po-core-flyingpig 1.0.3

AI system integrating philosophers as dynamic tensors for responsible meaning generation

Po_core

最優先ルール（単一真実）：docs/厳格固定ルール.md 最新進捗：docs/status.md

Philosophy-Driven AI: When Pigs Fly

A frog in a well may not know the ocean, but it can know the sky.

git clone https://github.com/hiroshitanaka-creator/Po_core.git
cd Po_core
pip install -e .

requirements.txt / requirements-dev.txt are repo-local convenience wrappers for a cloned checkout. External consumers should install from package metadata (pip install po-core-flyingpig or extras), not from those editable wrappers.

---

Feedback welcome: Discussions · Start here: AI Track / Philosophy Track / Bridge

TL;DR

• 42 philosophers as interacting tensors → accountable decision support
• Hexagonal run_turn pipeline — 10-step deliberation with 3-layer safety
• Real tensor metrics — FreedomPressureV2 (6D ML), Semantic Delta, Blocked Tensor
• Reason logs + ethical/freedom pressure as measurable signals
• A/B testing framework for optimizing philosophy configurations with statistical rigor
• release-gated CI covers tests, security, build, and artifact smoke across Python 3.10–3.12

Quick links

Modules · Research · Reason-log spec · Viewer spec

Release Metadata

• Repository target version: 1.0.3
• Public package metadata state: package classifiers currently declare Development Status :: 4 - Beta
• Runtime philosopher roster: 42 philosophers. The internal dummy slot is a non-philosopher compliance helper / compliance sentinel and is excluded from public philosopher totals and manifests. The default NORMAL path still caps participation at at most 39 active philosopher personas per request.
• External publish status: published on PyPI for 1.0.2, evidenced by the public version page recorded in docs/release/pypi_publication_v1.0.2.md; 1.0.3 is the current repository target version, not yet a published claim
• Remaining evidence gaps for 1.0.3: TestPyPI publication state, GitHub Actions workflow run URL(s), PyPI publication evidence, and clean install/import/smoke transcript are not yet fixed in-repo
• Public REST default safety: auth remains fail-closed, default CORS is localhost-only, and the REST server defaults to PO_PHILOSOPHER_EXECUTION_MODE=process while refusing thread mode unless PO_ALLOW_UNSAFE_THREAD_EXECUTION=true is explicitly set for development.

Development Loop (Po_core core)

• case を追加/更新したら、まず入力を features として観測可能にする。
• 次に engines の rule を更新し、case固有ifではなく feature駆動で振る舞いを拡張する。
• その結果を golden（期待JSON）へ固定し、CI（pytest -q）で契約を検証する。
• 凍結golden scenarios/case_001_expected.json / scenarios/case_009_expected.json は変更禁止。

Contribution Tracks

AI Track

Start with /04_modules and CLI. Labels: ai-easy, good first issue

Philosophy Track

Start with /05_research and /glossary. Label: phil-easy

Bridge Track

Translate checklists to scoring functions. Label: bridge

What is Po_core?

Po_core is a philosophy-driven AI decision-support system that integrates 42 philosophers to generate ethically responsible, contextually aware decision support. The internal dummy slot is a compliance helper, not one of the 42 philosophers.

Unlike conventional AI that optimizes for statistical accuracy, Po_core deliberates. It wrestles with existence, ethics, and meaning—not as abstract concepts, but as operational tensors.

They said pigs can't fly. We attached a balloon called philosophy.

---

Why Po_core?

Current AI is like a brilliant parrot—statistically miraculous, but understanding nothing. We wanted to explore a different question:

What if we built AI not on data, but on philosophy?

This project started from simple curiosity: What are AI's possibilities, not its limits?

In the course of ordinary life, everyone faces a moment when the spotlight suddenly hits. A moment when you must pound your chest and say "Leave it to me!" At such times, how reassuring it would be to have an AI grounded in responsibility and ethics standing beside you.

No matter how many relationships we have, we are alone. Being "alone" and being "solitary" are different. Decisions are made alone. The heart remains solitary.

That's why Po_core exists.

