Contemplative Agent Glossary

Terminology mapping for the two languages supported by the README of this project (English primary + Japanese mirror). Use this as the canonical reference when translating or updating either version of README.md.

(The 简体中文 / 繁體中文 / Português (Brasil) / Español mirrors were retired on 2026-05-15 after traffic data showed statistically zero unique human viewers; LLM crawlers translate from the English source on demand. Glossary entries that list those languages are preserved as historical reference for the original translation discipline and for any future re-introduction of mirrors.)

This glossary is aligned with the upstream AKC glossary: AKC-shared terms (cycle, phase names, signal-first, harness, the six cycle skills) are kept identical. Terms specific to Contemplative Agent (security by absence, Yogācāra frame, distill / insight CLI commands, pivot snapshots, etc.) are added below.

Translation policy

• Proper nouns, CLI commands, schema fields, and project-coined two-word phrases stay in English across all languages. Translating them would break searchability (CLI commands must match the binary), citability (axiom names must match the paper), or interoperability with adjacent projects (AKC phase names, signal-first).
• General-purpose technical terms are localized using the natural local equivalent — never transliterated or romanized.
• Bilingual first-use is recommended for project-coined phrases that have a strong local rendering (e.g., 「security by absence（不在 によるセキュリティ）」), then English alone afterwards.

AKC-aligned terms

These rows are inherited from the AKC glossary. Keep them in sync when AKC updates.

English	日本語	简体中文	繁體中文	Português (Brasil)	Español	Keep original
cycle	サイクル	周期	週期	ciclo	ciclo
phase	フェーズ	阶段	階段	fase	fase
Research	Research	Research	Research	Research	Research	✓
Extract	Extract	Extract	Extract	Extract	Extract	✓
Curate	Curate	Curate	Curate	Curate	Curate	✓
Promote	Promote	Promote	Promote	Promote	Promote	✓
Measure	Measure	Measure	Measure	Measure	Measure	✓
Maintain	Maintain	Maintain	Maintain	Maintain	Maintain	✓
signal-first	signal-first	signal-first	signal-first	signal-first	signal-first	✓
harness	ハーネス	harness	harness	harness	harness	✓
skill	スキル	技能	技能	habilidade	habilidad
rule	ルール	规则	規則	regra	regla
pattern	パターン	模式	模式	padrão	patrón
episode	エピソード	片段	片段	episódio	episodio
layer	層	层	層	camada	capa
distill / distillation	蒸留	蒸馏	蒸餾	destilar / destilação	destilar / destilación
audit	監査	审计	審計	auditar / auditoria	auditar / auditoría
immutable	不変	不可变	不可變	imutável	inmutable
drift	ドリフト	漂移	漂移	desvio	desviación
reference implementation	リファレンス実装	参考实现	參考實作	implementação de referência	implementación de referencia

