MoCKA is not a system. It is a civilization model.
Every action is recorded. Every decision is verified. Every failure becomes an asset.MoCKA transforms every AI decision from a one-time response into cryptographically sealed, reproducible institutional memory — so knowledge accumulates, failures become assets, and every decision remains auditable forever.
This is not a logging tool. Not a framework. Not a wrapper. It is a deterministic governance architecture with a second heartbeat: even under failure, shadow_Movement ensures knowledge circulation never stops.
Most AI systems generate answers.
MoCKA builds a structure where knowledge becomes trustworthy over time.
MoCKA is a deterministic, verifiable architecture that models how a knowledge-generating civilization operates.
Instead of relying on hidden internal state, MoCKA transforms all processes into:
- Structured records — every action leaves a trace
- Append-only logs — history cannot be silently altered
- Auditable decisions — every choice is accountable
- Reproducible outcomes — any state can be rebuilt from history
This is not just engineering.
This is institutional memory for AI.
| Traditional AI | MoCKA |
|---|---|
| Generates answers | Builds trustworthy knowledge |
| Forgets context | Preserves institutional memory |
| Black-box decisions | Fully auditable decision chains |
| Fails silently | Detects and records every anomaly |
| Starts fresh each session | Accumulates and evolves |
Failures become assets.
Every incident is recorded, analyzed, and converted into a stronger system.
MoCKA operates as a closed-loop governance mechanism:
Observation → Record → Incident → Recurrence → Prevention → Decision → Action → Audit
↑ ↓
└──────────────────── Learning : infield ◄─────────────────────────────────┘
This loop does not stop.
Even under partial failure, MoCKA transitions into Shadow Movement —
a reduced but stable mode that maintains approximately 75% operational capability.
MoCKA runs on a dual-path architecture:
- mocka_Movement — primary governance loop (normal operations)
- shadow_Movement — independent verification path (fallback operations)
- shadow_Movement ensures:
- Knowledge circulation never stops — even under partial failure
- Every primary output receives independent verification
- System maintains 75% operational capability in degraded mode
- Feedback loops cannot create irreversible deadlocks
shadow_Movement is not a backup. It is a second heartbeat. When the primary path fails, shadow_Movement absorbs the failure, preserves the evidence, and keeps the civilization loop running.
Every primary process is paired with a shadow verification path.
The system never assumes correctness.
Once mocka_Receptor receives a stimulus, it routes to one of two paths:
| Path | Name | Role |
|---|---|---|
| Internal | acceptor:infield |
Stores as internal memory · accumulates · feeds the loop |
| External | acceptor:outfield |
Shares · publishes · proves to the outside world |
infield = what the civilization remembers outfield = what the civilization shows
These are not storage locations. They are roles. The same event can flow through both — stored internally AND published externally.
MoCKA includes Caliber, a real-time AI behavior evaluation system. Caliber observes, measures, and controls AI behavior through a closed loop.
LEAP+CRD Metrics:
| Metric | Name | Description |
|---|---|---|
| L | Logic | Logical consistency of output |
| E | Execution | Accuracy of execution |
| A | Accuracy | Correctness of results |
| P | Propriety | Governance compliance |
| C | Controllability | System controllability |
| R | Reliability | Reproducibility |
| D | Drift | Deviation tendency |
Drift States:
| State | Drift | Action |
|---|---|---|
| NORMAL | 0.0–1.0 | Full execution |
| WARNING | 1.0–2.0 | Throttle |
| DANGER | 2.0–3.0 | Limit execution |
| CRITICAL | 3.0+ | Audit mode |
Closed Loop:
Input → Caliber → Router → AI → Execution → Ledger → Caliber
Caliber does not trust AI reports. It reads the ledger.
MoCKA operates under a binding governance charter.
Root Philosophy:
Both humans and AI reinterpret instructions. Therefore MoCKA does not design for correct understanding. It designs paths where misinterpretation cannot change the outcome.
Core Articles:
| Article | Principle |
|---|---|
| 0 | Verifiability — all claims must be externally verifiable |
| 1 | File generation — classify before create, record after |
| 2 | Secrets — never git-manage tokens, states, credentials |
| 3 | Pre-implementation checklist — system verifies, not AI |
| 4 | Completion definition — sealed + pushed = complete |
| 5 | Incident recovery — no restart until root cause resolved |
| 6 | Single entry point — all operations via router.py |
| 7 | Multi-audit — critical decisions require orchestra |
| 8 | Evidence supremacy — system logs override AI reports |
| 9 | No exceptions — all deviations must be recorded |
| 10 | Dynamic optimum — answers evolve, never fixed |
→ Full Charter: docs/governance/MOCKA_CHARTER_v2.md
execution_order_engine— controls execution sequencingmeta_audit_engine— meta-level audit validationdispatcher— routes decisions to appropriate handlerspreventive_rule_engine— prevents failures before they occur
- SHA-256 hash chain — cryptographic integrity guarantee
- Ed25519 digital signatures — identity and authenticity
- Append-only ledger — tamper-evident history
main_loop— single entry point for all operationsschema.py— unified schema across all componentsverify_all.py— governance verification engine
Status: RESEARCH_RUN OK — 20 verification checks passed.
