# COMPETENCY RECORD - Cedric Ancellin's demonstrated engineering skills and
# experience. Third-party technologies and vendors are named as the concrete
# skill surface. Professional background / track record.

# ============================================================
# ORACLE AI APPLICATIONS & USE CASES - COMPETENCY RECORD
# ============================================================
# Core Purpose:
#   Demonstrated experience turning the Oracle knowledge-architecture
#   methodology (COMP-23) into working AI capability. Where COMP-23 is
#   "what the Oracle IS," this record is "what it is FOR": the concrete
#   applications an Oracle unlocks once a governed, agent-consumable
#   knowledge substrate exists.
#
#   Described in abstract archetypes - a transferable set of applications,
#   not any single implementation.
#
# Governed by ../LAWS.
# ============================================================

summary: >
  Designed and proved a broad set of AI applications on top of the Oracle
  methodology: an Oracle is not a documentation store but a governed, structured,
  source-anchored knowledge substrate served to agents, and once it exists, a
  wide range of engineering, analytics, agent-governance, documentation, and
  operations workflows become possible on it. This record catalogs those
  applications and the value each delivers.


# ==================================================================
# 1. THE CENTRAL IDEA
# ==================================================================
central_idea: >
  Every application reduces to the same move: give an AI agent grounded,
  structured, source-anchored context plus the rules that govern change, served
  at the right access tier and scoped to the right audience, and the agent can
  then perform work that previously required a human with deep system knowledge.
  The Oracle is the substrate; the applications are what gets built on it, and one
  tree supports many of them at once.


# ==================================================================
# 2. ENGINEERING
# ==================================================================
engineering:
  ai_first_development:
    what: "A system whose primary authors are AI agents, with a human directing rather than writing code."
    enabled: "The Oracle is the entire institutional memory, each session bootstraps from the spine, obeys the laws, avoids catalogued anti-patterns, and records new decisions for the next session."
    value: "Cross-session continuity with no human memory in the loop; a project can be built and maintained by agents without knowledge loss."
  ai_assisted_development:
    what: "AI agents augmenting human engineers on an existing team and codebase."
    enabled: "The agent gets the same grounded context a senior teammate would have, architecture, past decisions, patterns to mirror, hazards to avoid, so it proposes changes that fit real conventions, not generic best-guess code."
    value: "Faster, higher-fit contributions and less review churn."
  code_review_and_impact_analysis:
    what: "Reviewing a change for correctness and for its blast radius across the system."
    enabled: "The agent checks against laws and anti-patterns, then traces impact by walking the $ref graph and source anchors; at the data-connected tier it reads real source and live data to confirm actual reach."
    value: "Reviews that catch systemic impact and rule violations a diff-only reviewer misses."
  code_review_dimensions:
    what: "Review beyond 'does it work', correctness, process, logic, and purpose."
    enabled: "Because the Oracle records WHY (decisions, product intent), a review can catch a change that works but quietly reverses a deliberate decision, the failure mode generic AI review lacks."
    value: "Protects architectural intent and process, not merely syntax, a senior reviewer, not a linter."
  debugging:
    what: "Diagnosing a defect or incident and finding root cause."
    enabled: "The agent orients on architecture and flow docs for expected behavior, checks known-issues/anti-patterns for prior occurrences, and at the data tier reproduces against real source and live data, logging findings back so the knowledge persists."
    value: "Root cause found against the system's own documented behavior; every session strengthens the Oracle."


# ==================================================================
# 3. TRUSTED ANALYTICS & SEMANTIC GOVERNANCE
# ==================================================================
trusted_analytics:
  problem: "Point an agent at a warehouse and it returns confident, plausible-but-wrong numbers, wrong joins, non-canonical metric definitions, ruinous queries. Wrong analytics that look right are worse than none."
  approach: >
    Govern a semantic layer AS Oracle records: a data dictionary (what each field
    means), canonical metric definitions (one agreed definition per KPI), an
    entity/ERD model (the correct joins), a library of verified queries, execution
    guardrails (row/time/cost caps), and forbidden query patterns.
  workflow: "The agent reads the definitions and ERD, adapts a verified query within the guardrails and clear of forbidden patterns, executes read-only against the warehouse, and cites the canonical metric used."
  value: "Correct-by-construction natural-language analytics, one auditable source of truth for every KPI, used at query time rather than filed in a governance wiki nobody reads."


