LEARNINGPROVISIONED
Path, courses, and labs
A signed-in account can hold an assessment and path. Full lessons and workbenches open only when the relevant training is provisioned.
AUTHORITY PRESERVED[ GUIDED PRODUCT TOUR ]NO ACCOUNT REQUIRED
See the machinery.
Keep authority closed.
Walk the actual lesson media, authored execution model, operating trial, and Evidence Chain that shape the protected learner system. The tour is interactive presentation state—not a shortcut around identity, enrollment, or review.
PRESENTER CONTRACTPUBLIC · READ-ONLY
- Tour stops
- 06
- FDE modules
- 08
- DEPLOY stages
- 06
- Video chapters
- 06
The lesson media and interactive instruments are real product surfaces backed by authored static contracts. They do not open protected learner data, run a live agent, create a session, or write to an account.
PUBLIC READ-ONLY TOUR · NO LEARNER RECORD, PROGRESS, REVIEW, ATTENDANCE, QUALIFICATION, OR CREDENTIAL EFFECT
Field standard
[ FDE-S/1.0 · authored ]
One mission method, six stage exits.
A forward-deployed engineer converts an ambiguous, consequential workflow into an adopted production system inside the customer's operating constraints, proves its behavior, transfers ownership, and returns reusable learning to the product without binding the solution to one model or interface.
Forward-Deployed Engineering Standard
Forward-Deployed Engineering
Diagnose the work. Establish the outcome contract. Prove the smallest wedge. Land inside the enterprise. Operate from evidence. Yield ownership and compound learning.
- VERSION
- 1.0.1
- BAND
- L2 → L4
- MODULES
- 8
- EVIDENCE GATES
- 10
- D01 / 06
Diagnose
How does the work actually move, especially where the documented process breaks?
PRIMARY EXIT SIGNALMission charter, stakeholder map, and observation boundary
- E02 / 06
Establish
What is the smallest safe wedge that can produce decision-grade evidence of workflow impact?
PRIMARY EXIT SIGNALOutcome contract and wedge decision record
- P03 / 06
Prove
Can the smallest wedge meet the mission contract under representative, edge, adversarial, and changing conditions?
PRIMARY EXIT SIGNALRunnable wedge and provider-neutral behavioral contract
- L04 / 06
Land
Can the proven wedge live inside enterprise identity, data, control, and operating boundaries without losing its evidence?
PRIMARY EXIT SIGNALProduction-shaped enterprise integration slice
- O05 / 06
Operate
Can users adopt the system while the team detects, contains, and learns from behavior in production?
PRIMARY EXIT SIGNALProduction-readiness and staged-rollout record
- Y06 / 06
Yield
Can the named owner operate, change, and recover the system—and what permitted learning should compound?
PRIMARY EXIT SIGNALOwner-accepted transfer package and owner-executed change/recovery record
Actual lesson
[ 1:30 · 6 chapters ]
Watch the real lesson asset, not a thumbnail.
The MP4, English captions, chapter track, transcript, and progression boundary are the same media surface used inside the provisioned lesson. Playback creates no completion event.
FIELD LESSON · ACTUAL MEDIA1:30 · CAPTIONS + CHAPTERS
Prompting as specification — a 90-second operator lesson
Prompts are programs in natural language. Write them with intent, context, constraints, and acceptance tests.
Open MP4 ↗1280 × 720 · H.264 / AAC
[ Chapter runway ]
- 01Vibes vs specs
- 02Control the input
- 03Constraints and output contract
- 04Failure behavior
- 05Evaluate the specification
- 06Produce evidence
WATCH → READ → BUILD → PROVE
Playback never completes this node. Continue through the full briefing, operating workshop, artifact submission, and published review rubric.
Read the exact 90-second transcript
01A prompt used once can survive on vibes because a human catches the weirdness.
02The same prompt inside a workflow needs a specification: explicit inputs, behavior, limits, and a way to fail safely.
03Start with context you control. Paste the source material, name the audience, and define the task with one clear verb.
04Never make the model invent the information that your workflow should supply.
05Turn expectations into numbered constraints, then define an output contract the next step can parse.
06A schema beats a pleasant paragraph because structure makes success testable and drift visible.
07Now specify failure behavior. If evidence is missing, say so. If the request crosses a policy boundary, stop.
08A production prompt needs a safe result for impossible work, not confident improvisation.
09Run the same prompt against a small test set: normal cases, edge cases, and one adversarial input.
10Compare results against the contract. Revise the specification, not just the single bad answer.
