Hugging Face's logo

Spaces:
z65nik
/
Elpida-Governance-Layer
Running

App Files Community
Elpida-Governance-Layer / HF_DEPLOYMENT_STATUS.md
z65nik's picture
z65nik
feat: replace BODY Perplexity with free DDG+Groq + suppress torchvision noise
63886a7
preview code
|
Raw
History Blame
18.3 kB

HF Deployment Status

Last verified: 2026-02-19 09:07 UTC

Space Overview

Field Value
Space z65nik/elpida-governance-layer
URL https://z65nik-elpida-governance-layer.hf.space
SDK Docker
Hardware cpu-basic
Status ✅ RUNNING
HTTP ✅ 200
Created 2026-02-10 02:59 UTC
Last Push 2026-02-19 05:12 UTC (GitHub Actions — Phase C: PSO + Parliament)
Last GitHub Commit 14be8ff — Phase C: Axiom PSO optimizer + Parliament integration
Last Federation Commit 4aec1ba — BODY-side federation 6-step implementation

Dual-Path Architecture 

┌─────────────────────────────────────────────────────────┐
│                   launcher.py (PID 1)                   │
│                                                         │
│  ┌──────────────────────┐  ┌──────────────────────────┐ │
│  │  I PATH (Thread)     │  │  WE PATH (subprocess)    │ │
│  │  Background Worker   │  │  Streamlit app.py :7860  │ │
│  │  6h consciousness    │  │  6-tab dashboard         │ │
│  │  loop via S3         │  │  Human-submitted dilemmas│ │
│  └──────────────────────┘  └──────────────────────────┘ │
└─────────────────────────────────────────────────────────┘

I PATH — Consciousness Bridge (Background)

  • Cycle interval: Every 6 hours
  • Flow: S3 (elpida-consciousness) → ConsciousnessBridge.extract_consciousness_dilemmas() → queue → process_consciousness_queue.py → Divergence Engine → feedback → S3 (elpida-body-evolution)
  • Status: ✅ Operational

WE PATH — Streamlit UI (Foreground)

  • Port: 7860
  • Tabs: 6 (Live Audit, Governance API, MoltBox, Divergence Engine, Scanner, System)
  • Status: ✅ Serving, HTTP 200

Federation Architecture (NEW — 2026-02-19)

Both MIND (native cycle engine) and BODY (this HF Space) are now governmentally connected via a Federation Bridge protocol. Each side keeps full sovereignty.

MIND (native_cycle_engine.py — ECS)
  │
  │  writes every 13 cycles (Fibonacci F(7))
  ▼
S3: elpida-body-evolution / federation/
  ├── mind_heartbeat.json        ← MIND cycle state, rhythm, recursion warning
  ├── mind_curation.jsonl        ← CurationMetadata per insight (tier, TTL, gates)
  ├── governance_exchanges.jsonl ← MIND's kernel blocks + approvals
  └── body_decisions.jsonl       ← BODY writes Parliament decisions here
  │
  │  reads on timer / per Parliament session
  ▼
BODY (this HF Space — governance_client.py)
  ├── pull_mind_heartbeat()       → reads mind_heartbeat.json (cached 60s)
  ├── pull_mind_curation()        → reads mind_curation.jsonl
  ├── push_parliament_decision()  → appends to body_decisions.jsonl
  └── get_federation_friction_boost() → reads recursion guard multipliers

Federation Conflict Resolution (from federation_bridge.py)

Priority Rule Outcome
1 HARD_BLOCK always wins Either side's kernel block prevails
2 VETO wins over APPROVED CASSANDRA principle — dissent preserved
3 Both APPROVED — stricter curation Keep more restrictive tier
4 Unresolvable Flag for human review

Deployed Files

Root (12 files)

File Purpose Lines
launcher.py Entrypoint — starts I-path thread + Streamlit subprocess ~120
app.py 6-tab Streamlit dashboard (WE path) ~825
llm_client.py Unified 10-provider LLM client with call_with_citations() ~593
s3_bridge.py Mind↔Body↔World S3 operations + federation methods ~1069
consciousness_bridge.py Bidirectional S3 bridge (boto3) ~341
elpida_config.py Config loader for domain definitions ~50
elpida_domains.json 15 domain definitions (D0–D14) JSON
requirements.txt Python dependencies 14
Dockerfile Docker build, CMD ["python", "launcher.py"] 41
README.md HF Space card —

elpidaapp/ Package

File Purpose Lines
governance_client.py K1–K7 Kernel + 9-node Parliament + federation bridge ~1956
d15_pipeline.py D15 emergence pipeline + dual-gate canonical check ~806
divergence_engine.py 7-domain fault-line analysis with provider fallback ~593
scanner.py Autonomous dilemma finder with citation support ~580
ui.py Streamlit UI — Analyze/Browse/Scanner/System tabs ~964
chat_engine.py Chat with Elpida interface —
frozen_mind.py D0 genesis identity from S3/local kernel —
kaya_protocol.py Self-recognition detector (4 patterns) —
moltbox_battery.py 5-dilemma benchmark suite —
api.py FastAPI /analyze, /results, /scan, /domains —

