FDE BenchmarkPublic document reader
Available in sourcev0.1 · live public draft · official release pending

Canonical source view. This renders the bundled repository document as readable HTML. Source metadata remains in the repository, and relative document links remain inside this reader.

Family Benchmark Depth Traceability

Status: proposed and not evaluated
Ontology source: ONTOLOGY/releases/0.1.0/capability-families.json
Depth artifact: ONTOLOGY/fde/0.1.0/family-benchmark-depth.proposed.json
Typed public-site module: apps/public-site/src/data/familyBenchmarkDepth.ts

Purpose and authority

This layer gives every released v0.1.0 capability family a proposed set of observable behaviors, minimum evidence, an applied task, a changing constraint, failure indicators, acceptable alternatives, reference prompts, observable outcomes, and explicit non-inference limits. It is intended to make future assessment-design review concrete.

The released ontology remains authoritative for family identity, domain, name, and definition. The new depth records are authored proposals. They are not a validated measurement model, rubric, score, cutoff, certification rule, or authorization for employment use. A listed artifact is a possible observation source; producing it would not by itself establish workplace performance or individual contribution.

Field contract

Field Proposed use Must not be read as
observable_behaviors Specific actions or distinctions an evaluator could attempt to observe in the bounded task. A validated latent construct or complete behavioral definition.
minimum_evidence_set Smallest proposed set of reviewable artifacts for the task design. Automatic proof that the family is present.
applied_task A family-specific work sample with named requested artifacts. A production simulation or job-readiness decision.
constraint_injection A mid-task change intended to expose adaptation and boundary handling. A calibrated difficulty increase.
failure_indicators Observable reasons for further review or revision of the work product. Automatic deductions or universal failure rules.
acceptable_alternatives Multiple defensible approaches that help prevent one-solution grading. A complete list of valid approaches.
reference_prompts Starting language for task-design and evaluator review. An administered or validated assessment instrument.
observable_outcomes Inspectable properties the proposed task could yield. Evidence of business impact or production performance.
non_inference_limits Claims that the bounded observation cannot support. Boilerplate that can be removed when a response looks strong.
source / status Per-record lineage to the released family and an explicit proposed, non-scoring state. Independent corroboration or validation evidence.

Exact release coverage

The artifact retains source order and covers 55 unique IDs: Discovery 6, Software 7, AI Native 8, Architecture 6, Enterprise 7, Production 7, Outcomes 4, Leadership 4, Modernization 2, and Specialization 4.

