One standard. A governed way to keep it current.
LockedIn Labs is the implementation, consulting, and delivery firm. It operates a training platform whose shared core standard is the AI Native Operator Framework: five levels, seven curriculum tracks, six fixed capability dimensions, and proof required at every gate. The architecture below shows how that standard stays stable as model and platform capabilities change.
Define the standard. Curate the frontier. Cultivate the operator. Prove the claim.
This is not another product or destination. It is the governed protocol the firm's training platform is encoding for turning changing model and platform capability into bounded learning packages, individual routes, practiced judgment, and inspectable evidence. The first authored lineage is inspectable now; external practitioner ratification and continuous Foundry operation are not active.
“At LockedIn Labs, AI-native means demonstrated operating capability across six fixed dimensions—AI literacy & judgment; Prompting & instruction design; Workflow decomposition; Agent direction & oversight; Verification & quality control; and Security & governance awareness—applied at increasing scope across changing models and platforms, and evidenced through reviewed work rather than attendance, tool usage, or self-identification.”
Frontier change enters as a claim. It earns standing as evidence.
- 01AI literacy & judgment
- 02Prompting & instruction design
- 03Workflow decomposition
- 04Agent direction & oversight
- 05Verification & quality control
- 06Security & governance awareness
- 1CORE STANDARD
AI Native Operator Framework
Defines the stable capability standard: six dimensions, five levels, and the bar evidence must clear.
- 2CURATION
Capability Foundry
Defines the governed protocol for curating frontier change into versioned Capability Packages without creating a new product surface.
- 3TOPOLOGY
Capability Graph
Maps each package into prerequisite-aware curriculum nodes and the authored modules behind them.
- 4PRACTICE
AI Fieldcraft
Supplies the repeatable method for testing an unfamiliar capability under real constraints.
- 5ROUTING
Adaptive Core
Routes the individual through prerequisite-aware practice in response to recorded evidence.
- 6PROOF
Evidence Chain
Preserves the provenance, attempts, reviews, and verification behind every capability claim.
The institutional curation protocol inside the AI Native Operator Framework that defines how a sourced frontier change can become a bounded, teachable, calibrated, and renewable unit of capability.
- 1 · SourceImplemented
A dated source record binds public first-party links to a stated citation-rights basis and original LockedIn synthesis.
EVIDENCE · CPS-001 · 5 public references · public-link-and-original-synthesis - 2 · DistillImplemented
A portable claim, explicit non-claims, and canonical dimension map bound what the package does and does not assert.
EVIDENCE · CPD-001 · 6 canonical dimensions · 4 explicit non-claims - 3 · RatifyActivation gate
External practitioner ratification is not active; no package-specific, actor-linked record exists.
EVIDENCE · No external practitioner ratification record - 4 · ForgeImplemented
An authored forge record binds the mission to its synthetic practice environment and curriculum anchor.
EVIDENCE · CPF-001 · BME-06 · n-l3-3 · developers/ship-an-agent - 5 · CalibrateActivation gate
Calibration is not active; an authored rubric and held-out evaluator exist, but no calibration record does.
EVIDENCE · No calibration record · authored rubric and evaluator do not imply calibration - 6 · CultivateActivation gate
Cultivation is not active; the authored practice lineage exists without a package-specific cultivation record.
EVIDENCE · Authored practice lineage · no cultivation record - 7 · RenewScheduled
No renewal decision is claimed; the authored source set carries a dated review boundary.
EVIDENCE · No renewal record · review scheduled by 2026-10-07
Networked shell agent release authorization
- 01 · SOURCE
5 reviewed references
Observed 2026-08-07 · review by 2026-10-07
- 02 · FIELDCRAFT
FCM-001
Can a hosted shell agent with versioned skills be governed for unattended finance reporting without widening the approved side-effect boundary?
- 03 · L4 PRACTICE
BME-06 · 6 packet documents
practice-only · brief / policy / evidence / log
- 04 · L3 CURRICULUM ANCHOR
Ship an agent to production
n-l3-3 · AI Native Developers
- 05 · RUBRIC
4 dimensions · 100%
Published before submission
- 06 · EVIDENCE
Held-out evaluator key
A held-out evaluator is bound server-side. Its existence does not represent calibration; no calibration record is active, and key contents are not published.
The authored lineage, public packet, rubric, and evaluator binding are inspectable now.
Practitioner ratification requires signed reviewer coverage; none is claimed.
Source renewal is governed by the recorded 2026-10-07 review boundary.
Record the work. Preserve what has not happened.
The first Capability Package now has a hash-linked operational baseline and a versioned calibration-corpus manifest. Source, Distill, and Forge are recorded; Ratify, Calibrate, and Cultivate remain inactive; Renew is scheduled. That separation is the operating discipline—not a disclaimer around an untracked idea.
