Every platform builds a vocabulary. This is ours. If a term appears in ORCA documentation, on our website, or in conversation with our team, you will find its definition here.
Intelligence Platform
Infrastructure for organisational memory. Not an AI model, not a chatbot, not a document search engine. An intelligence platform captures expertise, classifies how much to trust it, governs who can access it, and compounds it over time. The distinction matters: an AI tool generates output; an intelligence platform accumulates institutional knowledge that makes every future output better.
Brain
A governed knowledge store within ORCA. Each brain is an isolated collection with independent encryption, access controls, and retention policies. No brain bleeds into another without explicit, governed promotion. ORCA has four brain levels, each serving a different purpose.
Practice Brain
The shared organisational memory. Every employee and every AI agent can read it. Writes are governed — every entry passes a generalisation adequacy checklist, PII tokenisation, and multi-model adversarial review before it lands. The practice brain holds no client-identifying information. It contains only generalisable case law: the collective validated knowledge of the entire firm.
Engagement Brain
One per client or project. Team-scoped and fully isolated. All content is PII-tokenised before storage. Subject to 12-month retention from engagement close. Knowledge enters by direct entry or by governed promotion from personal brains.
Personal Brain
Private to each user. AES-256-GCM encrypted. One isolated collection per person. Invisible to administrators, managers, and every other user — including the founder. This absolute privacy is the foundation of adoption. An employee who trusts ORCA with their genuine thinking will use it every day. Personal brains have no quality gates — they are a frictionless capture space.
Governance Brain
Policy, compliance, and leadership intelligence. Three sub-types exist: the Founder brain (founder writes and reads, no other access), Director brains (each director writes and reads their own only), and the OrgOS brain (founder writes, directors and founder read). Content flows through deliberate owner-initiated promotion from personal brains — never automatic.
Confidence Classification
The five-level grading system that tells users how much to trust each answer. Every knowledge entry carries a confidence tag. This is the honest grading system — it does not claim certainty where none exists.
Confidence Levels: Definitive, Grounded, Inferred, Advisory, Insufficient
Definitive means verified fact from multiple independent sources — safe to act on. Grounded means strong evidence — verify before irreversible decisions. Inferred means connected dots — useful direction, not proof. Advisory means weak or older evidence — a hint, not an answer. Insufficient means the brain does not know — a signal to find out and feed it back.
OODA Loop
Observe, Orient, Decide, Act. A decision-making framework developed by Colonel John Boyd to explain why certain fighter pilots consistently won engagements. ORCA applies the OODA loop to organisational intelligence: Observe captures what your people learn; Orient classifies it by confidence and places it in the right brain; Decide surfaces the right intelligence at the right moment; Act closes the loop by capturing the outcome. The side that loops fastest wins.
Knowledge Equity
The economic footprint of captured expertise. Every time a piece of organisational knowledge is reused — a junior consultant retrieves a validated answer, a support engineer finds a prior resolution, a director accesses institutional context — it generates downstream value. Knowledge equity measures that compounding return. Not a score. Not a badge. An economic footprint.
Constitutional Governance
Rules enforced in executable code, not in policy documents. In ORCA, write gates, PII gates, and retention policies are code that runs on every operation. A policy document can be ignored. A code gate cannot be bypassed. The adversarial auditor, the PII tokeniser, and the confidence classifier all operate constitutionally — they execute every time, without exception.
Write Gate
The governed write path that every entry destined for the practice brain must pass through. Includes YAML frontmatter validation, PII gate, four-stage PII tokenisation, Phi-4 combined quality gate, and a three-model adversarial challenger pipeline. The write gate is constitutional — it cannot be skipped, paused, or overridden.
PII Vault / PII Tokenisation
The deterministic token store that replaces personally identifiable information with encrypted tokens before knowledge enters shared brain stores. Uses a four-stage pipeline: regex scanning, NER scanning, role-organisation tokenisation, and AI entity resolution. Tokens are stored in Azure SQL with deterministic UUIDv5 mappings, enabling GDPR Subject Access Requests and right-to-erasure by deleting the token mapping.
Envelope Encryption
The third layer of PII defence. Every knowledge entry is encrypted with a unique per-entry data encryption key (AES-256-GCM, random 256-bit). That DEK is then encrypted by a Key Encryption Key held in Azure Key Vault. The ciphertext and the encrypted DEK are stored together. Decryption requires Key Vault access — infrastructure-level enforcement that cannot be circumvented by application logic.
MCP (Model Context Protocol)
An open standard for connecting AI models to external tools and data sources. ORCA is MCP-native — its gateway speaks MCP natively, meaning any MCP-compatible AI surface can connect to ORCA's brain stores, search organisational knowledge, and receive confidence-classified results. MCP is the integration layer that prevents vendor lock-in.
Adversarial Auditor
One of ORCA's five autonomous agents. The Auditor challenges individual entries and deliverables through blind scoring and multi-model review. Every entry that enters a shared brain faces adversarial scrutiny from independent frontier models. Challenger verdicts — AGREE, CAUTION, CHALLENGE, or OPPOSE — surface disagreements to humans. The system never self-modifies.
The Heartbeat
ORCA's operational health monitoring agent. The Heartbeat checks infrastructure health, MCP server connectivity, knowledge drift, and design alignment on a regular schedule. It compares the live system against its own architecture specification and scores the gap. It asks one question: "Is the system working?"
The Gardener
ORCA's population health maintenance agent. The Gardener manages brain hygiene across all non-personal brains: deduplication, archival, retention enforcement, relevancy scoring, coherence checking, and non-redundancy. It asks one question: "Is the knowledge base as a whole healthy?"
Compounding Intelligence
The principle that an intelligence system gets more valuable every day it runs. Week 1 is the worst it will ever be. Each new entry creates connections with every entry that already exists, producing an exponential rather than linear growth curve. A firm that has been running ORCA for six months holds six months of validated institutional knowledge that a competitor starting today cannot replicate regardless of which AI tools they adopt.
The Typewriter Trap
The pattern where organisations use new technology to do the same things faster rather than to do fundamentally different things. Named after the typewriter, which made writing faster but did not make the writing better. Most organisations today are using AI as a faster typewriter — producing more documents, emails, and summaries at higher volume. ORCA exists to break this pattern by building persistent, governed memory instead of accelerating transient output.
Cognitive Load
The framing for the problem ORCA solves. Your people are not underperforming — they are carrying weight that the organisation should be carrying for them. Cognitive load is the mental overhead of remembering what was learned, finding what was documented, and reconstructing context that should have been preserved. ORCA handles the weight so your people can focus on judgement.