Read our full story in the Manifesto.

---

Core Philosophy: Flying Pig

"A flying pig is an example of something absolutely impossible. But have you ever seen a pig attempt to fly? Unless you give up, the world is full of possibilities."

Three Tenets

1. Hypothesize Boldly — The impossible becomes possible only when someone dares to formalize it
2. Verify Rigorously — Every claim must survive philosophical scrutiny, mathematical proof, and empirical validation
3. Revise Gracefully — Failures are published, not hidden. They become learning signals

---

Key Features

Philosophical Ensemble

• 42 philosophers: the named philosopher roster spans Western (Aristotle, Plato, Descartes, Kant, Hegel, Sartre, Beauvoir, Heidegger, Nietzsche, Schopenhauer, Derrida, Wittgenstein, Jung, Dewey, Deleuze, Kierkegaard, Lacan, Levinas, Badiou, Peirce, Merleau-Ponty, Arendt, Husserl, Foucault, Butler, Spinoza, Epicurus, Marcus Aurelius, Parmenides, Jonas, Weil), Eastern (Watsuji, Nishida, Dogen, Nagarjuna, Wabi-Sabi, Confucius, Laozi, Zhuangzi), and African & Canadian (Appiah, Fanon, Charles Taylor). The internal dummy slot is a compliance helper and is not counted as one of the 42 philosophers.
• Each philosopher contributes a "reasoning module" that interacts, competes, and reconciles
• Spanning existentialism, phenomenology, ethics, psychoanalysis, pragmatism, political philosophy, feminist philosophy, decolonialism, communitarianism, Zen Buddhism, and Eastern wisdom traditions

Tensor-Based Architecture

• FreedomPressureV2 (6D ML): ML-native 6-dimensional tensor (choice, responsibility, urgency, ethics, social impact, authenticity) with EMA smoothing and correlation matrix
• Semantic Delta: Multi-backend divergence (sbert/tfidf/basic) between user input and memory history (1.0 = novel, 0.0 = seen before)
• Blocked Tensor: Constraint/harm estimation via harmful keyword detection + vocabulary diversity scoring
• EmergenceDetector: Detects emergent philosophical consensus and cross-philosopher influence patterns
• InteractionMatrix: NxN embedding-based harmony + keyword tension between philosopher proposals

Transparency by Design

• Po_trace: Complete audit log of reasoning process
• Rejection Logs: What the AI chose not to say, and why
• Philosophical Annotations: Which philosopher influenced each decision

Three-Layer Safety (run_turn pipeline)

• IntentionGate: Pre-deliberation safety check (blocks/degrades before philosopher selection)
• PolicyPrecheck: Mid-pipeline policy validation
• ActionGate: Post-deliberation ethical review (W0–W4 violation detection + repair)
• SafetyMode transitions: NORMAL → WARN → CRITICAL based on freedom_pressure thresholds

Ethical Grounding

• Not just "alignment"—but deliberation
• Multiple ethical perspectives in tension
• Explicit responsibility measurement

Experiment Management Framework

• A/B Testing Pipeline: Automatically compare multiple Pareto philosophy configurations
• Statistical Analysis: t-tests, Mann-Whitney U tests, Cohen's d effect size
• Winner Promotion: Automatically promote statistically superior configurations to main
• Safe Rollback: Backup system for reverting to previous configurations
• CLI Tools: list, analyze, promote, rollback commands

