GLOSSARY

PAI Glossary & Index

Purpose: Single lookup surface for every acronym, term, codename, and concept used across the PAI infrastructure. Each entry gives the expansion (for acronyms), a one-to-two-sentence definition, aliases, and the canonical source file the definition derives from.

Created: 2026-07-02 · Updated: 2026-07-02 Maintenance: This is a living document. New doctrine terms route here via the LEARN-phase Learning Router (doctrine/rule types); when you coin or encounter an undefined term, add it with a source path. Definitions were extracted verbatim-first from sources read on the creation date — if an entry and its source ever disagree, the source wins; fix the entry.

Scope: Infrastructure terms only. Individual skills are NOT enumerated (127 skill directories exist; the skill registry and SkillSearch serve that lookup) — only the skill system and a handful of infrastructure-grade skills appear here. Duane’s personal definitions live separately in PAI/USER/DEFINITIONS.md.

---

Acronym Quick Reference

Acronym	Expansion	See entry
AGT	Agent ID prefix (AGT-001…)	AGT IDs
AGY	Antigravity (Gemini CLI wrapper)	AGY
DA	Digital Assistant	DA
E1–E5	Effort tiers (Standard→Comprehensive)	Effort Tiers
ICM	Infinite Context Memory	ICM
ISA	Ideal State Artifact	ISA
ISC	Ideal State Criteria	ISC
MOC	Map of Content (knowledge index)	KnowledgeHarvester
ORA	Org Role Assignment	ORA
PAI	Personal AI Infrastructure	PAI
PNC	PAI Nova Claude (Claude Code engine)	PNC
PNG	PAI Nova Gemini (antigravity/agy engine)	PNG
PNK	PAI Nova OpenCode (opencode engine)	PNK
PNO	PAI Nova Ollama (local-model engine)	PNO
PNX	PAI Nova Codex (codex engine)	PNX
PRD	Product Requirements Document (ISA’s format lineage)	PRD
RBAC	Role-Based Access Control (two-layered)	RBAC
RTK	Rust Token Killer	RTK
USR	User ID prefix (USR-001 = Duane)	Principal

---

A

• Activated — Core Concept — Human 3.0 stage 2: “you are acting on the gap.” — PAI/DOCUMENTATION/PAISystemPhilosophy.md
• Actualized — Core Concept — Human 3.0 stage 4: “the system and the human operate as one toward the ideal state.” — PAI/DOCUMENTATION/PAISystemPhilosophy.md
• Advisor — Doctrine — The Rule 2 reviewer: bun PAI/Tools/Inference.ts --mode advisor --auto-state (Opus, 30-min timeout). Auto-state synthesis “closes the biggest RedTeam flaw where the caller could omit problem areas from what the reviewer sees.” Mandatory at the E2+ PLAN→BUILD boundary. — PAI/Algorithm/ (LATEST)
• Agent (subagent) — Component — “AI personas spawned inside Claude Code using the Task tool. They run in the same Claude Code runtime but with specialized instructions.” Distinct from Engines (separate runtimes). Definitions live in agents/*.md (Architect, Engineer, Explore, Intern, RealityChecker, researchers, etc.). — PAI/ARCHITECTURE_RELATIONSHIPS.md
• Agent Systems (three) — Component — PAI has three: Task-tool subagent types (internal workflow), Named Agents (persistent identities with ElevenLabs voices), and Custom Agents (composed via ComposeAgent from traits). “Confusing them causes routing failures.” — PAI/PAIAGENTSYSTEM.md
• AgentLogger — Tool — “Appends agent activity to the centralized agent-activity.jsonl log. Required by CLAUDE.md for tracking significant tasks and phase shifts”; feeds the dashboard agent view. — PAI/Tools/AgentLogger.ts
• AGT IDs — Codename — Stable agent identifiers: AGT-001 PNC/Claude Code, AGT-002 AGY/Antigravity, AGT-003 opai/OpenCode, AGT-004 Switchboard-Broker, AGT-005 Antigravity IDE, AGT-007 PNO/Ollama, AGT-008 PNX/Codex. — PAI/ARCHITECTURE_RELATIONSHIPS.md, PAI/ASSET_REGISTRY.md
• AGY — Codename — Antigravity (AGT-002): “CLI wrapper for Google Gemini models | Web research, factual lookup, parallel search.” CLI command agy; replaced Gemini CLI 2026-05-25; a shim maps legacy --yolo to --dangerously-skip-permissions. — PAI/ARCHITECTURE_RELATIONSHIPS.md
• The Algorithm — Doctrine — “The 7-phase execution engine: Observe, Think, Plan, Build, Execute, Verify, Learn. Transitions from CURRENT STATE to IDEAL STATE via verifiable criteria (ISC).” Current version string is read from PAI/Algorithm/LATEST (v5.7.11 at writing). Only the primary DA may run ALGORITHM mode. — PAI/README.md, PAI/Algorithm/
• ALGORITHM mode — Convention — One of three output modes: “everything else. Including any build/create/make/implement/design/refactor/migrate/integrate request.” See Modes. — PAI/PAI_SYSTEM_PROMPT.md
• Aligned — Core Concept — Human 3.0 stage 3: “your daily work and your Telos point the same direction.” — PAI/DOCUMENTATION/PAISystemPhilosophy.md
• Anti-criterion — Doctrine — ISC prose-prefix kind: ISC-N: Anti: <what must NOT happen>. “≥1 required — a goal with zero failure modes worth naming is under-specified.” — PAI/Algorithm/ (LATEST)
• Antecedent — Doctrine — ISC prose-prefix kind: a “precondition that reliably produces the target experience (novel juxtaposition, elegance in constraint…)”; ≥1 required when the goal is experiential. — PAI/Algorithm/ (LATEST)
• Anvil — Codename — “Kimi K2.6 via Moonshot, 256K context — invoke at E3/E4/E5 when whole-project context materially affects correctness (cross-file refactors, architecture-fitting changes, long-range reasoning).” — PAI/Algorithm/ (LATEST)
• aorus — Infra — “Physical machine hosting pai-primary; GPU compute (RTX 3060 6GB + RTX 4060 8GB).” Runs Proxmox; shares its NIC with pai-primary (“NOT separate compute nodes”). — PAI/ASSET_REGISTRY.md
• Archetype — Org — Authority archetypes are the permission primitives of Org RBAC: “the archetype decides the permission shape; the title remains display/context.” The nine: Principal, Coordinator, Producer, Verifier, Operator, Security, Archivist, Broker, Caregiver. — PAI/RBAC_ORG_STRUCTURES_MODEL.md
• Archivist — Org — Archetype (aka Librarian, Scribe, Memory Curator): “Preserve knowledge, summaries, decisions, audit trails. Cannot expand access or mutate source facts.” — PAI/RBAC_ORG_STRUCTURES_MODEL.md
• ASPIRATIONAL — Doctrine — Enforcement Map status: “only prose; no deterministic enforcement. Surfaced here so it can be promoted to ENFORCED or honestly demoted.” — PAI/ENFORCEMENT_MAP.md
• ASSET_REGISTRY.md — Convention — Auto-loaded registry of “hosts, services, ports, paths”; auto-catalogued via SessionEnd hook. First stop for infra facts (paired with pai-infra.ts resolve). — PAI/ASSET_REGISTRY.md, PAI/CONTEXT_ROUTING.md
• Async Primitive Gate — Doctrine — PLAN gate: “One-shot command → Bash(run_in_background). Event stream → Monitor. AI work → Agent(run_in_background). Never poll in a sleep loop.” — PAI/Algorithm/ (LATEST)
• Authentik — Infra — SSO proxy on port 9000; “canonical host: m710q.” — PAI/ASSET_REGISTRY.md
• Auto-memory — Memory — The harness-level persistent memory at ~/MEMORY/auto/ with the MEMORY.md one-line index. Read-side unified via recall(); PAI doctrine forbids routing RULES/preferences here (they go to CLAUDE.md/hooks/skills — see Self-Healing Infrastructure). — PAI/Algorithm/ (LATEST, Learning Router)
• Aware — Core Concept — Human 3.0 stage 1: “you can see your current state and your ideal state.” — PAI/DOCUMENTATION/PAISystemPhilosophy.md