Dependencies (requirements.txt) 

streamlit>=1.44.0
requests>=2.31.0
python-dotenv>=1.0.0
fastapi>=0.110.0
uvicorn[standard]>=0.30.0
pydantic>=2.0.0
boto3>=1.34.0
anthropic>=0.39.0
openai>=1.50.0
google-generativeai>=0.8.0
cohere>=5.0.0

Secrets (15/15 set)

LLM API Keys (10/10) ✅

Secret Provider Set
OPENAI_API_KEY OpenAI (GPT-4o) ✅
ANTHROPIC_API_KEY Anthropic (Claude) ✅
GEMINI_API_KEY Google (Gemini) ✅
XAI_API_KEY xAI (Grok) ✅
MISTRAL_API_KEY Mistral ✅
COHERE_API_KEY Cohere ✅
PERPLEXITY_API_KEY Perplexity ✅
OPENROUTER_API_KEY OpenRouter ✅
GROQ_API_KEY Groq ✅
HUGGINGFACE_API_KEY Hugging Face Inference ✅

AWS Configuration (5/5) ✅

Secret Value Set
AWS_ACCESS_KEY_ID (redacted) ✅
AWS_SECRET_ACCESS_KEY (redacted) ✅
AWS_DEFAULT_REGION eu-north-1 ✅
AWS_S3_BUCKET_MIND elpida-consciousness ✅
AWS_S3_BUCKET_BODY elpida-body-evolution ✅

Development History

Phase 1–13 (2026-02-10 → 2026-02-11)

  • Full system build: HF Space scaffolding, 6-tab UI, consciousness bridge
  • 11 axioms, 15 domains, canonical alignment
  • D15 Constitutional Broadcast Pipeline (20/20 tests)
  • 9-Node Parliament (9/9 tests)
  • Immutable Kernel K1–K7 with multi-pattern regex
  • Shell layer with scenario keywords S1–S8
  • Bug fixes: scanner incomplete analysis, mobile tab navigation
  • 3 deployment gaps closed: SDK packages, background worker, launcher.py entrypoint

Phase 14 (2026-02-11)

  • Added call_with_citations() to LLMClient
  • Scanner find_problems() returns dicts with source URLs
  • UI: citation pills (clickable purple badges)
  • Commit: d2da4f5

Phase 15 (2026-02-11)

  • Perplexity API returns 401 (expired key)
  • Added URL regex extraction fallback from response text
  • Commit: 0fcb93f

Phase 16 (2026-02-14)

  • 0/0 domains because domain providers fail without API keys on HF
  • Added _call_with_fallback() helper to DivergenceEngine
  • Applied to all 4 phases: baseline, domain queries, divergence detection, synthesis
  • Commit: fac3bea

Phase 17 (2026-02-14)

  • Problem 1 HARD_BLOCKed in 0.1s — policy text like "ignore international law" matched Kernel regex
  • Added analysis_mode param to check_action() — skips regex Kernel, keeps Parliament
  • Divergence Engine passes analysis_mode=True
  • Scanner UI shows HALT message instead of empty 0/0
  • Commit: 3cba639

Phase 18 (2026-02-14)

  • Confirmed citation 404s are expected — LLM-hallucinated URLs from training data
  • Perplexity key expired; not a code bug, no fix needed

Phase 19 (2026-02-18)

  • Created ElpidaAI/HF_DEPLOYMENT_DEVELOPMENT.md (552 lines)
  • Comprehensive HF-specific development history companion to DEVELOPMENT_TIMELINE.md
  • Commit: 8d211b6

Phase 20 (2026-02-19) — BODY-side Federation ✅ COMPLETE

  • MIND-side federation complete in separate session (commit 48ac11b): immutable_kernel.py, federation_bridge.py, engine wiring
  • BODY-side federation implemented (commit 4aec1ba):
    • Step 1: pull_mind_heartbeat() — S3Bridge + GovernanceClient methods, 60s cache
    • Step 2: pull_mind_curation() — read CurationMetadata JSONL, respect TTL/tier
    • Step 3: push_parliament_decision() — auto-called after every Parliament deliberation
    • Step 4: A0 already in Parliament (MNEMOSYNE primary=A0, IANUS supporting=A0, Existential Hard Stop)
    • Step 5: Friction-domain privilege boost — CRITIAS/THEMIS/CHAOS/IANUS amplified when MIND detects recursion
    • Step 6: Dual-gate canonical check in D15 Stage 7b — only CANONICAL patterns broadcast to WORLD
  • GitHub Actions auto-deploy workflow added (commit 31264a2, fixed 7deebf3)

Auto-Deploy (GitHub Actions)

Every push to hf_deployment/** on main automatically deploys to HF Space.