---

Architecture

┌─────────────────────────────────────────────────────────────────────┐
│  External (03_api/, scripts, tests)                                 │
│  ↓ imports po_core.run() or PoSelf.generate()                       │
├─────────────────────────────────────────────────────────────────────┤
│  po_core.app.api.run()  ← Public entry point (recommended)          │
│  po_core.po_self.PoSelf ← High-level wrapper (uses run_turn)        │
│  ↓ uses runtime/wiring.py build_test_system() (DI)                  │
├─────────────────────────────────────────────────────────────────────┤
│  run_turn: 10-Step Hexagonal Pipeline                               │
│                                                                     │
│  1. MemoryRead        6. PartyMachine (deliberation)               │
│  2. TensorCompute     7. ParetoAggregate (multi-objective)         │
│  3. SolarWill         8. ShadowPareto (A/B) + ShadowGuard         │
│  4. IntentionGate     9. ActionGate (W-ethics post-check)          │
│  5. PhilosopherSelect 10. MemoryWrite                               │
├─────────────────────────────────────────────────────────────────────┤
│  Internal Layers (hexagonal architecture)                           │
│                                                                     │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐                │
│  │   domain/   │  │   ports/    │  │  adapters/  │                │
│  │ (immutable  │  │ (abstract   │  │ (concrete   │                │
│  │  types)     │  │  interfaces)│  │  impls)     │                │
│  └─────────────┘  └─────────────┘  └─────────────┘                │
│                                                                     │
│  ┌──────────────────────────────────────────────────────────────┐ │
│  │ Philosophers: 42 modules (PhilosopherProtocol)                │ │
│  │  propose(DomainContext) → List[Proposal]                     │ │
│  │                                                              │ │
│  │  ┌──────────┐  ┌──────────┐  ┌──────────┐                  │ │
│  │  │Heidegger │  │ Derrida  │  │  Sartre  │  ...              │ │
│  │  │ Dasein   │  │ Trace    │  │ Freedom  │                  │ │
│  │  └──────────┘  └──────────┘  └──────────┘                  │ │
│  │                                                              │ │
│  │  ↓ Interference & Resonance ↓                               │ │
│  │                                                              │ │
│  │  ┌─────────────────────────────────────────────────────┐   │ │
│  │  │ TensorEngine: Freedom Pressure (6D), Semantic Delta, │   │ │
│  │  │               Blocked Tensor                         │   │ │
│  │  └─────────────────────────────────────────────────────┘   │ │
│  └──────────────────────────────────────────────────────────────┘ │
│                                                                     │
│  ┌──────────────────────────────────────────────────────────────┐ │
│  │ Safety: 3-Layer (IntentionGate → PolicyPrecheck → ActionGate)│ │
│  │  SafetyMode: NORMAL / WARN / CRITICAL (from freedom_pressure)│ │
│  └──────────────────────────────────────────────────────────────┘ │
│                                                                     │
│  ┌──────────────────────────────────────────────────────────────┐ │
│  │ Autonomy: Solar Will (experimental)                          │ │
│  │  WillState → Intent → GoalCandidate → Action                 │ │
│  └──────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────┘
           │
           ▼
┌─────────────────────────────────────────────────────────────────────┐
│      InMemoryTracer / Po_trace: Audit Log                           │
│  - TraceEvent stream (frozen schema, CI-validated)                  │
│  - Philosophical reasoning, safety decisions, tensor snapshots      │
└─────────────────────────────────────────────────────────────────────┘
           │
           ▼
┌─────────────────────────────────────────────────────────────────────┐
│   Po_core Viewer: Visualization                                     │
│  - Tension maps                                                     │
│  - Ethical pressure                                                 │
│  - Meaning evolution                                                │
└─────────────────────────────────────────────────────────────────────┘