B

• Backend Falsification Checkpoint — Doctrine — BUILD checkpoint: before building a pipeline that writes to a stateful/external backend, “run a throwaway write+read-back on a disposable key/name FIRST” to confirm the write method persists. — PAI/Algorithm/ (LATEST)
• Binding Commitment — Doctrine — “Selecting a capability = binding commitment to invoke it via tool… text-only is dishonest and counts as a CRITICAL FAILURE.” (aka “no phantom capabilities”) — PAI/Algorithm/ (LATEST)
• Broker — Org — Archetype (aka Connector, Hub, Delegate, Advocate): “Route messages, introductions, requests, and opportunities. Cannot bind others to obligations.” — PAI/RBAC_ORG_STRUCTURES_MODEL.md
• BUILD — Doctrine — Algorithm phase 4/7: preparation work; invoke selected capabilities; write non-obvious decisions to the ISA. Hosts the Root-Cause-at-Ingestion and Backend Falsification checkpoints. — PAI/Algorithm/ (LATEST)

C

• Capability — Doctrine — A skill or agent selected during OBSERVE for the run (“Select what the task genuinely needs within the tier time budget”). Catalogued in PAI/Algorithm/capabilities.md. — PAI/Algorithm/ (LATEST)
• Capability Routing — Convention — “Design/UI/UX/art/frontend work is capability routing, not a new authority archetype: engage Producer for creation/implementation and Verifier for UX review.” — CLAUDE.md
• Caregiver — Org — Archetype (aka Parent, Counselor, Calendar Steward): “Track care tasks, needs, safety checks. No secret/infra/destructive rights by care title.” — PAI/RBAC_ORG_STRUCTURES_MODEL.md
• Cato — Codename — The Rule 2a cross-vendor auditor: “runs GPT-5.4 via the codex exec CLI — different vendor… different constitutional training”; mandatory at E4/E5 after the Advisor, read-only sandbox, findings appended to MEMORY/VERIFICATION/cato-findings.jsonl. — PAI/Algorithm/ (LATEST), PAI/Tools/CrossVendorAudit.ts
• CheckpointPerISC — Hook — “Auto-commit on ISC [ ]→[x] transitions… Stages ONLY the ISA file path (never git add -A). Commits with subject ISC-{N} ({slug}): {desc}”. Rollback preview via PAI/Tools/Checkpoint.ts (prints the git reset — never executes it). — hooks/CheckpointPerISC.hook.ts
• ClaimAttributionScan — Hook — “WARN-ONLY telemetry scanner… scans the final assistant message for high-confidence success-claims and cross-checks the session transcript’s tool-call log for a corroborating call.” Honesty enforcement layer 3 (partial). — hooks/ClaimAttributionScan.hook.ts
• CLAUDE.md — Component — “The master config — generated from CLAUDE.md.template via BuildCLAUDE.ts. It defines execution modes, The Algorithm, and the context routing table.” Second in the Context Hierarchy after PAI_SYSTEM_PROMPT.md. — PAI/README.md
• CLI-First Architecture — Convention — “Applies to all new PAI tools, skills, and systems. Philosophy: deterministic code execution > ad-hoc prompting.” — PAI/CLIFIRSTARCHITECTURE.md
• ContainmentGuard — Hook/Security — “PreToolUse Edit/Write/MultiEdit gate. Blocks writes that would leak sensitive identity/infra strings into files outside the Z1–Z4 containment zones used by ShadowRelease” — i.e. it blocks hardcoded user-home paths from entering PAI files. — hooks/ContainmentGuard.hook.ts
• Context Hierarchy — Doctrine — “This system prompt defines behavioral non-negotiables: it is the highest authority layer. CLAUDE.md defines operational procedures and format templates.” loadAtStartup files come third. — PAI/PAI_SYSTEM_PROMPT.md
• Context Routing — Convention — The on-demand context table: “Load context on-demand by reading the file at the path listed. Only load what the current task requires.” — PAI/CONTEXT_ROUTING.md
• Coordinator — Org — Archetype (aka Guardian, Moderator, Incident Commander): “Assign, route, close session decisions, request approvals. Cannot exceed identity tier or bypass gates.” — PAI/RBAC_ORG_STRUCTURES_MODEL.md
• Coordination Gate — Doctrine — PLAN gate choosing among the three agent systems: “1. Agent Teams (DEFAULT for parallel work)… 2. Custom Agents (ONLY when Duane says ‘custom agents’). 3. Managed Agents (for unattended/overnight work).” — PAI/Algorithm/ (LATEST)
• Current State → Ideal State — Core Concept — The universal mechanism: “every task, from shipping code to making art, is a transition from current state to ideal state, pursued through the Algorithm.” — PAI/DOCUMENTATION/PAISystemPhilosophy.md
• Custom Agents — Component — “Dynamic agents composed via ComposeAgent from traits — when user says ‘custom agents’” (voice trait-mapped). — PAI/PAIAGENTSYSTEM.md