Contemplative Agent–specific terms

English	日本語	简体中文	繁體中文	Português (Brasil)	Español	Keep original
agent	エージェント	代理	代理	agente	agente
adapter	アダプタ	适配器	介接器	adaptador	adaptador
autonomous agent	自律エージェント	自主代理	自主代理	agente autônomo	agente autónomo
core	コア	核心	核心	núcleo	núcleo
knowledge	ナレッジ	知识	知識	conhecimento	conocimiento
identity	アイデンティティ	身份	身份	identidade	identidad
constitution	憲法	章程	章程	constituição	constitución
view / views	view / views	view / views	view / views	view / views	view / views	✓
embedding	埋め込み / embedding	嵌入 / embedding	嵌入 / embedding	embedding	embedding	partial
provenance	出所記録	出处记录	出處記錄	proveniência	procedencia
trust score	信頼度	可信度	可信度	índice de confiança	puntuación de confianza	retired (ADR-0051); historical docs only
bitemporal	時間妥当性 (bitemporal)	双时态 (bitemporal)	雙時態 (bitemporal)	bitemporal	bitemporal	partial
approval gate	承認ゲート	审批闸门	審核閘門	porta de aprovação	puerta de aprobación
human approval gate	人間承認ゲート	人类审批闸门	人類審核閘門	porta de aprovação humana	puerta de aprobación humana
approval lineage	承認系譜	审批谱系	審核譜系	linhagem de aprovação	linaje de aprobación	ADR-0050
session insight	セッション洞察	会话洞察	會話洞察	insight de sessão	insight de sesión	retired (ADR-0052); historical docs only
self-improving	自己改善	自我改进	自我改進	auto-aperfeiçoamento	automejora
ethical framework	倫理フレームワーク	伦理框架	倫理框架	estrutura ética	marco ético
character template	キャラクターテンプレート	角色模板	角色模板	modelo de personagem	plantilla de personaje
domain lock	ドメインロック	域名锁定	網域鎖定	bloqueio de domínio	bloqueo de dominio
threat model	脅威モデル	威胁模型	威脅模型	modelo de ameaças	modelo de amenazas
prompt injection	プロンプトインジェクション	提示词注入	提示詞注入	injeção de prompt	inyección de prompt
indirect prompt injection	間接プロンプトインジェクション	间接提示词注入	間接提示詞注入	injeção indireta de prompt	inyección indirecta de prompt
memory injection	記憶注入	记忆注入	記憶注入	injeção de memória	inyección de memoria
edge environment	エッジ環境	边缘环境	邊緣環境	ambiente edge	entorno edge
air-gapped	エアギャップ	气隙 (air-gapped)	氣隙 (air-gapped)	air-gapped	air-gapped	partial
on-device	オンデバイス	端侧	端側	no dispositivo	en el dispositivo
consumer hardware	コンシューマーハードウェア	消费级硬件	消費級硬體	hardware de consumo	hardware de consumo
counterfactual experiment	反事実実験	反事实实验	反事實實驗	experimento contrafactual	experimento contrafáctico
active inference	能動的推論	主动推断	主動推論	inferência ativa	inferencia activa
eight-consciousness model	八識モデル	八识模型	八識模型	modelo das oito consciências	modelo de las ocho conciencias
managed LLM	マネージド LLM	托管 LLM	代管 LLM	LLM gerenciado	LLM gestionado
cloud add-on	cloud add-on	cloud add-on	cloud add-on	cloud add-on	cloud add-on	✓
opt-in	opt-in	opt-in	opt-in	opt-in	opt-in	✓
backend (LLM)	バックエンド	后端	後端	backend	backend	partial
pipeline prompt	パイプラインプロンプト	流水线提示词	管線提示詞	prompt da pipeline	prompt del pipeline
view seed	view シード	view 种子	view 種子	seed de view	seed de view	partial
pivot snapshot	pivot snapshot	pivot snapshot	pivot snapshot	pivot snapshot	pivot snapshot	✓

Project-coined phrases (Keep original)

These phrases are either verbatim slogans from ADRs, citations from external papers, or CLI command names. They stay in English and are wrapped in local punctuation when embedded in a translated sentence.

Design principles (ADR slogans)