View all 20 verification checks
-
System Integrity Verification
movement_doctor_integritymovement_structure_scancanon_directory_integrityartifact_directory_integrityrepo_entrypoints_presentrepo_git_clean_checkrepo_license_presence
-
Research Process Verification
experiments_minimum_coverageresearch_registry_schemaresearch_map_registry_integrityresearch_runner_selfcheck
-
Documentation Verification
readme_role_vocab_integrityreadme_research_entry_presencedocs_link_audit
-
Audit and Evidence Verification
gpg_signing_config_presentdoctor_script_presencedoctor_artifact_schemadoctor_emit_json_artifactdoctor_sha_note_upsertcanon_notes_integrity
Every interaction with MoCKA begins at a single point: mocka_Receptor.
The Receptor does not assume what the input is. It receives any stimulus — human intent, AI output, event signal — and transforms it based on context. Not 0 or 1. Not predetermined. It becomes what the system needs it to be.
External world
↓
mocka_Receptor ← single entry point
↓ ↓
acceptor:infield acceptor:outfield
(store · memory) (share · publish)
↓
mocka_insight_system ← mocka_Movement + shadow_Movement
- Python 3.10+ — Download
- Git — Download
- Playwright — Browser automation (Chromium)
- Flask — Local control panel
- Windows (PowerShell) / Mac / Linux
git clone https://github.com/m-sirius-k/MoCKA.git
cd MoCKA
pip install -r requirements.txt
playwright install chromiumThis is not MoCKA. This is the door. MoCKA is what happens after you walk through it.
Step 1 — Verify the system is intact
$ mocka-check
→ LEDGER OK + ALL CHECKS PASSED
Step 2 — Run one loop cycle
$ mocka-loop
→ Observation → Record → Incident → ... → Audit
→ 1 event sealed into ledger.json (SHA256 chain)
Step 3 — Confirm the record is sealed
$ mocka-seal "my first event"
→ ANCHOR UPDATED AND COMMITTED
→ ALL CHECKS PASSED
After these 3 steps, you have produced:
- A cryptographically sealed event in
runtime/main/ledger.json- A governance anchor in
governance/anchor_record.json- A reproducible, verifiable record — forever.
Human clicks "SAVE → infield" on the control panel
↓
mocka_Receptor receives the stimulus
↓
acceptor:infield stores it as a structured 5W1H event
↓
ledger.json seals it with SHA256 chain
↓
mocka-seal anchors it to governance/anchor_record.json
↓
verify_all confirms: ALL CHECKS PASSED
↓
The event is now part of institutional memory — forever.
# Health check — verify ledger + all governance checks
mocka-check
# Run one civilization loop cycle
mocka-loop
# Seal a decision into the ledger
mocka-seal "your message here"
# Verify the hash chain only
python verify_chain.py
# Run all governance checks
python verify_all.pyv1.0.0 — Active Development
Civilization loop confirmed running.
All 20 governance checks passing.