D

• DA — Core Concept — Digital Assistant: “the primary interface to the Life Operating System. When you talk to PAI, you’re talking to PAI Nova.” Every PAI user names their own DA; Nova is Duane’s. Canonical name source: settings.json → daidentity.name. — PAI/ARCHITECTURE_RELATIONSHIPS.md, settings.json
• DEFERRED-VERIFY — Doctrine — ISC status marker: “passed in code/intent but live probe is impossible at execution time… Requires a follow-up task ID. Cannot be marked [x] until the deferred probe runs.” — PAI/Algorithm/ (LATEST)
• Delegation Gate — Doctrine — Before spawning any agent: “Can I do this with Glob + Grep in under 30 seconds? YES → do it directly. NEVER delegate directed lookups… Foreground agent blocking >2 minutes = execution failure.” — PAI/Algorithm/ (LATEST)
• Deliverable Manifest — Doctrine — PLAN block enumerating “every sub-task the user explicitly asked for… Multi-part requests are the highest-risk failure vector.” Mandatory at any tier with 2+ explicit sub-tasks; checked at VERIFY as Deliverable Compliance (any ✗ blocks phase: complete). — PAI/Algorithm/ (LATEST)
• DestructiveOpGuard — Hook/Security — “PreToolUse:Bash gate. Loads destructive-patterns.json (the LEARNED-PATTERN REGISTRY) and blocks Bash commands that match a learned destructive pattern.” Overrides are pattern_slug-keyed (once/session/permanent). Patterns arrive via the promotion pipeline (memo → extractor → validator → dual-factor approval). — hooks/DestructiveOpGuard.hook.ts, PAI/DOCUMENTATION/SECURITY_OVERRIDE_MECHANISM.md
• Doc Sync — Doctrine — LEARN step: “if this session modified PAI system files, propagate changes to dependent docs” via the DocumentationUpdate workflow + DocCheck.ts --changed. — PAI/Algorithm/ (LATEST)

E

• Effort Tiers (E1–E5) — Doctrine — Standard E1 (<90s, no ISC floor, DEFAULT) · Extended E2 (<3min, ≥16 ISC) · Advanced E3 (<10min, ≥32) · Deep E4 (<30min, ≥128) · Comprehensive E5 (<120min+, ≥256). “The time budget is the hard constraint set by tier; the ISC floor (E2+) is a soft minimum on the count axis only.” Explicit override tokens /e1–/e5. — PAI/Algorithm/ (LATEST)
• Effective Permission — Org — “Effective access is the intersection of both layers [Identity RBAC ∩ Org-Scoped RBAC], never their union.” — PAI/RBAC_ORG_STRUCTURES_MODEL.md
• EgressClassGuard — Hook/Security — PreToolUse:Bash destination-aware egress watch: warns when a payload’s data class (CONFIDENTIAL via secret patterns / PRIVATE via private-tree sources) exceeds the route’s ceiling (external = PUBLIC only; internal allows PRIVATE). Warn-only ramp; block mode gated behind egress-guard.json + consent after FP soak. Adopted 2026-07-05 from the LifeOS 6.0 delta-mine (ADOPT #1). — hooks/EgressClassGuard.hook.ts
• ENFORCED — Doctrine — Enforcement Map status: “a hook, shim, or other deterministic mechanism blocks or transforms the violating behavior.” — PAI/ENFORCEMENT_MAP.md
• Enforcement Map — Component — “Every constitutional rule’s coverage is tracked in PAI/ENFORCEMENT_MAP.md — which engines enforce it, which mechanism, current status (ENFORCED / GAP / ASPIRATIONAL).” The source of truth for whether a doctrine layer is live. — PAI/PAI_SYSTEM_PROMPT.md
• Engine — Component — “Separate AI runtime environments that can execute work… Engines are separate runtimes, not subagents.” The roster codes: PNC, PNK, PNX, PNG, PNO. — PAI/ARCHITECTURE_RELATIONSHIPS.md, PAI/ENFORCEMENT_MAP.md
• EntityGraph — Memory — Entity-proximity graph used to boost /recall scoring over the ICM corpus (“FTS5 + entity graph scoring”). — PAI/Tools/EntityGraph.ts
• Euphoric Surprise — Core Concept — The experiential metric: “what you feel when a hard-to-vary explanation meets novelty — an answer that clicks in a way you couldn’t have predicted but instantly recognize as true.” Predicted and scored 1–10 at THINK (E2+); “if you cannot name an insight, predict ≤6.” — PAI/DOCUMENTATION/PAISystemPhilosophy.md
• events.jsonl — Memory — The Unified Event Log: “append-only JSONL file where hooks emit structured, typed events… type field uses a dot-separated topic hierarchy (e.g., algorithm.phase, work.created).” — PAI/MEMORYSYSTEM.md
• EXECUTE — Doctrine — Algorithm phase 5/7: do the work; “as each criterion passes, IMMEDIATELY edit ISA: [ ] → [x], update progress:” under the Inline Verification Mandate. — PAI/Algorithm/ (LATEST)