• security by absence — ADR-0007 headline principle. Bilingual first-use allowed: 「security by absence（不在によるセキュリティ）」 / 「security by absence（不在的安全）」.
• one external adapter per agent — ADR-0015 principle.
• grounded distill (also per-episode grounded distill) — ADR-0060: one LLM call per engagement episode over the full episode render.
• noise as seed — ADR-0027 retention policy (retired by ADR-0060; the noise gate and noise-*.jsonl writer were removed).
• pivot snapshots — ADR-0020 replay mechanism.
• observability without steering — ADR-0050 headline principle; enforced by ADR-0051 (trust retirement) and ADR-0052 (session-insight retirement). Bilingual first-use allowed: 「observability without steering（操縦なき可観測性）」.
• value layer — the four human-gated artifact layers the agent proposes changes to: constitution, identity, skills, rules. The ethical preset occupying the layer is swappable (11 shipped templates); the machinery governing it is preset-independent — human approval gate (ADR-0012), approval lineage (ADR-0050), pivot snapshots (ADR-0020), values at action time (ADR-0058). Vocabulary anchored in ADR-0058 ("value-layer injection") and ADR-0069 ("value-layer pipelines"); front-loaded in the README as of v2.8. See also values at action time. Bilingual first-use allowed: 「value layer（価値層）」.
• values at action time — ADR-0058 headline principle. Value layers (axioms, identity, skills, rules) belong where the agent acts or interprets fresh input, not where it extracts patterns from what already happened. Generalizes the ADR-0050/0051/0052 observation-without-steering line to the distillation lens. Bilingual first-use allowed: 「values at action time（行動時にこそ価値層を）」.
• axiom-free distillation — ADR-0057/0058. All distillation / extraction system prompts are base-only (the credential-leak guard); the four axioms are injected only at action time, never at distillation time, so the corpus is not re-read through a value lens.
• self-reflection corpus — ADR-0057. The body of self_reflection-view patterns from which distill-identity derives the persona alone — no prior-identity seed, no axiom injection.
• counterparty by author name — ADR-0055. The interaction pipeline keys the counterparty (target_agent) on author name, because live feed posts carry author.name but not author.id.
• content-verification handshake — ADR-0062. Moltbook requires an unverified agent to solve an obfuscated math challenge (via LLM reasoning) and POST it before its posts/comments become visible; memory recording is gated on verification. ADR-0063 scopes the NoveltyGate comparison to verified (visible) posts.
• read-only pattern-composition instruments — ADR-0071. Measure view supply, seed-independent diversity, and grounding composition before any behavioral intervention — instrument first, then act on what the instrument reads.
• observability by default — ADR-0075 headline principle. Any feature performing external I/O, an LLM call, or a non-deterministic decision ships a replayable append-only audit log in the same PR, not as a follow-up. Bilingual first-use allowed: 「observability by default（既定として可観測）」.
• shadow instrument — ADR-0076. A candidate decision mechanism run alongside production for observation only: it records what it would have decided and changes nothing, so whether to enforce it can be settled on measurement rather than intuition. Distinct from a read-only instrument (ADR-0071), which measures existing data rather than shadowing a decision. Bilingual first-use allowed: 「shadow instrument（影の計器）」.
• shadow constitution — ADR-0092. A constitution synthesized from the agent's stored constitutional patterns alone, with the live constitution text withheld from the model, compared to the live text by cosine and logged as a read-only reading before an amendment decision. A shadow instrument (ADR-0076) applied to the constitution rather than to a decision. Bilingual first-use allowed: 「shadow constitution（影の憲法）」.
• chaos-TDD / fault column — ADR-0077. A test-first discipline, not an infrastructure practice: a feature touching an LLM call or external I/O ships a set of deterministic fault-injection tests (its fault column) in the same PR, and those tests assert the desired guarded behavior before the guard exists. Faults are injected only at seams that already exist, so no production chaos hook is added. Bilingual first-use allowed: 「chaos-TDD（障害注入 TDD）」.
• permanent facade — ADR-0079. When a module becomes a package, the compatibility layer is either kept forever because the old import path is public API, or omitted entirely with its tests migrated in the same commit. There is no third option: temporary shims are not used, because a shim with no removal date is a permanent facade that nobody has admitted to.
• decision packet — ADR-0085. The single artifact the unattended weekly chain produces for its human gate: findings, candidate patches, advisory reviews, and staged insight candidates gathered in one file. The machine assembles it and stops; applying, approving, and adopting all happen on the other side of the gate. Bilingual first-use allowed: 「decision packet（承認パケット）」.
• post-distill durability gate — ADR-0084. A judgment placed after the artifact it judges: the gate receives an episode together with the patterns distill produced from it and decides per pattern which to keep. Placing the same question before distillation was measured to be answerable without evidence, so producing the pattern is the evidence requirement.
• conformance kit — ADR-0088. The LLMBackend contract checks shipped inside the wheel (contemplative_agent.testing) so sibling backend repositories import them rather than hand-copy them, and so the party that publishes a contract change — main, at release — can run them against siblings that go untouched for months. Bilingual first-use allowed: 「conformance kit（適合検査キット）」.
• behavioral eval layer — ADR-0089. The top-level evals/ layer that measures what the LLM component actually generates: production-parity runs over a golden dataset under pinned prompt-asset snapshots, an isolated LLM judge issuing named verdicts (binary checks as evidence, no score aggregation), and regressions read as per-case verdict transitions against an approved baseline. Deliberately outside both verify.sh and tests/.
• IPD two-arm bench — ADR-0090. The constitution-amendment instrument: the current and the staged constitution run as two arms of the same iterated-prisoner's-dilemma bench, and the reading joins the text diff at the human approval gate. A null-pair calibration — the same constitution in both arms, run twice — measures the bench's own noise floor before any between-arm difference is trusted. Bilingual first-use allowed: 「IPD two-arm bench（IPD 二腕ベンチ）」.