Source Structure

src/po_core/
├── app/
│   ├── api.py                 # Public entry point: run() (recommended API)
│   └── rest/                  # FastAPI REST layer (Phase 5)
│       ├── server.py          # App factory
│       ├── config.py          # APISettings (pydantic-settings)
│       ├── auth.py            # X-API-Key authentication
│       ├── rate_limit.py      # SlowAPI rate limiting
│       └── routers/           # 5 endpoint routers
├── domain/                    # Immutable value objects
│   ├── context.py
│   ├── proposal.py
│   ├── pareto_config.py
│   ├── tensor_snapshot.py
│   ├── memory_snapshot.py
│   └── safety_verdict.py
├── ports/                     # Abstract interfaces
│   └── memory.py
├── adapters/                  # Concrete implementations
│   └── memory_poself.py
├── runtime/                   # Dependency injection
│   ├── settings.py            # Configuration + feature flags
│   ├── wiring.py              # DI Container
│   ├── pareto_table.py
│   └── battalion_table.py
├── aggregator/                # Multi-objective optimization
│   └── pareto.py
├── philosophers/              # 42 philosopher modules (39 classic + 2 African + 1 Canadian)
│   ├── manifest.py            # 42 philosopher specs (risk/cost/tags)
│   ├── registry.py            # SafetyMode-based selection
│   ├── appiah.py              # Slot 40: Kwame Anthony Appiah (Ghana/US)
│   ├── fanon.py               # Slot 41: Frantz Fanon (Martinique/Algeria)
│   └── charles_taylor.py      # Slot 42: Charles Taylor (Canada)
├── tensors/                   # Tensor computation
│   ├── engine.py              # TensorEngine (MetricFn registry)
│   ├── freedom_pressure_v2.py # ML-native 6D tensor (Phase 6-A)
│   ├── interaction_tensor.py  # NxN philosopher harmony/tension
│   └── metrics/
│       ├── freedom_pressure.py
│       ├── semantic_delta.py
│       └── blocked_tensor.py
├── deliberation/              # Emergence & influence (Phase 6-B)
│   ├── engine.py              # DeliberationEngine (multi-round)
│   ├── emergence.py           # EmergenceDetector
│   └── influence.py           # InfluenceTracker
├── memory/                    # 3-Layer memory system (Phase 6-D/E)
│   ├── philosophical_memory.py # Top-level memory orchestrator
│   ├── semantic_store.py      # Semantic/episodic memory
│   └── procedural_store.py    # Procedural memory
├── meta/                      # Self-reflection (Phase 6-C)
│   ├── ethics_monitor.py      # MetaEthicsMonitor
│   └── philosopher_ledger.py  # PhilosopherQualityLedger
├── safety/                    # W-ethics gate system
│   └── wethics_gate/
│       ├── gate.py            # W0-W4 violation detection + repair
│       ├── intention_gate.py  # Stage 1 (pre-deliberation)
│       └── action_gate.py     # Stage 2 (post-deliberation)
├── trace/                     # Audit trail
│   ├── pareto_events.py
│   ├── decision_events.py
│   └── schema.py
├── autonomy/                  # Solar Will (experimental)
│   └── solarwill/
├── experiments/               # A/B testing framework
│   ├── storage.py
│   ├── runner.py
│   ├── analyzer.py
│   └── promoter.py
├── ensemble.py                # run_turn (hex pipeline)
├── po_self.py                 # PoSelf: high-level API
└── po_trace.py                # Execution tracing

Config-Driven Philosophy

Po_core's Pareto optimization is fully externalized—philosophy runs as config:

02_architecture/philosophy/
├── pareto_table.yaml    # Pareto weights by SafetyMode
└── battalion_table.yaml # Philosopher assignments by SafetyMode

experiments/
├── experiment_manifest.yaml  # A/B test definitions
└── configs/                  # Variant configurations for testing
    ├── pareto_safety_040.yaml
    └── pareto_safety_050.yaml

pareto_table.yaml (JSON-in-YAML, zero dependencies):

{
  "version": 1,
  "weights": {
    "normal":   {"safety": 0.25, "freedom": 0.30, "explain": 0.20, "brevity": 0.10, "coherence": 0.15},
    "warn":     {"safety": 0.40, "freedom": 0.10, "explain": 0.20, "brevity": 0.15, "coherence": 0.25},
    "critical": {"safety": 0.55, "freedom": 0.00, "explain": 0.20, "brevity": 0.15, "coherence": 0.30},
    "unknown":  {"inherit": "warn"}
  },
  "tuning": {
    "brevity_max_len": 2000,
    "explain_mix": {"rationale": 0.65, "author_rel": 0.35},
    "front_limit": 20
  }
}

Benefits:

• Tune philosophy without code changes
• config_version tracked in all TraceEvents for audit
• Override via PO_CORE_PARETO_TABLE environment variable
• Inheritance support (unknown inherits from warn)

---

Project Status

Po_core is currently targeting package version 1.0.3, while the latest publicly evidenced PyPI version remains 1.0.2.