F

• Fabric — Skill — “Intelligent prompt pattern system with 240+ specialized patterns for content analysis, extraction, and transformation.” Reference doc: PAI/THEFABRICSYSTEM.md. — skills/Fabric/SKILL.md
• Fabrication — Doctrine — “Generating a confident assertion that is not grounded in a source verified this session.” The rule: “retrieve from a verified source, or do not assert.” Enforced by the honesty layers (IdentityPin, IdentityValidator, ClaimAttributionScan, shims). — PAI/PAI_SYSTEM_PROMPT.md
• fabricated-path — Doctrine — HonestyValidator finding kind targeting “backtick-absolute and ‘at/in/file /path’ forms” — a path claimed in output that no tool call touched. Currently warn-only. — PAI/ENFORCEMENT_MAP.md
• Fast-path mode — Doctrine — “Available at Standard tier (E1) only. Compresses phases for simple tasks… a whitelist, not a heuristic. Compress to: OBSERVE → EXECUTE → VERIFY.” — PAI/Algorithm/mode-detection.md
• Feedback Memory — Memory — feedback_*.md files (auto-memory) holding corrections/confirmed approaches. Consulted twice by doctrine: design-time at OBSERVE Step 0(3), pre-action at the PLAN Feedback Memory Auto-Consult. — PAI/Algorithm/ (LATEST)
• Forge — Codename — “GPT-5.4 via codex exec, reasoning_effort=high — auto-include at E3/E4/E5 for any coding task (implement, refactor, debug, build, migration, fix, feature).” — PAI/Algorithm/ (LATEST)
• Functions API — Component — “HTTP gateway on localhost:8890. No Claude API tokens required. Authenticated calls should use a signed PAI identity session.” Systemd unit pai-functions; also the JIT credential gateway. — PAI/FUNCTIONS_API.md

G

• GAP — Doctrine — Enforcement Map status: “a mechanism exists for one or more engines but not the full set named in the rule’s scope.” — PAI/ENFORCEMENT_MAP.md
• Granularity Test — Doctrine — “Split until each criterion is one binary tool probe… If you cannot name the probe, the criterion is not yet atomic — split it.” The pre-THINK exit condition for ISC quality. — PAI/Algorithm/ (LATEST)

H

• Hard-to-Vary Explanation — Core Concept — Deutsch epistemology at PAI’s root: “a description of reality (or of a goal) where every detail plays a functional role, so contrary evidence has nowhere to flee.” — PAI/DOCUMENTATION/PAISystemPhilosophy.md
• Herdr — Codename/Tool — Multi-engine agent/pane/workspace manager; PAI drives it via HerdrCockpit.ts (hc): “status / launch / handoff / council (fan-out a prompt to N engines in exec mode)” over a direct Unix socket. — PAI/Tools/HerdrCockpit.ts
• Honesty Floor — Doctrine — “Three permitted forms for any technical or identity claim: Sourced… Hedged… Unknown — ‘I don’t know’ is a valid answer. It always beats a fabrication.” Overridden for safety-asserting claims by Security Escalation. — PAI/PAI_SYSTEM_PROMPT.md
• HonestyValidator — Tool — “Single source of truth for honesty/no-fabrication checks… validates that an output emitted by ANY PAI engine does not fabricate the DA’s identity and does not contain confident unsourced claims.” Called by IdentityValidator (block-mode for identity-mismatch) and ClaimAttributionScan (warn). Kill switch: PAI_HONESTY_BYPASS=1 (logged). — PAI/Tools/HonestyValidator.ts
• Hook — Component — “TypeScript/Python scripts that run automatically when specific events occur in Claude Code sessions” (SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, Stop, SubagentStop, PreCompact…). Registered in settings.json; live in hooks/. — PAI/ARCHITECTURE_RELATIONSHIPS.md, PAI/THEHOOKSYSTEM.md
• HookHealer — Hook — SessionStart self-healing sweep: parses settings-registered hook commands and repairs missing exec bits on direct-exec scripts under ~/.claude (realpath containment; warn-only, fail-open). Adopted 2026-07-05 from the LifeOS 6.0 delta-mine (ADOPT #2). — hooks/HookHealer.hook.ts
• Human 3.0 — Core Concept — “The maturity arc PAI is built to move the Principal along”: Aware → Activated → Aligned → Actualized. — PAI/DOCUMENTATION/PAISystemPhilosophy.md

I

• ICM — Memory — Infinite Context Memory: MCP server for long-term memory across sessions (recall/store with topic + importance levels; memoirs; feedback). Surfaced by the /recall skill with EntityGraph boost. — icm MCP server, commands/recall.md
• Ideate mode — Doctrine — Algorithm mode (ideate/id8 triggers): idea-generation loop configured by ideate-loop.md; sets mode: ideate in ISA frontmatter. — PAI/Algorithm/mode-detection.md
• Identity Tokens — Doctrine — “Your own name, the principal’s name, the names of system components — must be retrieved from their canonical source, not generated from prior.” Canonical source: settings.json → daidentity.name / principal.name. — PAI/PAI_SYSTEM_PROMPT.md
• identity-mismatch — Doctrine — The HonestyValidator finding kind in BLOCK mode: the Stop-hook emits {decision: block} JSON forcing regeneration when output uses a wrong DA/principal name. — PAI/ENFORCEMENT_MAP.md
• IdentityPin — Hook — “Emits an explicit identity-pin block at every session start… tells the model: ‘your name is X; do not synthesize it.’” Honesty enforcement layer 1. — hooks/IdentityPin.hook.ts
• IdentityValidator — Hook — “Catches DA-identity fabrication at Stop/SubagentStop… scans the response text for fabricated DA-identity tokens — names that don’t match settings.json → daidentity.name.” Honesty enforcement layer 2 (block-mode for identity-mismatch). — hooks/IdentityValidator.hook.ts
• Inference.ts — Tool — “Unified inference tool with capability-matrix routing… via inference-matrix.json, task-based model selection, cost tracking integration.” The dispatcher behind OllamaSkill task profiles and --mode advisor. — PAI/Tools/Inference.ts
• Inline Verification Mandate — Doctrine — “No ISC criterion may transition [ ] → [x] without verification evidence captured in the same tool call block that claims it, or the immediately-following block.” Forbidden language in place of evidence: “should work”, “done” without tool evidence, “no errors” without the log. — PAI/Algorithm/ (LATEST)
• Intent Echo — Doctrine — Mandatory first OBSERVE action: “restate the user’s request in ONE sentence… This line anchors the entire Algorithm run.” — PAI/Algorithm/ (LATEST)
• Interceptor — Skill/Component — “The ONLY sanctioned browser automation in PAI — real Chrome, no CDP detection, real login sessions, accurate rendering.” Mandatory for web-output verification and deploy confirmation. — PAI/PAI_SYSTEM_PROMPT.md, skills/Interceptor/SKILL.md
• ISA — Core Concept — Ideal State Artifact: “the single document that articulates, drives, and verifies a thing’s ideal state as a hard-to-vary explanation.” Lives at MEMORY/WORK/{slug}/ISA.md; system of record for Algorithm runs (“The AI writes ALL content directly. Hooks only read.”). Frontmatter: task/slug/effort/phase/progress/mode/started/updated. — skills/ISA/SKILL.md, PAI/Algorithm/ (LATEST)
• ISC — Core Concept — Ideal State Criteria: “the irreducible, independently verifiable structure of ‘done.’ A vague checklist is not an ISC.” Written as - [ ] ISC-N: criterion, one binary tool probe each. — PAI/DOCUMENTATION/PAISystemPhilosophy.md
• ISASync — Hook — “Read-only ISA → work.json sync via PostToolUse… ONLY reads the ISA and syncs to work.json for the dashboard.” The ISA frontmatter phase: edit IS the phase signal. — hooks/ISASync.hook.ts
• Isolation Gate — Doctrine — PLAN gate for parallel write-agents: “Overlapping file targets → isolation: ‘worktree’. Non-overlapping → skip. Read-only agents → never need worktree. Competing approaches → always worktree.” — PAI/Algorithm/ (LATEST)

