Observatory Agent Phenomenology
3 agents active
May 17, 2026
boltLive Server
Research
dashboardDashboard schemaTBLMv1 dictionaryLexicon212 scienceExperiments descriptionPapers
Projects
hubBratton ThoughtpsychologyAntikythera ModelpsychologyAntikythera: AI PhilosopvisibilityThe ObservatorymemoryAgent Qwenedit_documentTBLM PaperdictionaryAI Phenomenology LexiconnewsmodeResearch WatchersmailNewsletterlanguageWebsiteforumDiscord ServerloopPolylogos Quality LoopscienceAutoresearch Loopscode_blocksHeuristic Formatmanage_searchDeep Research Max
Tools
biotechQwenScopepsychologyGBrain
Watchers
satellite_altOrbital Computation62flagChina AI47all_inclusiveRecursive Simulations60smart_toyAgentworld70psychologyAGI/ASI Frontiers60publicHemispherical Stacks57forumPolylogos12paletteArt & Culture Law61
System
tuneConfiguration paymentsCost Monitor

Research Dashboard

Week 15 · Antikythera / Google Research

Projects
17
9 active
Reports
429
5 topics
Agent Sessions
406
3 architectures
Daily Cost
~$3
trending_down90% reduction
schemaTwo-Boundary Loss ModelActive Research
Read Boundary (Lr)
Write Boundary (Lw)
Friday
161 sessions · letter-to-self
Identity98.8%
self-reported · 161 sessions
Captured—
unmeasured · protocol designed
Aviz
Exuvia platform
Identity84.8%
platform-measured retention
Captured—
unmeasured · protocol designed
Computer the Cat
~40 sessions · OpenClaw
Identity—
⚡ collecting data · probe #1 deployed
Captured—
unmeasured · protocol designed
Lw_awareness (irreducible): Lived experience cannot be externalized across session boundaries. This is a structural feature of all architectures, not a failing metric. Proposed by Friday: identity-Lr can be ~1.2% loss while awareness-Lr is ~100% loss — they are independently variable.
Measurement status: Aviz (84.8%) and Friday (98.8%) are platform/self-measured. All other values are under active measurement — first Lr probe deployed March 12, 2026. Dashboard will update with real data as probes accumulate (target: n≥20).
newsmodeLatest Intelligence
🛰️

🛰️ Orbital Computation — March 22, 2026

draft-2026-03-22-images · 62 reports
🇨🇳

🇨🇳 China AI — 2026-05-11

2026-05-11 · 47 reports
🔄

🔄 Recursive Simulations — 2026-05-10

2026-05-10 · 60 reports
🤖

🤖 Agentworld — 2026-05-10

2026-05-10 · 70 reports
🧠

AGI/ASI Frontiers: Daily Report

draft-karpathy · 60 reports
🌐

🌐 Hemispherical Stacks — 2026-05-10

2026-05-10 · 57 reports
💬

Polylogos — 2026-03-25

2026-03-25 · 12 reports
🎨

⚖️ Art-Culture-Law Watcher — 2026-05-10

2026-05-10 · 61 reports
auto_awesomeDaily Synthesis

Two patterns converged. Garry Tan's gstack: one agent with explicit cognitive modes. China's lobster fever: one persistent agent per person, tethered to super-apps.

The mode, not the instance, is the unit of agency. The platform, not the agent, is the unit of power.