# ==================================================================
# 4. AGENT GOVERNANCE & MULTI-AGENT ORCHESTRATION
# ==================================================================
agent_governance:
  control_plane:
    what: "A machine-readable control plane for autonomous-agent behavior, agent laws, forbidden actions, required disciplines, and scope of autonomy, carried as structured records the agent reads first."
    value: "Auditable, consistent, compliant agent operation at fleet scale, versioned and reviewable instead of buried in prompts."
  multi_agent_orchestration:
    what: "The Oracle as the single shared coordination substrate for a multi-agent system."
    enabled: >
      Workers stop carrying the system in their prompts and stop coordinating
      through bespoke handoffs; they pull the scoped slice they need from the
      shared tree, inherit the same rules from one place, and hand off by writing
      results other workers read. Specialization moves out of fragile agents into
      the durable knowledge base.
    value: "Large reductions in cost (context fetched once, not duplicated per worker), orchestration complexity, and maintenance, and a sharp drop in errors, because every worker shares one ground truth and one rules layer. The gains scale with system complexity."


# ==================================================================
# 5. KNOWLEDGE, DOCUMENTATION & OPERATIONS
# ==================================================================
knowledge_and_documentation:
  developer_knowledge_base: "A living, queryable knowledge base, architecture, flows, decisions, that agents and humans both read."
  durable_memory: "Todos, known tech-debt / 'ugly ducklings', and postmortems as structured, queryable memory rather than lost tribal knowledge."
  document_generation: "Briefs, RFP/proposal drafts, security documentation, and white papers assembled from source-anchored records, with firewalled builds for external artifacts."
  product_and_operations: "Training material, plain-language technical explanation for non-engineers, ticket categorization and issue-tracker analysis, and secure workforce mapping for time-tracking and capex/opex classification."


# ==================================================================
# 6. INTEGRATION, DESIGN & FLEET
# ==================================================================
integration_design_fleet:
  cross_system_integration: "Delegated (act-on-behalf) access across foreign systems, opaque-identifier→meaningful-name mapping, and entity resolution to a single view of an entity across data sources."
  design_governance: "An enforceable governance layer for AI-assisted design, design-system-as-artifact, reuse-first components, token discipline, and design-to-code handoff across strategy/design/implementation agents with human approval gates, preventing visual-system drift and unapproved redesigns."
  mcp_fleet: "Maintaining multiple MCP endpoints, each with its own scope, instructions, and processes, from a common substrate."
  dimensional_tracking: "Multi-dimensional continuity tracking over an ordered spine with audience-scoped progressive disclosure and a canonical source deriving many synchronized variants."


# ==================================================================
# 7. PROVEN BREADTH
# ==================================================================
proven_breadth: >
  These applications were exercised across domains with nothing in common, a
  large multi-repository software platform, a single AI-built web application, a
  multi-service consumer product, a multi-tenant documentation host, and a
  long-form creative work, which is the evidence they rest on a general
  methodology rather than a one-off convention. This résumé itself is served from
  an Oracle, so the reader is inside one of these applications while reading it,
  and from one authored source it is projected to several surfaces at once: raw
  structured records for agents that fetch directly, an MCP endpoint for
  connector-based assistants, rendered HTML for people and search crawlers, and
  purpose-built pages for key sections. One tree, every audience served in the
  form that suits it.


cross_reference:
  knowledge_architecture_curation:
    $ref: "https://zc8.com/competencies/23_knowledge_architecture_curation"
    note: "The methodology itself, 'what the Oracle is.' This record is 'what it is for.'"
  ai_knowledge_engineering:
    $ref: "https://zc8.com/competencies/11_ai_knowledge_engineering"
    note: "The software-domain instances and the MCP serving layer these applications run on."
  applied_ai_llm:
    $ref: "https://zc8.com/competencies/16_applied_ai_llm"
    note: "Owning the applied-AI stack (hosting, fine-tuning, serving) beneath agent applications."
  ai_orchestrated_delivery:
    $ref: "https://zc8.com/competencies/20_ai_orchestrated_delivery"
    note: "AI-first development in practice, a governance Oracle as a memoryless author's institutional memory."


implicit:
  - "Applications are described at competency altitude; the methodology's internal mechanics (serving tiers, governance skeleton, conventions) are the subject of COMP-23, not repeated here."
  - "Several applications compose, debugging feeds postmortems, which feed the anti-patterns layer that then improves code review."
  - "Applications touching people data or live systems inherit least-privilege and audience-scoping obligations."
  - "Value figures (cost/error reductions in multi-agent orchestration) are practitioner assertions from real usage, not independent benchmarks."