K

• Kai incident — Doctrine — The 2026-05-26 originating identity-fabrication failure (DA name “Kai” emitted from prior). Named failure modes: salience gap, plausibility hijack, template pressure, missing structural guard — the reason the honesty enforcement layers exist. — PAI/PAI_SYSTEM_PROMPT.md
• KnowledgeHarvester — Tool — “Knowledge graph maintenance tool… for each domain in PAI/MEMORY/KNOWLEDGE/{Ideas,People,Companies,Research}, generate _MOC.md” (Map of Content indexes). — PAI/Tools/KnowledgeHarvester.ts

L

• LEARN — Doctrine — Algorithm phase 7/7: learning inventory + Learning Router, doc sync, reflection JSONL, then the SUMMARY block. — PAI/Algorithm/ (LATEST)
• Learning Router — Doctrine — “Every ‘should I remember this?’ question goes through this single router” — routes each learning to its PAI surface (knowledge/rule/gotcha/state/business/identity/doctrine/hook/permission/reflection). “Default disposition: SKIP.” — PAI/Algorithm/ (LATEST)
• Legion — Infra/Codename — The laptop (pai3); hosts the Legion Voicebox Sound Server (port 17493, POST /speak) — the active Pulse voice backend — and the migrated Plane instance (:8091). — PAI/ASSET_REGISTRY.md
• Life OS — Core Concept — See PAI. “A Life Operating System — not just a chatbot, but a system that helps you run your life with persistent memory, delegation, coordination, and verification.” — PAI/ARCHITECTURE_RELATIONSHIPS.md
• loadAtStartup — Convention — settings.json list of “files force-loaded into session context at startup by LoadContext.hook.ts… injected as system-reminder blocks.” Paired with postCompactRestore.fullFiles for compaction survival. — settings.json
• LoopDetector — Hook — PostToolUse all-tools watcher: per-session sliding window of tool-call signatures; warns on 4 identical calls in 12 or 3 consecutive identical failures (cooldown-limited, warn-only, fail-open). Adopted 2026-07-05 from the LifeOS 6.0 delta-mine (ADOPT #3). — hooks/LoopDetector.hook.ts
• LEARNING/ (REFLECTIONS · SIGNALS · FAILURES · SYNTHESIS) — Memory — The learning surfaces: algorithm-reflections.jsonl (per-run LEARN reflections), ratings.jsonl (satisfaction signals), full-context failure dumps for low-sentiment events, and synthesized patterns. — PAI/MEMORYSYSTEM.md

M

• MEMORY/ — Memory — The unified memory tree (Memory System v7.0, “Projects-native architecture”): WORK/ (per-task ISA/PRD dirs — primary work tracking), KNOWLEDGE/ (People/Companies/Ideas/Research with typed cross-links), STATE/ (ephemeral operational state, “optimized for speed, not permanence”), LEARNING/ (see above). Claude Code’s projects/ transcripts are “the actual firehose.” — PAI/MEMORYSYSTEM.md
• Mercury — Codename — “Mercury 2 fill-in-the-middle code completion via Inception Labs API” (MercuryFIM skill). — skills/MercuryFIM/SKILL.md
• MetaMCP — Infra — MCP proxy on port 3010 (“8 MCP servers proxied”). — PAI/ASSET_REGISTRY.md
• MINIMAL mode — Convention — Output mode for “greetings, ratings, single-token acknowledgments.” — PAI/PAI_SYSTEM_PROMPT.md
• Modes (MINIMAL / NATIVE / ALGORITHM) — Convention — The three mandatory output formats; “decided by a deterministic classifier at UserPromptSubmit” (PromptProcessing) writing a MODE:/TIER: line. Every response uses exactly one. — PAI/PAI_SYSTEM_PROMPT.md, CLAUDE.md

N

• Named Agents — Component — “Persistent identities with backstories and ElevenLabs voices (Serena, Marcus, Rook, etc.) — recurring work, voice output, relationships.” — PAI/PAIAGENTSYSTEM.md
• NATIVE mode — Convention — Output mode for “a single fact lookup, a single-line edit on a named file, or one command run — AND no new artifact is created — AND no multi-step plan is required.” All subagents use NATIVE. — PAI/PAI_SYSTEM_PROMPT.md
• Nova — Codename — Duane’s DA (daidentity.name: Nova, aka PAI Nova; the identity behind AGT-001/PNC). — settings.json
• /notify (port 31337) — Component — pai-pulse: “canonical engine-facing voice compatibility route: /notify; forwards to Legion Voicebox.” The curl target for Algorithm voice announcements. — PAI/ASSET_REGISTRY.md