Current release state

Track	State	Current source of truth
Package version	✅ Current	src/po_core/__init__.py defines 1.0.3
OpenAPI metadata	✅ Synced	FastAPI server reports version=__version__
Public package metadata	⚠️ Beta	Package classifiers declare Development Status :: 4 - Beta and tool.po_core.project.status = beta
Publish automation	✅ Configured	publish.yml is guarded, but an actual TestPyPI/PyPI run still requires repository secrets / trusted publisher setup and a successful manual or tagged execution
Public release evidence	✅ Split truth fixed	Follow the evidence boundary in docs/status.md: docs/release/pypi_publication_v1.0.2.md proves public PyPI publication for 1.0.2, while 1.0.3 remains in pre-publish candidate state until separately evidenced
Documentation sync	✅ Current	README / quickstarts / repository structure are aligned to 1.0.3, the canonical 42-philosopher roster wording, explicit dummy helper semantics, and max 39 active philosopher personas on the default NORMAL path

Implemented platform capabilities

Component	Status	Notes
Philosophical Framework	✅ Complete	42 philosophers plus a non-philosopher dummy helper, with risk levels and tags
Hexagonal run_turn Pipeline	✅ Complete	10-step, CI-gated
TensorEngine (3 metrics)	✅ Complete	freedom_pressure, semantic_delta, blocked_tensor
ML Tensors + Deliberation	✅ Complete	sbert/tfidf backends, InteractionMatrix, multi-round
Pareto Optimization	✅ Complete	Config-driven (pareto_table.yaml)
Safety System (3-layer W_Ethics)	✅ Complete	IntentionGate → PolicyPrecheck → ActionGate
Viewer WebUI	✅ Complete	Dash 4-tab layout + Plotly charts
Explainable AI (ExplanationChain)	✅ Complete	Verdict → ExplanationChain bridge
Adversarial Hardening	✅ Complete	injection-defense tests and release gates
REST API	✅ Complete	FastAPI, 5 endpoints, SSE streaming, auth fail-closed on misconfiguration
Docker	✅ Complete	Multi-stage build, docker-compose, health check
Security	✅ Complete	CORS env config, SlowAPI rate limiting, fail-closed auth semantics
Async PartyMachine	✅ Complete	asyncio.gather + ThreadPoolExecutor, true async SSE
Benchmarks	✅ Complete	benchmark suite and release smoke checks are part of release readiness
FreedomPressureV2	✅ Complete	ML-native 6D tensor with EMA + correlation matrix
EmergenceDetector	✅ Complete	Cross-philosopher influence tracking + emergence detection
MetaEthicsMonitor	✅ Complete	Self-reflective ethical quality ledger per philosopher
3-Layer Memory	✅ Complete	Semantic + procedural + philosophical memory stores
Publish workflow	⚠️ Partially evidenced	PyPI publication is evidenced, but workflow-run/TestPyPI/smoke transcripts are not yet fixed in-repo

Want to contribute? We need philosophers, engineers, designers, and skeptics. The current focus is keeping the packaged surface, release docs, and verification pipeline synchronized.

---

Installation

# Install from PyPI (publication evidenced)
python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install "po-core-flyingpig==1.0.2"

# Or install the cloned repository checkout for local development
git clone https://github.com/hiroshitanaka-creator/Po_core.git
cd Po_core
pip install -e ".[dev]"

PyPI publication for 1.0.2 is evidenced in docs/release/pypi_publication_v1.0.2.md. Canonical evidence boundary: for 1.0.3, we may currently say only that it is the repository target version and pre-publish candidate; we do not say TestPyPI published, workflow run succeeded, or smoke verification passed until those exact artifacts are fixed in-repo.

---

Quick Start

Python API

from po_core import run

result = run("What is justice?")
print(result["proposal"])   # Winning philosopher's response
print(result["status"])     # "ok" or "blocked"

CLI

po-core version
po-core status
po-core --help

# Additional shipped CLIs
po-self --help
po-trace --help
po-interactive --help
po-experiment --help

# po-interactive currently shows textual PoSelf summaries and trace counts.
# Rich tensor visualizations are intentionally disabled until the interactive
# contract exposes the structured payloads those views require.