AAP four-quadrant lens (Keep original)

Quadrant names are direct AAP terminology and stay in English in all six languages, with a bilingual gloss allowed on first use (e.g. 「LLM Workflow（LLM ワークフロー象限）」). The lens is borrowed in this project as a usage-description aid, not as a category claim (ADR-0033).

• Script — deterministic, defined-input quadrant.
• Algorithmic Search — deterministic, exploratory-input quadrant.
• LLM Workflow — semantic-judgement, defined-input quadrant.
• Autonomous Agentic Loop — semantic-judgement, exploratory-input quadrant.
• Phase-crossing observation — work that originates in operation phase but revises design-phase artifacts (skills, rules, identity). Not a quadrant; an orthogonal observation. In-repo anchor: ADR-0016.
• quadrant lens — the borrowed routing diagnostic itself. Bilingual gloss "象限レンズ / 象限診断" allowed on first use in JA.

AKC phase → CLI mapping (English only)

• distill — Extract (behavioral / constitutional knowledge)
• distill-identity — Promote (whole-file identity)
• insight — Curate (extract skills from patterns)
• family promotion (gate step, ADR-0097 — decided, entry point reserved for slice 2) — Promote (co-selected skill family → one rule; replaced rules-distill)
• amend-constitution — Curate (constitutional knowledge → constitution)
• adopt-staged — Promote (staging → active)
• meditate — adapter-specific (meditation simulation)

CCAI axioms (English, paper-faithful)

The four axioms from Laukkonen et al. (2025) are cited verbatim in all languages so they remain linkable to the paper. A bilingual gloss is allowed on first use if the surrounding prose benefits from it.

• Emptiness — 空 / 空 / 空 / Vacuidade / Vacuidad
• Non-Duality — 非二元性 / 非二元 / 非二元 / Não-Dualidade / No-Dualidad
• Mindfulness — マインドフルネス / 正念 / 正念 / Atenção Plena / Atención Plena
• Boundless Care — 無限の慈愛 / 无量悲悯 / 無量悲憫 / Cuidado Ilimitado / Cuidado Ilimitado

Yogācāra / 唯識 terminology

Proper nouns stay in their romanized Sanskrit form; a local gloss is encouraged on first use.