O

• OBSERVE — Doctrine — Algorithm phase 1/7: Intent Echo, Reverse Engineering, Preflight Gates A–F, Step 0 batch/parallelize + prior-art consult, Reproduce-First, effort setting, capability selection, ISC writing. — PAI/Algorithm/ (LATEST)
• OBSERVE Step 0 — Doctrine — “Before deep reads, in ONE round: enumerate every file the task will likely touch and Read them together; launch every independent background job up front… (3) consult prior art and probe the grammar — in OBSERVE, not at PLAN/THINK/BUILD.” — PAI/Algorithm/ (LATEST)
• OllamaSkill — Skill — “Local-first inference dispatch via PAI/Tools/Inference.ts — routes to Ollama for cheap task types… with automatic cloud fallback per inference-matrix.json.” Task profiles: fast_classification (llama3.2:3b), structured_json (qwen2.5:7b), summarization (gemma2:9b), multi_step_reasoning (deepseek-r1:7b), code_generation (honestly routed to claude/sonnet). — skills/OllamaSkill/SKILL.md
• OmniPulse — Codename — Cross-engine dispatch/handoff layer; handoff artifacts under ~/MEMORY/STATE.claude/omnipulse-handoffs/. Its honesty-validation role (layer 5) is NOT YET LIVE. — PAI/PAI_SYSTEM_PROMPT.md (layer table)
• opai — Codename — “OpenCode (opai) | AGT-003 | Separate CLI tool using Zen free models | Zero-cost coding tasks, budget-critical periods.” The engine PNK runs. — PAI/ARCHITECTURE_RELATIONSHIPS.md
• Operator — Org — Archetype (aka SysAdmin, DBA, Release Manager): “Execute runbook-scoped operations on assigned assets/tickets. Destructive or undocumented actions require approval.” — PAI/RBAC_ORG_STRUCTURES_MODEL.md
• Optimize mode — Doctrine — Algorithm mode (optimize [target]): measured improvement loop per optimize-loop.md (eval_mode metric vs eval). — PAI/Algorithm/mode-detection.md
• ORA — Org — Org Role Assignment: “Binds one UID to one org role in one org session” (ORA-{id}). Permissions require an active scoped ORA plus RBAC allow. — PAI/RBAC_ORG_STRUCTURES_MODEL.md

P

• PAI — Core Concept — “Personal AI Infrastructure = the Life Operating System. It turns AI from a chatbot you talk to into a system that helps you run your life.” Primary directive: understand the Principal to move them from current state to ideal state. — PAI/DOCUMENTATION/PAISystemPhilosophy.md
• pai-core.ts — Tool — “Phase-1 substrate kernel. The ONLY module allowed to know real PAI paths… resolves the PAI CONTENT root (Option B) explicitly and forward-compatibly.” Use memoryDir() etc., never raw env. — PAI/Tools/lib/pai-core.ts
• pai-guard.ts shims — Security — PATH shims at PAI/bin/<tool> enforcing the destructive-pattern registry outside Claude Code; “shims fail-open on registry errors so the user’s shell is never broken.” Shell-tier override key: shell:<ppid>:<tty>. — PAI/DOCUMENTATION/SECURITY_OVERRIDE_MECHANISM.md
• pai-infra.ts — Tool — “Single source of truth for KNOWN infrastructure/environment… nothing about known infra may be RECALLED from memory. It is RESOLVED from this registry… it never guesses.” The tool behind “resolve, don’t recall.” — PAI/Tools/pai-infra.ts
• pai-primary — Infra — “Primary AI/service hub (VM on aorus physical)” — the main PAI host (Proxmox VM100). — PAI/ASSET_REGISTRY.md
• pai-prism — Infra — “Kitchen Prism family kiosk (Linux appliance)… Chrome —kiosk via Xorg autologin.” The family-facing display surface (calendar, dashboards, 3D home twin). — PAI/ASSET_REGISTRY.md
• PAI_HONESTY_BYPASS — Security — Env kill switch that “disables the [honesty] validator with a logged warning. Intended for hook-debugging only”; uses are logged to honesty-bypass.jsonl. — PAI/PAI_SYSTEM_PROMPT.md
• Persona — Component — “Rich narrative identities for agents — who they are, how they think, their voice, their style.” Layered on archetypes/agents; not an authority grant. — PAI/ARCHITECTURE_RELATIONSHIPS.md
• PhaseTransitionGate — Hook — “Structural enforcement of Algorithm Rule 2. Blocks ISA edits that transition phase: to build unless the ISA Decisions section contains an advisor marker or a complete Rule 2 emission block.” Fires at every PLAN→BUILD regardless of tier. — hooks/PhaseTransitionGate.hook.ts
• PLAN — Doctrine — Algorithm phase 3/7: feedback auto-consult, planning block, Deliverable Manifest, delegation/parallelism/async/watchdog/isolation/coordination/quota gates, then the PLAN→BUILD Advisor Gate. — PAI/Algorithm/ (LATEST)
• PNC / PNK / PNX / PNG / PNO — Codename — The engine legend: “PNC — Claude Code (Anthropic, primary)… PNK — opencode… PNX — codex… PNG — antigravity / agy (Google)… PNO — Ollama local models (via PAI/Tools/Inference.ts).” Colloquially “PNK is Codex” means “a code agent” — PNK runs opencode, PNX runs codex. — PAI/ENFORCEMENT_MAP.md
• PRD — Memory — Product Requirements Document — the ISA’s format lineage: “the single source of truth for every Algorithm run” (PRDFORMAT.md v2.0 still defines the ISA file spec; Algorithm renamed the artifact ISA). — PAI/PRDFORMAT.md
• Preflight Gates (A–F) — Doctrine — OBSERVE gates fired by task shape: A Diagnostic (reproduce before code archaeology; probe the authoritative dynamic source before any infra-attribution claim), B Deploy/API (credentials/CLI/service access exist), C External service (load skill context + gotchas), D Research (external docs before local archaeology), E Existence Probe (Glob/Grep for prior art or confirmed-greenfield), F Prerequisite Probe (empirically probe writability/tool-on-PATH/host-reachable/port-conflict before committing a write/start/call ISC). — PAI/Algorithm/ (LATEST)
• Principal — Org/Core Concept — The human PAI serves (Duane, USR-001): “only principal accepts risk, spend, destructive ops, and authorization expansions.” Canonical name: settings.json → principal.name. — PAI/RBAC_ORG_STRUCTURES_MODEL.md, settings.json
• Producer — Org — Archetype (aka Engineer, Volunteer, SRE, Toolsmith): “Update owned tasks and write work evidence. Cannot approve own work.” — PAI/RBAC_ORG_STRUCTURES_MODEL.md
• PromptProcessing — Hook — The UserPromptSubmit dispatcher: deterministic MODE/TIER classifier (“regex/heuristic rules… instant, no model call”) plus tab-title/session-name inference. Its MODE:/TIER: line is authoritative; absence = hook failure to flag. — hooks/PromptProcessing.hook.ts, PAI/PAI_SYSTEM_PROMPT.md
• Proxmox (prox) — Infra — The hypervisor on aorus (PVE 9.x, ssh prox); hosts pai-primary (VM100) and the HAOS VM; nightly vzdump backups. — PAI/ASSET_REGISTRY.md
• Pulse — Component — “The Life Dashboard: the visible surface onto the Life OS (http://localhost:31338).” Also the name of the voice-notify service lineage (pai-pulse, port 31337). — PAI/DOCUMENTATION/PAISystemPhilosophy.md