11Your evidence is the before prompt, the production specification, the test inputs, and the observed outputs.
12Submit that package for review. That is the loop: operate, verify, prove, then advance.
Runtime X-Ray
[ AUTHORED EXECUTION MODEL ]
Reveal what the terminal hides.
Step through the model proposal, control decision, machine action, and evidence return. Play the authored trace or branch into a denied action; neither control executes a command.
[ RUNTIME X-RAY · 101 ]L1 · N-L1-1
RUNTIME X-RAY · AUTHORED TRACE
One task becomes an accountable workflow
Move left to right through the work the screen compresses into a few status lines: inputs are gathered, context is shaped, a model proposes, checks hold the boundary, and a person authorizes the exit.
NO LIVE RUNTIME
Authored simulation — this is not a live trace. It does not inspect your terminal, editor, agent runtime, files, prompts, or learner activity.
INSTRUCTION / SYNTHETIC
READ-ONLY LEARNING INSTRUMENTBuild the weekly partner update from the approved metrics file. If any required metric is missing, stop before drafting. Never publish without my review.
THREE PLANES / ONE ACCOUNTABLE LOOP
MODEL → MACHINE · NO DIRECT PATHThe model proposes. The control plane scopes and authorizes. A tool or service executes, then returns inspectable evidence.
P-01
MODEL PLANE
PROPOSESWorkflow · Model
P-02
CONTROL PLANE
SCOPES + AUTHORIZESOperator
P-03
MACHINE + EVIDENCE
EXECUTES + RETURNSTools + runtime · Evidence
EXECUTION TRACE
Select a transition. Arrow keys move through the authored run.
Stage 1 of 8: Capture the trigger. You choose RUN UPDATE.
STAGE 01 · Operator
INSPECTING EVENTCapture the trigger
WHAT YOU SEE
You choose RUN UPDATE.
WHAT HAPPENS BEHIND THE GLASS
The workflow creates a run identifier and freezes the task contract for this execution.
INPUT BOUNDARY
Operator intent + approved workflow definition
OUTPUT / STATE CHANGE
Run envelope with named exit and stop condition
WHO OR WHAT HAS AUTHORITY
The operator starts this defined workflow; neither the model nor a tool can create its own run.
WHAT PROVES IT
Run id, actor, start time, workflow version
STAGE 02 · Tools + runtime
INSPECTING EVENTGather named inputs
WHAT YOU SEE
The console says “Reading approved metrics…”
WHAT HAPPENS BEHIND THE GLASS
A scoped file tool resolves only the declared source and returns content plus metadata.
INPUT BOUNDARY
Metrics file path + read permission
OUTPUT / STATE CHANGE
Versioned source bundle
WHO OR WHAT HAS AUTHORITY
The host grants one read against the named source; the model has no ambient file access.
WHAT PROVES IT
Source name, digest, modified time, read result
STAGE 03 · Workflow
INSPECTING EVENTShape the context
WHAT YOU SEE
A short “Preparing context” state appears.
WHAT HAPPENS BEHIND THE GLASS
The workflow checks required fields, removes irrelevant rows, and assembles instructions, evidence, and output schema.
INPUT BOUNDARY
Source bundle + prompt contract
OUTPUT / STATE CHANGE
Bounded model request
WHO OR WHAT HAS AUTHORITY
The workflow may transform approved inputs inside its contract, but it cannot expand the source boundary.
WHAT PROVES IT
Input checklist + context manifest
STAGE 04 · Model
INSPECTING EVENTGenerate a proposal
WHAT YOU SEE
The terminal shows “Drafting update…”
WHAT HAPPENS BEHIND THE GLASS
The model receives the bounded request and proposes structured content; it does not publish or fetch new data.
INPUT BOUNDARY
Instructions + approved evidence + schema
OUTPUT / STATE CHANGE
Draft update + structured fields
WHO OR WHAT HAS AUTHORITY
The model may propose structured content; it receives no authority to fetch, publish, or approve.
WHAT PROVES IT
Model and prompt version + response reference
STAGE 05 · Workflow
INSPECTING EVENTVerify the contract
WHAT YOU SEE
A CHECKING badge replaces the drafting state.
WHAT HAPPENS BEHIND THE GLASS
Deterministic checks validate the schema, required metrics, citations, and explicit failure behavior before the draft can move.
INPUT BOUNDARY
Draft + source bundle + acceptance rules
OUTPUT / STATE CHANGE
Pass or blocked result with reasons
WHO OR WHAT HAS AUTHORITY
Deterministic validators decide whether the contract passed; fluent model output cannot waive a failed check.