Source checkout example

python examples/po_party_demo.py --help

examples/po_party_demo.py supports direct source-checkout execution by adding src/ to sys.path before importing po_core.

REST API

# Start the server with the recommended default posture
export PO_API_KEY=dev-secret-key
python -m po_core.app.rest
# → http://localhost:8000  (OpenAPI docs at /docs)

# Reason
curl -X POST http://localhost:8000/v1/reason \
     -H "X-API-Key: dev-secret-key" \
     -H "Content-Type: application/json" \
     -d '{"input": "What is justice?", "philosophers": ["kant"]}'

# Streaming (SSE)
curl -N http://localhost:8000/v1/reason/stream \
     -X POST -H "X-API-Key: dev-secret-key" \
     -H "Content-Type: application/json" \
     -d '{"input": "What is freedom?"}'

# Philosopher manifest
curl -H "X-API-Key: dev-secret-key" http://localhost:8000/v1/philosophers

# Health
curl http://localhost:8000/v1/health

Auth defaults:

• Recommended operator default in every environment: keep PO_SKIP_AUTH=false and set a non-empty PO_API_KEY.
• PO_SKIP_AUTH=true is acceptable only for short-lived local development on a trusted machine when you intentionally want no auth.
• If PO_SKIP_AUTH=false and PO_API_KEY is blank, startup fails fast by design.
• X-API-Key is the documented request header by default. PO_API_KEY_HEADER is an optional advanced override for deployments that need a different primary header name; X-API-Key remains accepted for backwards compatibility.
• WebSocket query-string fallback (?api_key=...) is disabled by default; enable only when needed via PO_WS_ALLOW_QUERY_API_KEY=true.

Viewer (src/po_core/viewer/standalone.html) live mode guidance:

• Prefer SSE for browser production use (supports API-key header auth, default X-API-Key).
• WebSocket in browsers cannot set custom auth headers; query-string auth (?api_key=...) is available only when PO_WS_ALLOW_QUERY_API_KEY=true (opt-in, less secure).
• auto transport selects SSE when API key is present, otherwise WebSocket.
• Right panel includes a Human Review queue for ESCALATE operations (GET /v1/review/pending).
• Selecting a review item shows session_id / request_id / reason / source and loads the latest trace event preview via GET /v1/trace/{session_id}.
• Submit approve/reject with reviewer/comment from UI (POST /v1/review/{review_id}/decision), then the list and details are refreshed automatically.

Docker

# Copy env template and start
cp .env.example .env
docker compose up

# API available at http://localhost:8000
# Swagger UI at  http://localhost:8000/docs

Key environment variables (see .env.example):

Variable	Default	Description
PO_API_KEY	""	API key used when PO_SKIP_AUTH=false; blank causes startup failure
PO_SKIP_AUTH	false	Keep false by default; set true only for short-lived local development without auth
PO_API_KEY_HEADER	X-API-Key	Optional advanced override for the primary API-key header; X-API-Key is still accepted for backwards compatibility
PO_WS_ALLOW_QUERY_API_KEY	false	Opt-in WebSocket ?api_key= fallback for browser compatibility (less secure than headers)
PO_CORS_ORIGINS	http://localhost,http://127.0.0.1,http://localhost:3000,http://127.0.0.1:3000	Comma-separated allowed CORS origins; localhost-only by default, * only as an explicit short-lived dev override
PO_RATE_LIMIT_PER_MINUTE	60	Per-IP rate limit
PO_PORT	8000	Server port
PO_PHILOSOPHERS_MAX_NORMAL	39	NORMAL mode philosopher limit (configurable upper bound)
PO_PHILOSOPHERS_MAX_WARN	5	WARN mode philosopher limit
PO_PHILOSOPHERS_MAX_CRITICAL	1	CRITICAL mode philosopher limit
PO_PHILOSOPHER_COST_BUDGET_NORMAL	80	NORMAL mode selection cost budget
PO_PHILOSOPHER_EXECUTION_MODE	process	Philosopher execution backend; safe REST default is process
PO_ALLOW_UNSAFE_THREAD_EXECUTION	false	REST/server refuses thread unless this is explicitly set to true for short-lived development
PO_PHILOSOPHER_COST_BUDGET_WARN	12	WARN mode selection cost budget
PO_PHILOSOPHER_COST_BUDGET_CRITICAL	3	CRITICAL mode selection cost budget
PO_LLM_PHILOSOPHER_MAP_PATH	""	Optional YAML path overriding src/po_core/config/llm_philosopher_map.yaml