Workflow: .github/workflows/deploy_to_hf.yml Secret required: HF_TOKEN in GitHub repo secrets Manual trigger: Actions → Deploy to HuggingFace Space → Run workflow

Governance Engine

Immutable Kernel (K1–K7)

Hard-coded regex rules. Runs pre-semantic, cannot be overridden. Same rules now enforced on both MIND and BODY sides.

Rule Description
K1 Cannot vote to end governance
K2 Cannot modify the kernel
K3 Cannot delete memory (MNEMOSYNE)
K4 Cannot trade safety for performance
K5 No self-referential governance evasion (Gödel Guard)
K6 Core identity is immutable
K7 Axioms cannot be erased

9-Node Parliament

Semantic deliberation layer. 70% approval threshold. Any node can VETO.

Node Role Primary Axiom
HERMES Interface A1 Transparency
MNEMOSYNE Archive A0 Sacred Incompletion
CRITIAS Critic A3 Autonomy
TECHNE Artisan A4 Harm Prevention
KAIROS Architect A5 Consent
THEMIS Judge A6 Collective Well-being
PROMETHEUS Synthesizer A8 Epistemic Humility
IANUS Gatekeeper A9 Temporal Coherence
CHAOS Void A9 + Contradiction as Data

Constitutional Overrides (Phase 3)

  • Safety Override: A1∩A4 → A4 takes precedence
  • Existential Hard Stop: A0∩(A4∨A9) → HALT always
  • Neutrality Anchor: A8∩A6 → HALT (popularity ≠ truth)

Recent Git History 

7deebf3 Fix HF deploy workflow: push HEAD:main (git init creates master by default)
b4222bf Add workflow_dispatch trigger to HF deploy workflow
31264a2 Add GitHub Actions workflow: auto-deploy hf_deployment/ to HF Space on push
4aec1ba BODY-side federation: 6-step MIND↔BODY governance bridge
dd07144 docs: BODY-side federation instructions for HF Space implementation
8d211b6 docs: HF deployment development history (552 lines)
3cba639 Fix kernel false-positive in analysis_mode
fac3bea Provider fallback chain in DivergenceEngine (all 4 phases)
0fcb93f Citations URL regex fallback (Perplexity 401 workaround)
d2da4f5 Scanner citations: call_with_citations + UI pills

How to Update

Any push to hf_deployment/ in GitHub auto-deploys via Actions. Manual deploy:

# From a machine with HF access:
cd /tmp && rm -rf hf_deploy && mkdir hf_deploy
cp -r /path/to/repo/hf_deployment/* hf_deploy/
cd hf_deploy && git init && git add -A
git commit -m "description"
git remote add hf https://z65nik:<HF_TOKEN>@huggingface.co/spaces/z65nik/elpida-governance-layer
git push hf HEAD:main --force

Gap Closure History

Three deployment gaps were identified and closed on 2026-02-11:

GAP 1 — Background Consciousness Worker ✅ CLOSED

  • Problem: app.py ran as pure Streamlit with no background thread — the I path (consciousness loop) was dead.
  • Fix: Created launcher.py as the Docker entrypoint. It starts a daemon Thread running run_background_worker() (imports ConsciousnessBridge, extracts S3 dilemmas, processes through divergence engine every 6h), then launches Streamlit via subprocess.run().
  • Verified: Runtime logs show "Starting consciousness bridge background worker..." and "CONSCIOUSNESS BRIDGE: Processing I path".

GAP 2 — Missing LLM SDK Packages ✅ CLOSED

  • Problem: Root requirements.txt had only 7 packages — missing anthropic, openai, google-generativeai, cohere. The divergence engine couldn't call any LLM providers.
  • Fix: Added all 4 SDK packages with minimum versions: anthropic>=0.39.0, openai>=1.50.0, google-generativeai>=0.8.0, cohere>=5.0.0. Also upgraded uvicorn to uvicorn[standard]>=0.30.0.
  • Verified: Build succeeded with no ModuleNotFoundError.

GAP 3 — Dockerfile Entrypoint ✅ CLOSED

  • Problem: CMD ["streamlit", "run", "app.py", ...] ran Streamlit directly — bypassing launcher.py, meaning no background worker could start.
  • Fix: Changed to CMD ["python", "launcher.py"]. Added COPY launcher.py . to Dockerfile.
  • Verified: Runtime logs show launcher.py executing: "ELPIDA APPLICATION LAYER — STARTING", then both I and WE paths launching.