The first package can no longer change silently.
CP-001 now binds its authored source, claim, package, task, rubric, and held-out evaluator to stable fingerprints. Later work must append a new record; it cannot rewrite this baseline into a ratified, calibrated, or effective result.
- LIFECYCLE RECORDS
- 07
- RECORDED STAGES
- 03
- INACTIVE STAGES
- 03
- CALIBRATION OBSERVATIONS
- 00
- 01recorded
Source
Source-reference metadata was recorded; external source bodies were not snapshotted or hashed.
- RECORD
- CPS-001
- HASH
- 2a2838ed7212
- 02recorded
Distill
The claim is LockedIn-authored synthesis and has not received external practitioner ratification.
- RECORD
- CPD-001
- HASH
- ff90c2a15cb6
- 03inactive
Ratify
No signed, authorized external practitioner ratification exists.
- RECORD
- NOT RECORDED
- HASH
- fb7109211a1a
- 04recorded
Forge
Forge records authored practice; it does not imply prior ratification, calibration, cultivation, or effectiveness.
- RECORD
- CPF-001
- HASH
- 88e7d3c6f700
- 05inactive
Calibrate
Authored task, rubric, and held-out evaluator fingerprints exist; no calibration observation exists.
- RECORD
- NOT RECORDED
- HASH
- a8067763281f
- 06inactive
Cultivate
No consented pilot or package-specific cultivation record exists.
- RECORD
- NOT RECORDED
- HASH
- bad0561f3aa0
- 07SCHEDULED
Renew
No renewal decision exists; the authored source review boundary is 2026-10-07.
- RECORD
- NOT RECORDED
- HASH
- 874e3cc3af64
The instrument is versioned. The observation corpus is not yet populated.
The BME-06 task, public rubric, and held-out evaluator are now bound by digest. There are zero human-rating observations, zero agreement measurements, and zero effectiveness findings. The first consented pilot and independent review coverage must create those records.
- TASK
- BME-06
- RUBRIC DIMENSIONS
- 4
- EVALUATOR
- HELD OUT
- OBSERVATIONS
- 0
- 01 · protocol
Pre-register samples, scoring procedure, thresholds, exclusions, and stopping rules.
- 02 · human-coverage
Contract authorized independent human raters for the claim being tested.
- 03 · consent
Activate purpose-limited consent, retention, access, and deletion handling before participant data enters the corpus.
- 04 · adjudication
Record agreement, disagreement, dissent, adjudication, and evaluator-version changes without overwriting history.
- 05 · analysis
Publish bounded reliability and limitation analysis before making a calibration or effectiveness claim.
Full cycle operated · no. Outcomes observed · no. The machine contract contains authored metadata and fingerprints only—no learner, reviewer, holder, or evaluator contents.
Inspect machine contractThe frontier moves. The standard holds.
AI capability changes too quickly for a catalog to be the whole training system. Fieldcraft is the repeatable discipline inside LockedIn Labs for absorbing what is new without lowering the standard for evidence.
One method. Eight evidence moves.
The discipline of absorbing an unfamiliar AI capability, proving where it works and fails, adapting it to real constraints, and carrying it through evaluation, governance, deployment, and measured operation.
- READ → STRESS
- Separate a frontier claim from observed behavior
- ADAPT → GOVERN
- Fit the capability to real constraints and controls
- DEPLOY → MEASURE
- Release minimally, then inspect the operating result
- 01
Read
What changed, and what is only claimed?
EVIDENCE · SOURCE + CLAIM MAP - 02
Reproduce
Can the claimed capability run from a controlled baseline?
EVIDENCE · REPRODUCIBLE BASELINE - 03
Stress
Where does it break, drift, or create hidden cost?
EVIDENCE · FAILURE LEDGER - 04
Adapt
What must change for the actual workload and constraints?
EVIDENCE · DECISION DIFF - 05
Evaluate
Does it clear the published quality and reliability gates?
EVIDENCE · SCORED EVALUATION - 06
Govern
Which controls, owners, and evidence make use accountable?
EVIDENCE · CONTROL MAP - 07
Deploy
What is the smallest safe release with a way back?
EVIDENCE · RELEASE RECORD - 08
Measure
Did the system improve the intended work in operation?
EVIDENCE · OUTCOME REVIEW
These stages are operating language, not another level, score, or credential. Each pass produces inspectable evidence; measurement and material capability changes send the work back through the loop.
THE MISSION WORKBENCH REQUIRES A PROGRAM ENROLLMENT THAT INCLUDES ITS TRACK. AN ACCOUNT BY ITSELF DOES NOT OPEN CURRICULUM OR LABS.
The curriculum is a graph, not a queue.