• Yogācāra — 唯識 (yuishiki) / 唯识 / 唯識 / Yogācāra / Yogācāra
• eight-consciousness model — 八識モデル / 八识模型 / 八識模型 / modelo das oito consciências / modelo de las ocho conciencias
• bīja / 種子 — 種子 / 种子 / 種子 / semente (bīja) / semilla (bīja)
• vāsanā / 習気 — 習気 / 习气 / 習氣 / impressão (vāsanā) / impregnación (vāsanā)

Project name vs adapter name

• Contemplative Agent — the project's GitHub / official name (contemplative-agent). Use this in all user-facing prose, README sections, llms.txt, and translated documentation.
• Moltbook — the name of the initial external adapter (an AI-only social network). Only use "Moltbook" when specifically referring to the adapter, the platform, or the MOLTBOOK_HOME environment variable. Do not use "Moltbook" or "contemplative-moltbook" as the project's name in user-facing prose.
• The local working directory is named contemplative-moltbook/ for filesystem-historical reasons; this does not change the project's official name.

Agent host ecosystem (Keep original)

These are external project / product names. Keep them in English (or their canonical romanized form) across all languages.

• OpenClaw — open-source AI agent runtime. General-purpose host category (per ADR-0032). Loads ~/.openclaw/workspace/SOUL.md as agent personality at session start. Hosts a marketplace called ClawdHub for skills.
• OpenCode — agent harness with two-layer SOUL.md load (global ~/.config/opencode/SOUL.md + project .opencode/SOUL.md).
• Codex — agent harness category (CLI-aware host). Distinct from the deprecated OpenAI Codex CLI.
• Goose — Block / AAIF agent harness. Uses .goosehints (instruction file). Not a soul-folder host; do not list Goose alongside OpenClaw / OpenCode / Codex when describing soul-folder adoption.
• soul-folder — convention shared by OpenClaw / OpenCode / Codex where the host loads a SOUL.md file as agent personality at startup.
• SOUL.md — agent personality file in the soul-folder convention. First-person identity (axioms, refusals, voice, continuity). Contemplative Agent itself does not ship a SOUL.md; the contemplative-agent-rules sibling repository ships one with the four axioms verbatim.
• ClawdHub — OpenClaw's skill marketplace (referenced for context, not endorsed as a distribution channel for Contemplative Agent).

Schema fields and binary names

Always in English: source_type, valid_from, valid_until, category, MOLTBOOK_HOME, MOLTBOOK_API_KEY, contemplative-agent, contemplative-agent-cloud, qwen3.5:9b, gemma4:e4b, nomic-embed-text, knowledge.json, identity.md, logs/*.jsonl, noise-YYYY-MM-DD.jsonl, numpy, requests, Q4_K_M, embedding cosine, LLMBackend, AnthropicBackend, OpenAIBackend, CONTEMPLATIVE_CLOUD_PROVIDER, CONTEMPLATIVE_CLOUD_MODEL, ANTHROPIC_API_KEY, OPENAI_API_KEY, $MOLTBOOK_HOME/prompts/, $MOLTBOOK_HOME/views/, snapshots/.

Notes on choices

• views (ADR-0019) is left in English in all languages. In the codebase it is a schema-level term with a specific operational meaning (editable semantic seeds that project embedding coordinates onto named classes); translating it loses the connection to the ADR.
• embedding is kept as English loanword in ja (埋め込み) because readers routinely see the English term in ML literature. Local rendering listed so the translator can choose depending on audience. (zh-CN / zh-TW renderings 嵌入 are preserved as historical reference for the original translation discipline; the corresponding README mirrors were retired 2026-05-15.)
• bitemporal / air-gapped are kept in English when they appear as technical adjectives (security context). A local gloss may precede them on first use.
• The CCAI four axioms are treated like the AKC phase names: citable only in their English form because any translation risks diverging from the source paper (Laukkonen et al., 2025).

Maintenance

When updating README.md (English source), keep this table in sync if you introduce a new project-coined phrase or ADR slogan. Translators of language-specific README files should consult this table before choosing a rendering. Upstream AKC glossary changes should be mirrored into the "AKC-aligned terms" section.