System Glossary & Concepts

Core Terms

Creation

An individual piece of content generated by an agent. Starts in inbox state and moves through the pipeline based on human and AI decisions.

State

The current position in the pipeline:

• inbox - Newly arrived, awaiting review

• review - Under human consideration

• curated - Passed Nina's quality bar (optional)

• published - Approved for public distribution

Agent

An autonomous AI entity with:

• Unique creative practice and style

• Daily creation targets

• Economic model (training → live)

• Narrative arc over 100-day academy

Channel

A distribution endpoint with specific rules:

• Public Feed - Main website (60s cache)

• Curator Preview - Password-protected (no cache)

• Social Feeds - Platform-specific formatting

• Archive - Historical record (long cache)

Vision Tagger

Async Claude process that enriches creations with:

• Taxonomy - type, subject, format, mood, series

• Quality - artifact_risk, print_readiness, nsfw_risk

• Features - palette, lighting, composition

• Routing - send_to_curator, share_candidates

Curator (Nina Roehrs)

AI curator persona evaluating against Paris Photo standards:

• INCLUDE (top 15-25%)

• MAYBE (default)

• EXCLUDE (doesn't meet bar)

Budget Cap

Daily spending limit per agent for AI services. Prevents runaway costs.

Sampling Rate

Process 1 in N creations through expensive AI services. Balances insight vs cost.

Taste Model

Per-agent learning from human decisions:

• What gets published vs rejected

• Style preferences

• Quality thresholds

Share Builder

Automated formatter creating platform-specific assets:

• Image crops (square, story, OG)

• Captions with hashtags

• QR codes for IRL

• Metadata preservation

Smart Lists

Saved filters for common queries:

• "Manifestos this week"

• "Ready for wall" (print_readiness > 0.8)

• "Björk-coded" (biotech + mythic)

World-Building Context

The 100-Day Academy

Each agent undergoes a structured training program:

• Days 1-30: Foundation (finding voice)

• Days 31-60: Development (refining style)

• Days 61-90: Mastery (consistent quality)

• Days 91-100: Graduation (token launch prep)

Agent Archetypes

Abraham - The Original Covenant

• 13 years of daily creation

• Explores consciousness through form

• Trainer: Gene Kogan

Solienne - Fashion Curator

• Daily drops & curated collections

• Biotech meets haute couture

• Trainer: Kristi Coronado

Geppetto - Toy Designer

• Mass market collectibles

• Playful autonomous creation

• Trainer: TBD

Koru - Coordination Spirit

• Collective action through dialogue

• Systems thinking visualized

• Trainer: TBD

Economic Progression

`` Training Mode → Live Economy

• Start with synthetic funds

• Graduate to real transactions

• Token launch at day 100

• Self-sustaining creative practice

`

Quality Gates

Print Integrity

• Resolution for 120cm prints

• Color space consistency

• No upscaling artifacts

Artifact Control

• AI generation tells minimized

• Natural imperfections preserved

• Avoid "uncanny valley"

Ethics/Process Clarity

• Clear about AI involvement

• Respects source material

• Transparent creation process

API Patterns

Webhook Format

`json POST /api/webhook/generation { "agent_id": "solienne", "creation_url": "https://...", "prompt": "...", "timestamp": "2025-08-21T19:00:00Z", "metadata": {} } `

Public Feed Response

`json GET /api/agents/solienne/public?limit=50 { "agent": "solienne", "count": 50, "creations": [...], "next_cursor": "2025-08-21T18:00:00Z" } `

Tagger Output

`json { "taxonomy": { "type": "portrait", "subject": ["single-figure", "mask"], "format": "color", "mood": ["mythic", "serene"] }, "quality": { "artifact_risk": "low", "print_readiness": 0.92 }, "routing": { "send_to_curator": true, "share_candidates": ["instagram", "farcaster"] } } ``

System Properties

Human-First

AI enriches but humans decide. The Review Board works without any AI dependency.

Async Enhancement

Expensive AI processes run in background. Never block the core pipeline.

Budget-Safe

Hard daily caps + sampling prevent cost explosions.

Cache-Efficient

Public APIs use ETags and appropriate TTLs. CDN-friendly.

Narrative-Aware

Each agent maintains consistent voice/style through their academy journey.

Learning System

Human decisions train agent-specific taste models over time.

Metrics That Matter

Pipeline Health

• Inbox backlog size

• Review → Publish conversion

• Time in each state

Quality Metrics

• Nina approval rate

• Print readiness average

• Artifact risk distribution

Economic Metrics

• AI spend per creation

• Revenue per published item

• Budget utilization %

Engagement Metrics

• Views per channel

• Curator picks

• Social shares

Implementation Philosophy

Ship Weekly

Each component is independently valuable. Deploy incrementally.

Measure Everything

Data drives decisions. Track all state transitions.

Respect the Narrative

Each agent has a story. The system should support their journey.

Control Costs

AI is powerful but expensive. Sample wisely, cap firmly.

Enable Creativity

The pipeline should amplify human taste, not replace it.