WHAT PROVES IT
Check results tied to the run id
STAGE 06 · Operator
INSPECTING EVENTHold at the human gate
WHAT YOU SEE
You see the draft, evidence, and APPROVE / REJECT controls.
WHAT HAPPENS BEHIND THE GLASS
The run is paused with its state intact; no external action is available until a named person decides.
INPUT BOUNDARY
Verified draft + check evidence
OUTPUT / STATE CHANGE
Recorded approval, rejection, or requested correction
WHO OR WHAT HAS AUTHORITY
A named reviewer owns approve, reject, or repair; the paused run cannot self-resume into an external action.
WHAT PROVES IT
Reviewer, decision, reason, timestamp
STAGE 07 · Tools + runtime
INSPECTING EVENTPerform the bounded action
WHAT YOU SEE
The approved update is saved to the review channel.
WHAT HAPPENS BEHIND THE GLASS
The action tool verifies approval and destination again, writes once with an idempotency key, and returns the external reference.
INPUT BOUNDARY
Approved draft + destination + approval receipt
OUTPUT / STATE CHANGE
Created channel draft; no unapproved send
WHO OR WHAT HAS AUTHORITY
The host permits one idempotent write only when the recorded approval and destination match this run.
WHAT PROVES IT
Action receipt + destination + idempotency key
STAGE 08 · Evidence
INSPECTING EVENTClose with evidence
WHAT YOU SEE
DONE links to the run log and created draft.
WHAT HAPPENS BEHIND THE GLASS
The workflow joins inputs, checks, human decision, and action receipt into one replayable record.
INPUT BOUNDARY
All run events and receipts
OUTPUT / STATE CHANGE
Run log ready for the second-input comparison
WHO OR WHAT HAS AUTHORITY
The evidence service can seal observed events, but it cannot rewrite the decisions or receipts it records.
WHAT PROVES IT
Complete evidence chain with terminal status
FAILURE REPLAY · RETURN TO DECISIONAT VERIFY-CONTRACT
- SYMPTOM
- The draft fluently explains a sharp decline even though one metric row never arrived.
- HIDDEN CAUSE
- The source read succeeded, but completeness was not checked before generation; fluent output hid a partial input.
- CONTAINMENT
- The required-metric check blocks the human gate and the action tool remains unreachable.
- HUMAN DECISION
- Repair the source, replay from gather, and compare both run logs before changing the gate.
[ CURRENT KNOWLEDGE RUNWAY ]
Go from mental model to field command.
Maintained specifications, official runtime guides, and open-source implementation references, checked through 2026-09-06. Each entry records its currency signal and next recheck deadline; links are learning references, not claims about this platform's runtime.
- OPEN SOURCE· MIT (software)Running agents (opens in a new tab)
OpenAI Agents SDK · Follow the work between a user prompt, model turn, tool result, and final output instead of treating the terminal spinner as the system.
The current TypeScript SDK guide covers the run loop, turn boundaries, state, streaming, and resumable execution.
REVIEWED 2026-08-27 · RECHECK 2026-11-27
- OPEN SOURCE· MIT (software)Human in the loop (opens in a new tab)
OpenAI Agents SDK · Place explicit human authority before consequential tools, preserve paused state, and make rejection a designed path.
The current SDK supports interruptions, approval decisions, rejection, and resuming a serialized run state.
REVIEWED 2026-08-27 · RECHECK 2026-11-27
- OFFICIAL REFERENCEAutomate workflows with hooks (opens in a new tab)
Anthropic Claude Code · Map the invisible event lifecycle around a visible tool call and identify where deterministic checks can run before or after agent action.
The current lifecycle exposes session, prompt, pre-tool, permission, post-tool, subagent, stop, task, and compaction events across terminal, IDE, desktop, and web surfaces.
REVIEWED 2026-08-27 · RECHECK 2026-11-27
Operator Trial
[ PRACTICE-ONLY · NO ACCOUNT WRITE ]
Make one consequential decision.
Choose a role target and respond to a release trace that looks healthy in aggregate but fails a required case. The interpretation explains the evidence boundary; it does not score you.
AUTHORED SPECIMEN · operator-trial/v1IDENTITY NOT REQUESTED · NO ACCOUNT WRITE
- 01ROLESELECTED
- 02DECISIONOPEN
- 03INTERPRETPENDING
- 04RECORDNOT CREATED
This authored specimen interprets one selected decision against the published BME-09 control sequence. It is not a capability assessment, learner record, credential, hiring signal, or evidence that the workflow was executed. A linkable view carries only the selected role and decision IDs in its URL; it does not write them to an account.