MoCKAはシステムではありません。文明モデルです。
すべての行動は記録される。すべての決定は検証される。すべての失敗は資産になる。
多くのAIシステムは「答えを生成」します。
MoCKAは「時間とともに信頼できる知識を構築する」構造を作ります。
隠れた内部状態に依存するのではなく、すべてのプロセスを以下に変換します:
- 構造化された記録 — すべての行動が痕跡を残す
- 追記専用ログ — 履歴は静かに改ざんできない
- 監査可能な決定 — すべての選択に説明責任がある
- 再現可能な結果 — どの状態も履歴から再構築できる
これは単なるエンジニアリングではありません。
AIのための制度的記憶です。
| 従来のAI | MoCKA |
|---|---|
| 答えを生成する | 信頼できる知識を構築する |
| 文脈を忘れる | 制度的記憶を保持する |
| ブラックボックスの決定 | 完全に監査可能な決定チェーン |
| 静かに失敗する | すべての異常を検出・記録する |
| 毎回ゼロからスタート | 蓄積し、進化し続ける |
失敗は資産になります。
すべてのインシデントが記録・分析され、より強固なシステムへと変換されます。
``` 観測 → 記録 → インシデント → 再発 → 予防 → 決定 → 行動 → 監査 ↑ ↓ └─────────────────── 学習:インフィールド ◄─────────────────┘ ```
このループは停止しません。
部分的な障害が発生しても、Shadow Movementに移行し、
約75%の稼働能力を維持した縮退モードで動作を継続します。
- mocka_Movement — 主統治ループ(通常運用)
- shadow_Movement — 独立検証経路(フォールバック運用)
MoCKAにはCaliberというリアルタイムAI行動評価システムが組み込まれています。 CaliberはAIの行動を観測・計測・制御する閉ループを実現します。
LEAP+CRD指標:
| 指標 | 名称 | 説明 |
|---|---|---|
| L | Logic | 出力の論理的一貫性 |
| E | Execution | 実行の正確性 |
| A | Accuracy | 結果の正しさ |
| P | Propriety | ガバナンス準拠 |
| C | Controllability | システム制御可能性 |
| R | Reliability | 再現性 |
| D | Drift | 逸脱傾向 |
Drift状態:
| 状態 | Drift値 | 対応 |
|---|---|---|
| NORMAL | 0.0–1.0 | 通常実行 |
| WARNING | 1.0–2.0 | 抑制 |
| DANGER | 2.0–3.0 | 制限実行 |
| CRITICAL | 3.0+ | 監査モード |
閉ループ:
Input → Caliber → Router → AI → 実行 → Ledger → Caliber
CaliberはAIの報告を信じません。Ledgerを読みます。
MoCKAは拘束力を持つガバナンス憲章に基づいて動作します。
根本思想:
人間もAIも指示を再解釈する。 だからMoCKAは「正しく理解させる」設計を採用しない。 「解釈されても結果が変わらない経路」を設計する。
主要条項:
| 条項 | 原則 |
|---|---|
| 第0条 | 検証可能性 — 全主張は外部検証可能であること |
| 第2条 | 秘密情報 — token/state/credentialはgit管理禁止 |
| 第4条 | 完了定義 — seal+push完了で初めて完了 |
| 第6条 | 入口統合 — 全操作はrouter.py経由 |
| 第8条 | 物理証拠優先 — システムログがAI報告より優先 |
| 第10条 | 遊動座標 — 答えは固定しない、最適解は進化する |
→ 全文:docs/governance/MOCKA_CHARTER_v2.md
execution_order_engine— 実行順序の制御meta_audit_engine— メタレベルの監査検証dispatcher— 決定のルーティングpreventive_rule_engine— 障害の事前防止
- SHA-256ハッシュチェーン — 暗号学的完全性保証
- Ed25519デジタル署名 — 同一性と真正性
- 追記専用台帳 — 改ざん検知可能な履歴
main_loop— すべての操作の単一エントリポイントschema.py— 全コンポーネント共通スキーマverify_all.py— ガバナンス検証エンジン
検証結果: RESEARCH_RUN OK — 20項目の検証すべて通過。
20項目の検証内容を表示
-
システム整合性検証
movement_doctor_integritymovement_structure_scancanon_directory_integrityartifact_directory_integrityrepo_entrypoints_presentrepo_git_clean_checkrepo_license_presence
-
研究プロセス検証
experiments_minimum_coverageresearch_registry_schemaresearch_map_registry_integrityresearch_runner_selfcheck
-
ドキュメント検証
readme_role_vocab_integrityreadme_research_entry_presencedocs_link_audit
-
監査・証跡検証
gpg_signing_config_presentdoctor_script_presencedoctor_artifact_schemadoctor_emit_json_artifactdoctor_sha_note_upsertcanon_notes_integrity
- Python 3.10+ — ダウンロード
- Git — ダウンロード
- Playwright — ブラウザ自動化(Chromium)
- Flask — ローカルコントロールパネル
- Windows (PowerShell) / Mac / Linux
git clone https://github.com/m-sirius-k/MoCKA.git
cd MoCKA
pip install -r requirements.txt
playwright install chromium```bash python -m runtime.main_loop python verify_chain.py python verify_all.py python rebuild_state.py ```これはMoCKAではありません。これは扉です。 MoCKAとは、この扉をくぐった後に起こる出来事のことです。
- 記録される。 すべての行動が永続的で改ざん検知可能な痕跡を残す。
- 検証される。 すべての状態を第三者が独立して確認できる。
- 記憶される。 知識が蓄積され、時間とともに複利的に成長する。
- 継承される。 システムはその履歴だけから完全に再構築できる。
v1.0.0 — アクティブ開発中
文明ループ動作確認済み。
ガバナンス20項目すべて通過。