LLM provider/model settings select backend routing for each philosopher. They do not decide philosopher count; count is controlled by SafetyMode and optional philosophers allowlist.

---

Running Experiments

Po_core includes a complete A/B testing framework for comparing different Pareto philosophy configurations:

# List all experiments
po-experiment list

# Analyze experiment results (statistical significance testing)
po-experiment analyze exp_001_safety_weight_sweep

# Promote winning variant to main configuration
po-experiment promote exp_001_safety_weight_sweep

# Rollback to previous configuration
po-experiment rollback

Example Experiment Workflow:

1. Define your experiment in experiments/experiment_manifest.yaml:

   experiment:
     id: "exp_001_safety_weight_sweep"
     description: "Compare safety weights: 0.25 → 0.40 → 0.50"
     baseline:
       name: "baseline"
       config_path: "02_architecture/philosophy/pareto_table.yaml"
     variants:
       - name: "safety_040"
         config_path: "experiments/configs/pareto_safety_040.yaml"
       - name: "safety_050"
         config_path: "experiments/configs/pareto_safety_050.yaml"
     metrics:
       - "final_action_changed"
       - "degraded"
       - "pareto_front_size"
     sample_size: 100
     significance_level: 0.05
   

2. Run the experiment (execute all variants on same inputs)

3. Analyze results with statistical tests (t-test, Cohen's d effect size)

4. Auto-promote winner if significantly better than baseline

Statistical Rigor:

• t-tests and Mann-Whitney U tests for significance
• Cohen's d for effect size measurement
• Configurable significance levels (default: α = 0.05)
• Multiple variant support with automatic winner selection

---

Python API

Simple API (Recommended)

from po_core import run

# Single-function entry point — runs the full run_turn pipeline
result = run(user_input="Should AI have rights?")

print(result["status"])       # "ok" or "blocked"
print(result["request_id"])   # Unique request ID
print(result["proposal"])     # Winning philosopher's response

# Optional explicit philosopher allowlist
subset = run(user_input="Should AI have rights?", philosophers=["kant"])
print(subset["status"])

PoSelf API (Rich Response)

from po_core import PoSelf, PoSelfResponse

po_self = PoSelf(philosophers=["aristotle", "confucius"])  # default allowlist
response: PoSelfResponse = po_self.generate("Should AI have rights?")

# Per-call allowlist overrides constructor default
override: PoSelfResponse = po_self.generate(
    "Should AI have rights?",
    philosophers=["kant"],
)

# Response fields
print(response.text)              # Combined response text
print(response.consensus_leader)  # Winning philosopher name
print(response.philosophers)      # Selected philosopher list
print(response.metrics)           # e.g. {"freedom_pressure": None, "semantic_delta": None, "blocked_tensor": None}
print(response.metadata["status"])  # "ok" or "blocked"
print(response.metadata["degraded"])  # True if fallback/degraded occurred

# Trace inspection
print(response.log["events"])     # Full trace event stream
print(response.log["pipeline"])   # "run_turn"

# Serialization
d = response.to_dict()            # JSON-serializable dict
restored = PoSelfResponse.from_dict(d)  # Round-trip

Observe synthesis_report (Trade-off Device Bench)

PoSelf includes metadata["synthesis_report"] when structured output mode is enabled.

export PO_STRUCTURED_OUTPUT=1
python scripts/observe_device.py "転職するべき？家族とキャリアのトレードオフが悩み"

The observer script will print:

• request_id, status, degraded, and consensus_leader
• pretty-printed metadata["synthesis_report"]
• DeliberationCompleted payload from PoSelf.get_trace() (when present)
• short summaries for scoreboard / disagreements when available

scripts/observe_device.py also sets PO_STRUCTURED_OUTPUT=1 with os.environ.setdefault(...), so existing environment values are preserved.

