researcher | G0 · brand-count ≥20, pain-buckets ≥5 |
| N2 | ICP Catalog · 03 | Tricia | creative-director | G1 · 3+ personas with ≥6 fields each |
| N3 | Particles · 10 | Yivonne | autoresearch | G3 · ≥50 alphas + 8 structural archetypes |
| N4 | Templates · 11 | Yivonne | creative-director | G3 · 8 templates with valid skeletons |
| N5 | Synth Verdict · 15 | Jenn | qa-doctor | G6 · top-angle resonance ≥7.5/10 |
| N6 | Landscape · 23 | Tricia | creative-director | G6 · ≥30 brands profiled + steal-map complete |
| N+ | Live campaign creative (every ad) | Tricia + Yivonne | creative-director · super-auditor | G7 · 9.5/10 per god-audit; ad_id minted before fire |
Three explicit hook points where the brand's compound-learning loop should fire. Each = a slash-command + an outputs path. Each hook fires automatically after the named event.
| Hook | Trigger | Slash command | Output |
|---|---|---|---|
| H1 | Every ad creative shipped | /m1.brief → mints ad_id with CAPI + UTMs | Dashboard trace at /api/trace/<ad_id> |
| H2 | Every audit failure | /m1.audit-gate | 9.5/10 target; failed creatives auto-route to /m1.lens-gate for persona-lens diagnosis |
| H3 | Every weekly retro | zen-weekly | ROAS deltas + winning-pattern surface + uncaptured-wins log written to workspace/kyochon/compound-learnings/YYYY-MM-DD-weekly.md |
Provenance pattern: Inherited from zennith-skills standard. UUIDs minted via uuid4(); parent-pointer = the artefact that the current one consumes (so the chain forms a DAG, not a tree).
Fibonacci calendar: Adapted from the Karpathy autoresearch compound-loop pattern. The intervals (1, 1, 2, 3, 5, 8, 13, 21) mirror the cognitive-load curve: tight gates early when assumptions are fragile, expanding gates once signal stabilises.
Gate names (G0-G7): BMAD operating system gates as defined in the zennith-os m1.* slash-command chain.