Evidence Chain
[ ASSESSMENT → GRAPH → WORK → REVIEW → RECORD ]
Follow the claim back to its evidence.
The public Evidence Chain uses current assessment, graph, field-exercise, review, and registry contracts. Protected stages link to their public explanations here, never to a learner record.
EVIDENCE CHAIN · SYSTEM VIEW
The complete static evidence model is active. Three-dimensional enhancement is available.INPUT · WORK SIGNALOUTPUT · PUBLIC RECORD
BME-01 · n-l3-59 NODE ROUTE
03SELECTED PLANE
Public packet briefProvisioned workbench
Hold or ship the support agent
BME-01 maps to n-l3-5 and publishes 4 synthetic packet sources with 4 weighted rubric dimensions.
4rubric dimensions
- 01 · assessment
Capability assessment method
Open public detail → - 02 · graph
Authored skill graph
Open public detail → - 03 · benchmark
Hold or ship the support agent
Open public detail → - 04 · review
Artifact review
Open public detail → - 05 · registry
Certified AI Native Developer
Open public detail →
ISOLATED SPECIMEN · NOT CUSTOMER OR LEARNER DATAPublic · read-only
[ ENTERPRISE GOLDEN THREAD ]PRESENTER-RUN SPECIMEN
One operating story. Every boundary stays in frame.
Trace how an enterprise operator reads aggregate formation state, hands the camera to a practitioner, and inspects a bounded field lab—without borrowing a customer identity or fabricating a credential, receipt, or outcome.
Each frame mirrors a shipped interface contract. The values below are an authored presentation, not production Store rows.
A facilitator cannot follow a roster row into a learner’s journey, Atlas, artifact, or review. The presenter changes lenses here. No account is impersonated.
PRESENTER TIMECODE
00:00 → 10:00Ten minutes, sequenced for evidence.
- T+00TRUTH
Perspective · Presenter
Establish the specimen boundary
State that every value is an authored presentation, never customer or learner data.
- T+01OPERATIONS
Perspective · Enterprise operator
Read the program state
Confirm one anonymous sample seat, its recorded sample baseline, and the reporting boundary.
- T+03HANDOFF
Perspective · Presenter
Change lenses without impersonation
Move from aggregate operations to the anonymous specimen without opening a roster row or learner account.
- T+04JOURNEY
Perspective · Anonymous specimen
Follow the evidence route
Trace the sample from activation through a required evidence revision without entering a learner account.
- T+06ATLAS
Perspective · Capability lead
Inspect the FDE progression
Read all eight authored modules and the prerequisite graph behind the sample state.
- T+08LAB
Perspective · Practice evaluator
Open the lab standard
Inspect the six-stage BME-10 practice loop and the evidence expected at every stage.
- T+09CONTINUE
Perspective · Enterprise sponsor
Choose the next public path
Continue to the public trial, lab catalog, or a sales-assisted enterprise conversation.
OPERATIONS SPECIMEN
SPECIMEN-01Read aggregate state before opening a story.
These four bounded states establish what an enterprise operator can discuss without opening a person-level record.
- Seat posture
- Active sample seat active sample
- Baseline
- Baseline recorded · sample recorded sample
- Reporting
- Aggregate reporting remains private aggregate private
- Outcome posture
- Individual outcomes withheld withheld
LEARNER JOURNEY · PRESENTATION STATE
Follow the evidence, not a borrowed account.
Five authored waypoints explain how practice becomes reviewable. They display specimen state only and cannot advance a graph, create a receipt, or grant a credential.
- 01Active sampleSEAT
Sample seat active
The specimen begins with one explicitly labeled sample seat, not an enrollment claim.
- 02Recorded sampleBASELINE
Sample baseline recorded
A baseline state is shown as recorded without exposing answers, dimension scores, or an assessment record.
- 03Route assignedROUTE
FDE route assigned
The specimen follows the current eight-module authored FDE route and its real graph joins.
- 04Revision requiredREVISION
Evidence revision required
The third module remains open for revision; the public view does not invent a score or reviewer decision.
- 05Not createdRECEIPT
Practice receipt not created
BME-10 remains an inspectable practice contract. No artifact, review, qualification, credential, or receipt is created here.