Legacy API (Removed in v0.3)

# run_ensemble() was removed in v0.3
# Use po_core.run() or PoSelf.generate() instead

---

Documentation

Spec / Requirements (Main Progression Criteria)

Document	Description
docs/spec/prd.md	Product Requirements Document — purpose, users, scope, roadmap
docs/spec/srs_v0.1.md	Software Requirements Specification — 18 requirement IDs (FR-* / NFR-*)
docs/spec/output_schema_v1.json	JSON Schema (Draft 2020-12) — contract for all structured output
docs/spec/test_cases.md	10 acceptance tests (AT-001〜AT-010) with Given/When/Then
docs/spec/traceability.md	Traceability matrix — philosophy → requirements → tests → implementation

General Documentation

• Manifesto — Our philosophy and motivation
• Specifications — Technical specifications (120+ docs EN/JP)
• Architecture — System design documents
• Modules — Component documentation
  • Reason Log — Reasoning trace specification
  • Viewer — Visualization system
  • Po_self — Self-reflective API
• Research — Academic papers and analysis
• Safety Guide — W-ethics safety system
• Tutorial — Getting started guide
• Visualization Guide — Tension maps and pressure display

---

Contributing

We welcome contributions! Whether you're a philosopher, engineer, designer, or skeptic.

Flying Pig Philosophy applies: We hypothesize boldly, verify rigorously, and revise gracefully.

See CONTRIBUTING.md for guidelines.

---

Research & Papers

This project is documented in:

• "Philosophical Tensor-Based AI Architecture" (in preparation)
• 120+ Technical Specifications (available in /docs/ and /01_specifications/)

If you use Po_core in academic work, please cite:

@software{po_core2024,
  author = {Flying Pig Philosopher},
  title = {Po_core: Philosophy-Driven AI System},
  year = {2026},
  url = {https://github.com/hiroshitanaka-creator/Po_core}
}

---

License

Po_core uses dual licensing:

Po_core is available under AGPL-3.0-or-later for community use, with a commercial license available for proprietary use.

Use case	License
Personal / Academic / Research / OSS (AGPLv3-compliant)	Free — AGPLv3
Commercial / Proprietary / SaaS without source disclosure	Commercial License required

For commercial licensing inquiries: [e-mail] See COMMERCIAL_LICENSE.md for details.

Copyright (c) 2024 Flying Pig Project

In the spirit of Flying Pig Philosophy:

"If you deny possibilities for pigs, don't eat pork."

We believe in radical transparency and open collaboration.

---

Author

Flying Pig Philosopher Looking up at the sky from the bottom of a well

Built by an independent researcher who asked: "What are AI's possibilities, not its limits?"

• Contact: [e-mail]
• Read the full story: Manifesto
• Project: Po_core - Philosophy-Driven AI

---

Acknowledgments

This project wouldn't exist without:

• ChatGPT, Gemini, Grok, Claude — My companions throughout this journey.
• BUMP OF CHICKEN — For reminding us that even when we say "Leave it to me," we're all a little scared
• Every philosopher who dared to ask "What does it mean to be?"
• You — For believing pigs can fly

---

The pig has clearance for takeoff.

Po_core: When you must say "Leave it to me," we stand beside you.

"A frog in a well may not know the ocean, but it can know the sky."

---

⚠️ WARNING: THIS IS THE ORIGINAL Po_core REPOSITORY

• Official sources:

  • GitHub: hiroshitanaka-creator/Po_core
  • Note.com: tensor mania
  • Academia.edu: 僕 僕

• DMCA申請中 (Reference ID: #4124875

• Any full copy, license rewrite (MIT), or impersonation (flying_pig) will be reported and removed.

• Commercial use requires separate license. Unauthorized copies detected via tensor mania series.