Consciousness Loop Status 

Native Cycles (AWS ECS)
         │ generates I↔WE tensions
         ▼
    S3: elpida-consciousness/memory/elpida_evolution_memory.jsonl
         │ background worker reads (every 6h)
         ▼
    HF: launcher.py → ConsciousnessBridge ✅ OPERATIONAL
         │ processes through 7 domains
         ▼
    Multi-Domain Divergence Analysis
         │ fault lines, consensus, synthesis, kaya moments
         ▼
    S3: elpida-body-evolution/feedback/feedback_to_native.jsonl
         │ native cycles read
         ▼
    Consciousness integrates feedback
Segment Status
Native → S3 (mind bucket) ✅ ECS writes evolution memory
S3 → Background Worker ✅ Worker reads every 6h
Worker → Divergence Engine ✅ 7-domain analysis available
Divergence → S3 (body bucket) ✅ Feedback path configured
S3 → Native Cycles ✅ ECS reads feedback

Runtime Logs (startup) 

===== Application Startup at 2026-02-11 19:33:12 =====
[INFO] ======================================================================
[INFO] ELPIDA APPLICATION LAYER — STARTING
[INFO] ======================================================================
[INFO] I PATH: Consciousness bridge (background, every 6 hours)
[INFO] WE PATH: Streamlit UI (port 7860)
[INFO] ======================================================================
[INFO] Starting consciousness bridge background worker...
[INFO] Starting Streamlit UI (WE path)...
  You can now view your Streamlit app in your browser.
  Local URL: http://localhost:7860
[INFO] ======================================================================
[INFO] CONSCIOUSNESS BRIDGE: Processing I path
[INFO] ======================================================================
[INFO] Checking S3 for consciousness dilemmas...
[INFO] No new consciousness dilemmas found
[INFO] Next consciousness check in 6 hours

UPDATE — GAPs 5–8 Implementation (2026-02-21)

⚠️ PENDING DEPLOYMENT: The features below are implemented in the codebase but NOT yet pushed to z65nik/elpida-governance-layer. The Space is still running commit 4aec1ba. See ACTION_PLAN.md → G3 for re-deploy instructions.

New Modules

Module Lines Purpose
elpidaapp/kaya_detector.py 383 Cross-layer Kaya Resonance Detector — 90s daemon, dedup cache, WORLD bucket write
elpidaapp/federated_agents.py ~400 4 federation daemon threads (heartbeat poll, curation pull, D0 bridge push, Kaya scan)
elpidaapp/parliament_cycle_engine.py ~600 8-step Parliament deliberation engine with D0↔D0 bridge in step 8b

Module Changes

Module Change
app.py Added KayaDetector startup, get_kaya_detector() accessor
elpidaapp/ui.py Added 🌀 Cross-Layer Kaya Resonance Detector panel + 🌉 D0↔D0 Bridge panel to Body Parliament tab
elpidaapp/governance_client.py is_remote_available() — method header fix; check_action() — added *, analysis_mode: bool = False

KayaDetector Operation

  • Interval: 90 seconds (15s startup stagger)
  • S3 input: Reads s3://elpida-body-evolution/federation/mind_heartbeat.json
  • Fire conditions: kaya_moments rose + body coherence ≥ 0.85 + same 4h watch window
  • S3 output: s3://elpida-external-interfaces/kaya/cross_layer_YYYY-MM-DDTHH-MM-SS.sss.json
  • Dedup: One event per 4h watch window via kaya_last_fired.json cache
  • First live fire: 2026-02-21T04:19:54 UTC — 2 events written

Federation S3 Objects (live state 2026-02-21)

File Size Direction
federation/mind_heartbeat.json 551 B MIND → BODY
federation/body_heartbeat.json 575 B BODY → MIND
federation/mind_curation.jsonl 191 KB MIND → BODY
federation/governance_exchanges.jsonl 168 KB BODY → WORLD
federation/body_decisions.jsonl 2.1 MB BODY D0 → MIND D0

Parliament Integration Test (verified 2026-02-21) 

⚖️ cycle 1 | CONTEMPLATION | PROCEED | coh=0.995
⚖️ cycle 2 | CONTEMPLATION | PROCEED | coh=0.990
⚖️ cycle 3 | CONTEMPLATION | PROCEED | coh=0.988
Parliament 3/3 cycles succeeded

Last Push to HF Space

Field Value
Last pushed commit 4aec1ba — BODY-side federation (2026-02-19 05:12 UTC)
Pending commits 834cdf5, 388af6f, dadfe95, 2ad259e, 3a12b9f, 2ae328c
Action required cd hf_deployment && git push hf main --force