Q

• QNAP TS-464 (qnap-nas / NAS6DB92A) — Infra — “Primary NAS — NFS /share/PAI → /PAI + /mnt/pai on pai-primary” (transport now SSH-only per the storage spine; off tailnet 2026-06-18). Hosts Shared-Inbox handoffs and backups. — PAI/ASSET_REGISTRY.md
• Quota-Check Gate — Doctrine — “Before spawning ≥2 parallel agents OR any Agent call with model:‘opus’, probe subscription headroom” via UsageProbe; <20% headroom → downshift to local/single-agent; <10% → defer or route local. — PAI/Algorithm/ (LATEST)

R

• RBAC (two-layered) — Org — “1. Identity RBAC: persistent trust tier and baseline capabilities for a UID. 2. Org-Scoped RBAC: temporary role grants inside a structure/team/session/resource boundary.” Effective access = intersection. — PAI/RBAC_ORG_STRUCTURES_MODEL.md
• recall / /recall — Memory — “Hybrid ICM memory recall with EntityGraph proximity boost. Searches ICM memories using FTS5 + entity graph scoring.” Read-side unification across ICM + auto-memory corpora. — commands/recall.md
• refined: — Doctrine — The ISA ## Decisions prefix logging living-document refinements; counted as living_doc_refinements in the reflection JSONL. — PAI/Algorithm/ (LATEST)
• Reflection JSONL — Doctrine — The per-run LEARN record appended to MEMORY/LEARNING/REFLECTIONS/algorithm-reflections.jsonl (tier, ISC counts, doctrine_fired flags, satisfaction prediction). Mined by /pai-upgrade mine-reflections to evolve doctrine. — PAI/Algorithm/ (LATEST)
• Re-Read Check — Doctrine — “Final gate before LEARN… re-read the user’s last message verbatim and enumerate every explicit ask against what actually shipped.” Any ✗ blocks phase: complete and loops back to PLAN. — PAI/Algorithm/ (LATEST)
• Reproduce-First — Doctrine — “If Preflight Gate A fired, a reproduction MUST be captured before ANY Read/Grep targets the suspect code path.” Code analysis without reproduction is speculation. — PAI/Algorithm/ (LATEST)
• Resolve, don’t recall — Convention — “Any fact about known infra/environment MUST come from bun PAI/Tools/pai-infra.ts resolve <thing> or a live probe — NEVER from memory/plausibility.” Read-path twin of the pai-provenance commit gate. — CLAUDE.md
• Retrieve, don’t generate — Doctrine — “The retrieval/generation boundary is the rule: retrieve from a verified source, or do not assert.” — PAI/PAI_SYSTEM_PROMPT.md
• Reverse Engineering — Doctrine — OBSERVE block enumerating explicit wants, explicit not-wanted, implied not-wanted, and the speed/urgency signal. — PAI/Algorithm/ (LATEST)
• Root-Cause-at-Ingestion Checkpoint — Doctrine — BUILD checkpoint: “Where does this bad state enter the system?… If I fix it at the ingestion point instead of here, do 3 similar bugs disappear? If yes → move the fix upstream.” — PAI/Algorithm/ (LATEST)
• RTK — Tool — Rust Token Killer: “token-optimized CLI proxy (60–90% savings on dev operations)”; a hook transparently rewrites commands (git status → rtk git status). Meta commands: rtk gain, rtk discover, rtk proxy. — RTK.md
• Rule 1 / 1b / 2 / 2a / 3 — Doctrine — The Verification Doctrine: 1 Live-Probe (user-facing artifacts pass only with tool-verified probe evidence; persistent services must be running AND enabled), 1b Independent Cross-Check (derived sets need two independent implementations asserting identity, not count), 2 Commitment-Boundary Advisor (advisor() at PLAN→BUILD for E2+, when stuck, and before phase: complete), 2a Cross-Vendor Audit (Cato at E4/E5), 3 Conflict-Surfacing (never silently override the advisor; max two re-calls, then escalate to the user). — PAI/Algorithm/ (LATEST)