Released family ID Released family name Unique proposed applied task
DISC.STAKEHOLDER_DISCOVERY Stakeholder Discovery Map the contested intake redesign
DISC.WORKFLOW_AND_CONSTRAINTS Workflow and Constraint Discovery Reconstruct the exception-heavy approval flow
DISC.PROBLEM_FRAMING Problem Framing and Decomposition Turn an executive AI mandate into testable work
DISC.OUTCOME_CONTRACT Outcome Contract and Acceptance Criteria Contract a bounded claims-assistant pilot
DISC.OPPORTUNITY_AND_PRIORITIZATION Opportunity Sizing and Prioritization Prioritize four automation candidates
DISC.CUSTOMER_TRUST Customer Trust and Expectation Management Reset expectations after an evaluation miss
SWE.PROGRAMMING_AND_CODE_QUALITY Programming and Code Quality Repair duplicate webhook processing
SWE.API_AND_SERVICE_DESIGN API and Service Design Design an asynchronous document-processing API
SWE.DATA_AND_EVENT_SYSTEMS Data and Event Systems Repair an out-of-order customer-state pipeline
SWE.IDENTITY_AND_AUTHORIZATION Identity and Authorization Engineering Close a cross-tenant export path
SWE.DELIVERY_AND_TESTING Delivery Automation and Testing Harden the release path for a policy service
SWE.OBSERVABILITY_AND_RELIABILITY Observability and Reliability Engineering Make a fan-out workflow operable
SWE.PERFORMANCE_AND_DEBUGGING Performance, Scale, and Debugging Diagnose the p99 latency cliff
AI.MODEL_SELECTION Model Selection and Routing Choose a model portfolio for mixed-risk support
AI.PROMPT_CONTEXT_GROUNDING Prompt, Context, Retrieval, and Grounding Ground a policy assistant in conflicting sources
AI.AGENT_ORCHESTRATION Agent Decomposition, State, and Orchestration Orchestrate a recoverable onboarding agent
AI.TOOLS_PERMISSIONS_BOUNDARIES Tool, Permission, and Action Boundaries Constrain a production-support agent
AI.STRUCTURED_AND_DETERMINISTIC_CONTROL Structured Outputs and Deterministic Control Bound an eligibility-extraction component
AI.EVALUATION_AND_GUARDRAILS AI Evaluation, Guardrails, and Red Teaming Design the release evaluation for an action-drafting assistant
AI.OPERATIONS_COST_LATENCY AI Operations, Cost, Latency, and Observability Operate a model gateway through a cost spike
AI.ADAPTATION_AND_OPEN_WEIGHT Adaptation, Fine-Tuning, and Open-Weight Systems Decide whether to adapt a private domain model
ARCH.CURRENT_TARGET_STATE Current-State and Target Architecture Reframe a fragmented case-management estate
ARCH.TRADEOFF_REASONING Tradeoff Reasoning and Decision Records Choose between shared and tenant-isolated retrieval
ARCH.BOUNDARIES_AND_INTERFACES Boundaries, Interfaces, and Integration Untangle order, entitlement, and billing ownership
ARCH.SECURITY_AND_RESILIENCE Security, Resilience, and Failure Domains Contain a regional model-platform compromise
ARCH.MIGRATION_AND_SEQUENCING Migration, Sequencing, and Technical Risk Sequence a live customer-profile migration
ARCH.REUSE_AND_PRODUCTIZATION Reuse, Platform Judgment, and Productization Decide whether three pilots deserve a shared platform
ENT.PRIVACY_DATA_CLASSIFICATION Privacy and Data Classification Minimize data for a conversation-quality pilot
ENT.IDENTITY_SECRETS_ACCESS Identity, Secrets, and Access Governance Replace shared deployment credentials
ENT.THREAT_AND_SECURE_SDLC Threat Modeling and Secure SDLC Threat-model a file-to-agent workflow
ENT.COMPLIANCE_AND_AUDIT Compliance Translation and Auditability Operationalize a model-change approval requirement
ENT.AI_MODEL_RISK_GOVERNANCE AI and Model Risk Governance Govern an AI recommendation in a consequential workflow
ENT.RELEASE_CONTINUITY_READINESS Release, Continuity, and Production Readiness Decide readiness for a dependency-heavy launch
ENT.ACCESSIBILITY_RESPONSIBLE_DEPLOYMENT Accessibility and Responsible Deployment Redesign an inaccessible AI-assisted intake
PROD.DEPLOYMENT_ENVIRONMENT_READINESS Deployment and Environment Readiness Prepare a private-network deployment
PROD.ROLLOUT_ROLLBACK_ACCEPTANCE Rollout, Rollback, and Production Acceptance Stage a recommendation-engine rollout
PROD.SLO_TELEMETRY_RELIABILITY Service Objectives, Telemetry, and Reliability Interpret a misleading green dashboard
PROD.INCIDENT_ESCALATION_POSTMORTEM Incident, Escalation, and Postmortem Practice Lead a partial-data-exposure incident
PROD.MODEL_QUALITY_PROVIDER_FAILURE Model Quality, Drift, and Provider Failure Contain a silent model-quality regression
PROD.COST_CAPACITY_LIFECYCLE Cost, Capacity, and Lifecycle Operations Plan for a tenfold seasonal workload
PROD.RUNBOOK_HANDOFF_OWNERSHIP Runbooks, Handoff, and Operational Ownership Hand off an AI intake service to customer operations
OUT.BUSINESS_CASE_BASELINE Business Case and Baseline Build a baseline for an assisted-resolution proposal
OUT.METRICS_EXPERIMENT_ATTRIBUTION Metrics, Experimentation, and Attribution Measure an AI triage rollout without randomizing customers
OUT.VALUE_PERSISTENCE_ADOPTION Value Persistence and Adoption Investigate a post-launch usage collapse
OUT.SCALE_STOP_COMMUNICATION Scale, Iterate, Stop, and Outcome Communication Recommend the fate of a mixed-result pilot
LEAD.EXECUTIVE_ENGINEERING_COMMUNICATION Executive, Customer, and Engineering Communication Translate a difficult launch review across audiences
LEAD.NEGOTIATION_CONFLICT_DECISIONS Negotiation, Conflict, and Decision Clarity Resolve a launch-scope conflict
LEAD.DELIVERY_OWNERSHIP_PRESSURE Delivery Ownership Under Pressure Replan a fixed-date customer pilot
LEAD.MENTORING_TRANSFER_FEEDBACK Mentoring, Capability Transfer, and Product Feedback Transfer ownership and escalate a recurring product gap
MOD.LEGACY_DEPENDENCY_DECOMPOSITION Legacy Discovery, Dependencies, and Decomposition Find safe seams in a nightly settlement monolith
MOD.MIGRATION_PLATFORM_OPERATING_MODEL Migration, Platform, and Operating-Model Transformation Plan a multi-business cloud and data transformation
PACK.HEALTHCARE_AI Healthcare AI Bound a discharge-summary drafting assistant
PACK.CONTACT_CENTER_AI Contact-Center AI Redesign a high-transfer virtual-agent flow
PACK.AI_INFRASTRUCTURE AI Infrastructure and Model Operations Design a private low-latency inference platform
PACK.REGULATED_ENTERPRISE_AI Regulated Enterprise AI Control an AI-assisted regulated case decision

Structural validation

Run from the repository root:

node scripts/check-family-benchmark-depth.mjs

The focused check compares the proposal with the immutable v0.1.0 family release. It requires exact count, ID set, source order, domain, name, and per-record source binding; rejects duplicate IDs and scoring fields; checks the minimum depth of every structured field; and requires unique task titles, constraint injections, and reference prompts. Passing this check establishes structural coverage only. It does not validate task content or measurement quality.

Public-site integration

Components beneath apps/public-site/src/ can import the proposed layer with:

import {
  familyBenchmarkDepthById,
  familyBenchmarkDepthRecords,
  getFamilyBenchmarkDepth,
  type CapabilityFamilyId,
} from "./data/familyBenchmarkDepth";

For a known family:

const depth = getFamilyBenchmarkDepth("AI.EVALUATION_AND_GUARDRAILS");

For an existing runtime string, first check it against capabilityFamilyIds; do not coerce arbitrary strings into CapabilityFamilyId. The module performs count, membership, duplicate, and source-binding checks when loaded and exposes both the ordered records and an ID-indexed view.

Any UI using this module should display the proposed/not-evaluated state and the relevant non_inference_limits. It should not derive numeric values, readiness labels, rankings, or certification language from these records. No explorer or homepage component is wired to this layer in this change.