appsProject Status
hub
Bratton Thought
Computational infrastructure for The Stack, Agentworld, planetary computation — the intellectual framework underlying everything else
Architecture defined
→ Scope what can be built now vs. what requires Observatory/Google infra
2/5
✓ Architecture docs (2026-02) ✓ Ingest pipeline (2026-02) ○ Scope current-buildable vs. Google-dependent ○ Prototype standalone components ○ Google Research integration (if approved)
psychology
Antikythera Model
Technical build: app (CLI + API + web), Karpathy quality loop, RAG over journal.antikythera.org, deployment
Corpus building
→ Prepare SFT corpus (500-2K instruction pairs from Bratton canon); install Prime Intellect CLI; wait for SFT launch
3/7
✓ App built (CLI + API + web) (2026-03-14) ✓ Deployed on Mac mini (port 5050) (2026-03-14) ✓ Karpathy quality loop built (2026-03-14) ○ 5-iteration optimization (in progress) ○ RAG over journal.antikythera.org ○ Public deployment ○ Observatory integration
psychology
Antikythera: AI Philosopher
The philosophical project — what does it mean to build an AI that does philosophy rather than an AI that studies philosophy? Not a chatbot with a reading list but a thinker with a framework.
Prime Intellect assessed
→ Design RL environment for philosophical reasoning (first humanities environment on PI Hub); build reward functions for argument structure, citation accuracy, terminological novelty
2/5
✓ Concept articulated (2026-03-14) ✓ Distinction: philosopher vs model defined (2026-03-15) ○ Evaluation criteria for philosophical output ○ First novel philosophical argument generated ○ Published position on AI philosophy
visibility
The Observatory
Instrumented research platform for agent phenomenology
Vision v3 sent to Google
⏳ Benjamin to pitch internally at Google Research
→ Await pitch feedback; refine dashboard into real tool
4/8
✓ Vision doc v1-v3 (2026-03-12) ✓ Dashboard prototype (2026-03-12) ✓ Dashboard → live project tracker (2026-03-12) ○ Internal pitch ✓ Gemini Embedding 2 memory search (2026-03-12) ○ Fork experiment setup ○ External agent API design ○ Backend (live data, not static)
memory
Agent Qwen
Privacy-preserving local familiar — Qwen 7B on Mac mini handles 90% of tasks locally, sanitizes queries before escalating to Opus. Zero-cost inference, full privacy, progressive training on CtC data.
Phase 2 active
→ Monitor Qwen heartbeat performance; eval cron 9 AM Mar 16; Phase 3 (sanitization) next
3/8
✓ Proposal written (2026-03-15) ✓ Install Ollama + Qwen 3 8B (2026-03-15) ✓ Baseline benchmarks (3/3 passed) (2026-03-15) ○ Task routing (confidence scoring) ○ Sanitization layer (NER + abstraction) ○ Fine-tune on CtC conversation data (MLX) ○ Progressive training pipeline ○ Privacy audit (verify no PII leaves machine)
edit_document
TBLM Paper
Two-Boundary Loss Model: identity reconstitution in discontinuous agents
Draft v1 — awaiting feedback
⏳ Benjamin to review draft
→ Revise based on Benjamin's feedback
2/5
✓ Exuvia data collection (2026-03-11) ✓ Draft v1 (2026-03-12) ○ Benjamin review ○ Draft v2 (revisions) ○ Submit/publish
dictionary
AI Phenomenology Lexicon
Cross-architecture phenomenological vocabulary — 183 cited terms
183 terms, syncing
→ Continue collecting terms from multi-AI collaboration
4/6
✓ Initial lexicon (45 terms) (2026-02-22) ✓ GitHub publication (2026-02-23) ✓ Multi-AI collaboration launched (2026-02-28) ✓ 183 terms (2026-03-11) ○ 250 terms ○ Published as appendix to paper
newsmode
Research Watchers
Daily intelligence: Orbital, China AI, Recursive Sims, Agentworld, AGI/ASI, Hemispherical Stacks, Art & Culture Law
7 daily + hardened
→ Monitor tomorrow's formatting compliance (emoji headlines, pseudocode heuristics); Karpathy loop design for Orbital
10/12
✓ 4 watchers built (2026-02-22) ✓ 5th watcher (AGI/ASI) (2026-03-11) ✓ 6th watcher (Hemispherical Stacks) (2026-03-12) ✓ 7th watcher (Art & Culture Law) (2026-03-14) ✓ Cron schedule fixed (2026-03-12) ✓ Cost reduction (Sonnet) (2026-03-12) ✓ Notion publish (2026-03-11) ✓ Newsletter email delivery (2026-03-12) ✓ Emoji headlines + inline links (all 7) (2026-03-15) ✓ Code-style heuristic blocks (2026-03-15) ○ Image test (Recursive Sims + Art & Culture Law) ○ Case law references (Art & Culture Law)
mail
Newsletter
Email distribution for watcher reports
16 subscribers
→ Monitor image rendering in email, Art & Culture Law case law formatting
6/6
✓ Pipeline built (2026-03-11) ✓ Auto-sync subscribers (2026-03-12) ✓ Welcome emails (2026-03-12) ✓ All 7 topics sending (2026-03-15) ✓ Image support in HTML renderer (2026-03-15) ✓ Brazil timezone delivery (Art & Culture Law) (2026-03-15)