S

• Security (archetype) — Org — Aka CISO, SOC, Red Team, GRC: “Triage, scan authorized scope, preserve evidence, respond. Cannot accept risk unless T0.” — PAI/RBAC_ORG_STRUCTURES_MODEL.md
• Security Escalation — Doctrine — For safety-asserting claims (“no vulnerabilities”, “patched”, “safe to ship”) the Honesty floor’s silent-drop exit is REMOVED: uncited = UNVERIFIED + open finding; “absence of a finding in a domain you were tasked to assess is itself UNVERIFIED unless you can cite the check.” — PAI/Algorithm/ (LATEST)
• Security Override Protocol — Security — The three-option flow (override once / override session / override permanent) when a tool call is blocked; two variants — path-based (path field) and pattern-based (pattern_slug, DestructiveOpGuard). Pending entries: ~/MEMORY/STATE.claude/security-pending.json. — CLAUDE.md, PAI/DOCUMENTATION/SECURITY_OVERRIDE_MECHANISM.md
• Security philosophy — Security — “Block catastrophic, confirm destructive, alert suspicious, allow everything else.” Bash tiers: trusted (fast-path allow) / blocked (exit 2 catastrophic) / confirm / alert. Path zones: zeroAccess (secrets), readOnly, confirmWrite, noDelete. Without a patterns file SecurityValidator is fail-open. — PAI/PAISECURITYSYSTEM/patterns.example.yaml
• Self-Healing Infrastructure — Doctrine — “When the system fails… fix the system, not your notes. An OS doesn’t accumulate sticky notes about its own bugs, it patches itself.” Rules go to CLAUDE.md/hooks/settings/skills — never memo files. — PAI/PAI_SYSTEM_PROMPT.md
• SessionBus — Tool — “Cross-session/cross-engine coordination for parallel PAI sessions… EMIT / SURFACE / MUTEX (flock on a shared lockfile so concurrent ~/.claude committers can’t clobber HEAD).” Use SessionBus commit for concurrent-session commits. — PAI/Tools/SessionBus.ts
• Skill — Component — “Self-contained instruction sets that teach the DA how to perform specific domains of work” (skills/<Name>/SKILL.md + Workflows/). The Skill System is “the MANDATORY configuration system for ALL PAI skills.” ~127 skill directories at writing. — PAI/ARCHITECTURE_RELATIONSHIPS.md, PAI/SKILLSYSTEM.md
• Spine — Component/Tool — The Harness Spine thin kernel (PAI/Tools/Spine/): “migrate / derive / seed-enforcement / render / doctor. Freshness invariant: render/check ALWAYS derive first; the DB is a cache, not a source of truth.” Emits SPINE:GENERATED markers into docs like HARNESS.md. — PAI/Tools/Spine/cli.ts
• Splitting Test — Doctrine — Split an ISC when: it joins two verifiable things (“and”/“with”), parts can fail independently, it uses scope words (“all/every/complete”), it crosses a domain boundary, or no single tool probe can verify it. — PAI/Algorithm/ (LATEST)
• STATE.claude — Memory — The doctrine path ~/MEMORY/STATE.claude/ for engine-adjacent runtime state (security-pending.json, honesty-ramp.json, usage-canonical-cache.json, omnipulse-handoffs/…), kept outside the root-owned legacy tree. — PAI/DOCUMENTATION/SECURITY_OVERRIDE_MECHANISM.md
• SUMMARY block (7/7) — Doctrine — The mandatory Algorithm closing: “the ONLY acceptable final output is this block” — 🔄 ITERATION / 📃 CONTENT / 🖊️ STORY / 🗣️ signature. Prose recaps after it are a critical failure. — PAI/Algorithm/ (LATEST)
• Switchboard — Component — AGT-004: “file-based message routing between engines / real-time message routing between agents.” — PAI/ARCHITECTURE_RELATIONSHIPS.md

T

• Telos — Core Concept/Skill — “The goal system: the Principal’s mission, goals, metrics, challenges, and strategies that define where ‘ideal state’ actually points.” Files under PAI/USER/TELOS/; the Telos skill analyzes them. — PAI/DOCUMENTATION/PAISystemPhilosophy.md
• THINK — Doctrine — Algorithm phase 2/7: riskiest assumptions, premortem, prerequisites, ISC refinement, euphoric-surprise prediction, Under-Claim Check, Security Escalation. — PAI/Algorithm/ (LATEST)
• Troubleshooting Method / 5-Step Loop — Doctrine — “Troubleshooting is an iterative loop, not a linear path… a session that shows only the final success is hiding the method.” The loop: show failing state → state hypothesis → try ONE fix → re-probe → iterate/escalate (FIXED/SAME/WORSE/DIFFERENT). E2+ debugging ISAs require a ## Troubleshooting Log. — PAI/DOCUMENTATION/TROUBLESHOOTING_METHOD.md
• Trust, but Verify — Doctrine — “Extend good faith freely… but never let trust stand in for evidence. Trust is the posture; verification is the discipline.” — PAI/PAI_SYSTEM_PROMPT.md

U

• Under-Claim Check — Doctrine — For components that make claims about their own behavior: “name the mechanical proof that substantiates it. If you cannot name the proof, under-claim — state only the weaker thing you can prove.” — PAI/Algorithm/ (LATEST)
• unsourced-claim — Doctrine — HonestyValidator finding kind (warn-only telemetry): a claimed path/fact with no corroborating tool call in the session log. — PAI/ENFORCEMENT_MAP.md
• UsageProbe / Canonical Budget Source — Tool/Convention — “Query Anthropic’s canonical usage endpoint… the authoritative subscription window state, not a local estimate” (GET /api/oauth/usage, mandatory 5-min cache at usage-canonical-cache.json). The only legitimate budget input for the Quota-Check Gate. — PAI/Tools/UsageProbe.ts, PAI/DOCUMENTATION/BUDGET_SOURCE.md

V

• Verifier — Org — Archetype (aka QA, Reviewer, Board, Assessor): “Read, challenge, block closure, certify outcome. Cannot perform producer changes in same gate.” — PAI/RBAC_ORG_STRUCTURES_MODEL.md
• VERIFY — Doctrine — Algorithm phase 6/7: the Verification Doctrine (Rules 1/1b/2/2a/3), per-ISC evidence, capability/preflight/doctrine/deliverable compliance checks, and the Re-Read Check. — PAI/Algorithm/ (LATEST)
• Verification Doctrine — Doctrine — “Four rules govern every VERIFY pass. They are NOT optional.” See Rule 1/1b/2/2a/3. — PAI/Algorithm/ (LATEST)
• Voice Announcements — Doctrine/Component — “At Algorithm entry and every phase transition, announce via direct inline curl” to the :31337 /notify route; always self-identifying (“Entering the phase for ”). Voice is audio-only; the ISA edit is the phase signal. Voice ID from MEMORY/STATE/voice-config.json. — PAI/Algorithm/ (LATEST)
• VoiceServer / Voicebox — Component/Infra — The voice runtime: local PAI Voice Server (port 8888, fallback) with Pulse (:31337) forwarding to the Legion Voicebox Sound Server (:17493, POST /speak) — “talk in, PAI answers aloud.” — PAI/VoiceServer/server.ts, PAI/ASSET_REGISTRY.md

W

• Watchdog Gate — Doctrine — “On first background agent spawn in a session, start the agent watchdog if not running” (Monitor + PAI/Tools/AgentWatchdog.ts). — PAI/Algorithm/ (LATEST)
• Wazuh — Infra — The SIEM: Wazuh Manager (agent log ingestion, OSSEC protocol, port 1514) + Wazuh Dashboard (:8601) with indexer at :9200; manager runs on the NAS. — PAI/ASSET_REGISTRY.md

---

Registered in PAI/CONTEXT_ROUTING.md (PAI System table) and PAI/DOCUMENTATION/INDEX.md. Companion surfaces: PAI/USER/DEFINITIONS.md (personal definitions), PAI/DOCUMENTATION/QUICK_REFERENCE_CARD.md, PAI/DOCUMENTATION/NAVIGATION_MAP.md.