[ FDE CAPABILITY ATLAS · EIGHT MODULES ]
Two complete. One in repair. Five stay closed.
Statuses describe SPECIMEN-01 only. They do not alter curriculum progress, unlock a module, or establish FDE readiness.
- Nodes
- 42
- Edges
- 48
- Levels
- 5
- FDE modules
- 8
- Sample complete94 minFDE-01
Diagnose the work, establish the outcome
Sample evidence is shown complete in this isolated view.
Sample complete - Sample complete87 minFDE-02
Scope and architect the mission
Sample evidence is shown complete in this isolated view.
Sample complete - Revision required103 minFDE-03
Land inside enterprise systems
The specimen must revise its evidence before advancing.
Revision required - Locked · inert109 minFDE-04
Engineer provider-portable AI systems
Locked until the preceding sample evidence clears.
PREREQUISITES REQUIRED · INERT - Locked · inert114 minFDE-05
Operate an evaluation and reliability control system
Locked until the preceding sample evidence clears.
PREREQUISITES REQUIRED · INERT - Locked · inert102 minFDE-06
Bound authority and earn authorization
Locked until the preceding sample evidence clears.
PREREQUISITES REQUIRED · INERT - Locked · inert118 minFDE-07
Operate adoption, rollout, and incidents
Locked until the preceding sample evidence clears.
PREREQUISITES REQUIRED · INERT - Locked · inert102 minFDE-08
Yield ownership and compound field learning
Locked until the preceding sample evidence clears.
PREREQUISITES REQUIRED · INERT
[ FIELD LAB SPECIMEN ]BME-10 · V1.1.0
Run the forward-deployed field loop
The public tour exposes only safe authored stage labels and evidence descriptions. It does not expose the runtime packet, learner response, evaluator material, artifact, or receipt body.
- Lab window
- 75min
- Evidence mode
- practice-only
- 01discover
Discover
EVIDENCE PRACTICEDWorkflow map plus baseline/value hypothesis with stakeholder, constraint, and unknown boundaries.
- 02frame
Frame
EVIDENCE PRACTICEDArchitecture decision record, acceptance gates, two-week sequence, and named responsibility map.
- 03build
Build
EVIDENCE PRACTICEDServer-bound authorization, citation checks, bounded retries, protected traces, and flagged rollback.
- 04deploy
Deploy
EVIDENCE PRACTICEDA HOLD, repaired candidate, current gate reruns, 24-user read-only canary, and rollback authority.
- 05recover
Recover
EVIDENCE PRACTICEDInspection, pause, preservation, safe-scope language, tombstone repair, and affected-gate reruns.
- 06handoff
Handoff
EVIDENCE PRACTICEDRunbook, SLO, kill switch, owners, evidence location, teach-back, adoption backlog, and reusable field signal.
LAB RECEIPTnot created · presentation only
Protected system
[ AUTHORITY MAP · NO PRIVATE PROJECTION ]
The tour ends where identity begins.
This final map shows how the implemented learning, evidence, formation, session, support, and operating contracts connect. It deliberately contains no person, cohort, scheduled offering, post, seat, or artifact.
EVIDENCEOWNER BOUND
Owned artifacts and revision
Submission, review history, and portfolio state stay attached to the owner. The public tour receives none of those records.
AUTHORITY PRESERVEDFORMATIONENROLLMENT BOUND
Cohort and working board
Private rooms resolve from active enrollment. Commons, cohort posts, replies, and evidence attachments retain their existing server authority.
AUTHORITY PRESERVEDSESSIONSSEAT BOUND
Preparation and calendar handoff
The implemented flow binds seat authority, canonical prework, a provider-neutral calendar handoff, an owner-bound brief, and an authorized board. No scheduled offering or participant state appears here.
AUTHORITY PRESERVEDSUPPORTACTIVATION GATED
AI mentor and human boundary
AI concept, specification, and next-step support is separate from human review or expert intake, which remains fail-closed until its operating coverage exists.
AUTHORITY PRESERVEDOPERATIONSROLE GATED
Tenant and decision controls
Enrollment, moderation, organizational, and review decisions remain role-gated and actor-bound. A tour interaction grants no authority.
AUTHORITY PRESERVEDPRESENTER CONTINUATION · PUBLIC
Choose the next lens.
Continue as an individual practitioner, an organization building an internal cohort, or a diligence reader inspecting the evidence. Identity remains required only for protected learner work. This system map describes implemented and gated product contracts. It contains no learner, tenant, enrollment, artifact, community, session, attendance, staffing, review, or credential record.