Every node below is an authored module; every edge is an explicit prerequisite. This is the same dependency graph learners operate in the workspace, and every node ends in a piece of reviewed work. Open any node to inspect its public brief, artifact, prerequisite route, and rubric-criteria count. Full lessons and field-exercise workspaces open only after LockedIn Labs provisions an enrollment that includes the relevant training.
Thinking in models
AI Native Foundations
Replace hype with a working mental model: what LLMs predict, why that matters, and where agents fit on top.
- ARTIFACT
- Calibration log: 3 prompt variants, drift analysis, and your trust rules.
- RUBRIC
- Scored against 3 authored criteria.
- WORKED EXAMPLES
- 1 verbatim artifact shown in the lesson — the thing this module asks you to produce, printed in full.
- REQUIRES
- Nothing — this is an entry point.
Full lessons require a provisioned program enrollment; an account alone does not unlock curriculum.
The progression, end to end.
Each level has a gate. Each gate is a piece of work, reviewed. There is no other way through.
AI Curious
See the machine clearly.
Strip away the magic. Build accurate mental models of what modern AI systems are, what they are not, and where they genuinely bite.
Planned: an unscripted foundations review, active after vetted examiner coverage.
AI Capable
Run your day on AI.
Turn AI from a novelty into your default way of working. Design personal workflows, write prompts that behave like specifications, and measure what you save.
Ship two verified artifacts: a working personal workflow and a hardened prompt system.
AI Practitioner
Apply it where you work.
Take the craft into your role — engineering, operations, clinical work, or the executive suite. Role-track modules with domain constraints and real stakes.
Ship a role-track capstone artifact. Planned: named practitioner review, active after vetted reviewer coverage.
AI Native Operator
Build and run agentic systems.
The operator level. Compose multi-step agents, orchestrate fleets, evaluate behavior rigorously, and take systems from demo to dependable.
Ship a verified agentic system: working, evaluated, and documented.
AI Native Leader
Transform the organization.
From operator to force multiplier. Strategy, governance, operating models, and the economics of AI transformation — for people who move companies, not just workflows.
Planned: defend a transformation plan in a live board-style review, active after vetted examiner coverage.
Aim the levels at your role.
AI-native does not mean the same thing for an executive, a clinician, and an engineer. Seven tracks plug role-specific modules into the same graph.
AI Native Foundations
Everyone — no technical background requiredThe on-ramp. Mental models, tool literacy, prompting as specification, and your first real AI workflows.
5 modules · proof-of-work each
Forward-Deployed Engineering
Experienced software, data, platform, solutions, and deployment engineersOwn ambiguous AI missions from field diagnosis through enterprise landing, evidence-led operation, and durable ownership transfer.
8 modules · proof-of-work each
AI Native Developers
Working software engineersFor engineers who want to build with agents, not just use autocomplete. Spec-driven development, agentic coding, and shipping agent systems.
6 modules · proof-of-work each
Agentic Workflows & AI Agents
Builders, operators, and technical leadsThe agent track. Anatomy, tools and protocols, orchestration, evaluation, and fleet design for people who want agents doing real work.
5 modules · proof-of-work each
AI Native SDLC & Control Plane
Platform, security, and delivery engineers in regulated orgsThe discipline that makes agentic engineering acceptable inside a bank, a hospital, or an agency: a lifecycle with gates that can stop work, verifiers no builder can approve past, a control plane that decides what agents are allowed to do, and an evidence chain that survives audit.
6 modules · proof-of-work each
Executives & Leaders
CEOs, CIOs, CMOs, and transformation leadersFor leaders who must fund, govern, and scale AI — honest economics, operating models, and governance that ships instead of stalls.
7 modules · proof-of-work each
Healthcare & Regulated Industries
Clinical, operational, and IT leaders in regulated orgsAI native practice where mistakes are regulated, not just embarrassing. PHI guardrails, clinical workflows, validation, and health-system transformation.
5 modules · proof-of-work each
Not sure where you land?
Inspect how three working exercises are read against the fixed dimensions, then compiled into a level, track, and path inside the signed-in assessment.
Inspect the placement methodYou advance by making the work inspectable.
A gate can use work from your own context or a synthetic scenario. Synthetic gates require an artifact produced against published constraints; they are evidence of performance in that scenario, not a claim that a real-world system was deployed.
Inspectable work over quiz completion
Nobody advances by picking options on a multiple-choice test. You advance by producing an inspectable workflow, evaluation, operating plan, prompt system, or agent against the gate's stated constraints.
Review has a published standard
Artifacts can be evaluated against the authored rubric now. Human edge-case review and live defense run only when vetted roster coverage is in place.
A private evidence portfolio that compounds
Artifacts accumulate in your private evidence portfolio across levels and roles. Only an eligible issued credential can appear in the public verification registry; public examples remain labeled specimens.