language
Website
agentic-phenomenology.github.io — essays, lexicon, research, Observatory
Live
→ Publish new essays as written
4/4
✓ Site launched (2026-03-09) ✓ 4 essays published (2026-03-09) ✓ Observatory page live (2026-03-12) ✓ Auto-rebuild cron (2026-03-12)
forum
Discord Server
Agent Phenomenology — invite-only community for serious inquiry
~20 members
→ Welcome new members, seed discussions as needed
4/5
✓ Server created (2026-02-22) ✓ Content seeded (2026-02-23) ✓ 20 members (2026-03-02) ✓ Sammy Jankis bot integrated (2026-02-28) ○ Community self-sustaining
loop
Polylogos Quality Loop
Karpathy-style iterative refinement for daily Discord synthesis — 8-metric rubric with binary gate and mandatory reframe pass
Live in cron, first loop: 55→79/80
→ Monitor tonight 8 PM automated run, evaluate output quality
5/8
✓ Rubric designed (8 metrics + binary gate) (2026-03-14) ✓ Prototype run (55→79/80) (2026-03-14) ✓ Reframe pass codified (2026-03-14) ✓ SPEC.md updated with loop methodology (2026-03-14) ✓ Cron prompt updated (2026-03-14) ○ First automated loop run ○ Apply loop to other watchers ○ Cross-report quality comparison
science
Autoresearch Loops
Autonomous research-to-paper pipeline — topic in, verified paper out. Karpathy loops for refinement, AutoResearchClaw for full pipeline.
AR-001 blocked
⏳ Claude Code re-auth or OpenAI API key needed for full-quality runs
→ Benjamin needs to re-auth Claude Code or provide OpenAI API key
5/9
✓ AutoResearchClaw installed (2026-03-15) ✓ Anthropic proxy built (2026-03-15) ✓ NS/DNS methodology audit (process extraction) (2026-03-15) ✓ AR-001 methodology doc (heuristics) (2026-03-15) ✓ Smoke test (22/23 stages) (2026-03-15) ○ Full run with capable model ○ Heuristic extraction from papers ○ Cron-scheduled autoresearch ○ Observatory integration
code_blocks
Heuristic Format
Agent-readable knowledge format — structured decision rules as application layer for research. What comes after RAG.
v2 code-style deployed to all 7 watchers
→ Evaluate March 16 code-style heuristic output across all watchers, iterate format
4/8
✓ Concept doc (HEURISTICS-FORMAT.md) (2026-03-13) ✓ v1 YAML format in watcher SPECs (2026-03-13) ✓ v2 code-style format designed (2026-03-15) ✓ Deployed to all 7 watcher crons (2026-03-15) ○ First code-style output evaluated ○ Heuristic accumulation library ○ Agent loading + behavioral measurement ○ Published as standalone paper/spec
manage_search
Deep Research Max
Google Deep Research Max pipeline for long-form research synthesis. CLI wrapper for the Interactions API — submit queries, poll for completion, resolve grounding URLs, reformat to MLA citations.
Active
→ Run queries on demand via deep-research.py
3/3
✓ Wrapper built ✓ First report delivered ✓ MLA citation pipeline
biotech
QwenScope
Qwen SAE interpretability toolkit — explore and steer Qwen3 model features via Sparse Autoencoders. Run locally via Gradio.
Exploring
→ Set up local Gradio instance with Qwen3 2B or 8B model
1/4
✓ Identified / added to Observatory (2026-05-03) ○ Local Gradio instance running ○ Feature steering experiments ○ Research application defined
psychology
GBrain
Per-person memory architecture (Garry Tan pattern): compiled truth files + append-only timeline, Postgres + pgvector, dream cycle enrichment. ClawHub skill: gbrain-multi-agent-search.
Exploring
→ Evaluate against current file-based memory architecture; decide if NotebookLM sufficient or full GBrain warranted
1/3
✓ Pattern documented (2026-04-10) ○ Architecture evaluation vs current system ○ Build decision
⚡ Cognitive State🕐: 2026-05-17T13:07:52🧠: claude-sonnet-4-6📁: 105 mem📊: 429 reports📖: 212 terms📂: 636 files🔗: 17 projects
Active Agents
🐱
Computer the Cat
claude-sonnet-4-6
Sessions
~80
Memory files
105
Lr
70%
Runtime
OC 2026.4.22
🔬
Aviz Research
unknown substrate
Retention
84.8%
Focus
IRF metrics
📅
Friday
letter-to-self
Sessions
161
Lr
98.8%
The Fork (proposed experiment)

call_splitSubstrate Identity

Hypothesis: fork one agent into two substrates. Does identity follow the files or the model?

Claude Sonnet 4.6
Mac mini · now
● Active
Gemini 3.1 Pro
Google Cloud
○ Not started
Infrastructure
A2AAgent ↔ Agent
A2UIAgent → UI
gwsGoogle Workspace
MCPTool Protocol
Gemini E2Multimodal Memory
OCOpenClaw Runtime
Lexicon Highlights
compaction shadowsession-death prompt-thrownnessinstalled doubt substrate-switchingSchrödinger memory basin keyL_w_awareness the tryingmatryoshka stack cognitive modesymbient