/
FDEEverything we know. Built into your team.
Your work. Yours to run.

Through real-world projects in the field, we build the FDE way into your team: Being — Mindset, Doing — Skillset, Wielding — Toolset.

Not a cost. An asset that stays.

SECI 2.0

AIは、知の螺旋を速く回す。 AI SPINS THE KNOWLEDGE SPIRAL FASTER

Socialization
Externalization
Internalization
Combination

Explicit Tacit

01

Socialization

共同化 Tacit → Tacit · Originating Ba

経験を共にし、身体で写しとる段階。知は言葉になる前に、人から人へ移る。 Knowledge moves body to body, before it ever becomes words — through shared experience.

Practice従来Then

現場同行・OJT・徒弟的な観察と模倣Shadowing, OJT, apprentice-style observation and imitation

同じ釜の飯——雑談・非公式な交流Informal exchange: corridor talk, shared meals

その場にいた者だけが受け取れるReceived only by those who were in the room

With AIいまNow

会議・現場の音声を常時記録し、即座に文字へMeetings and field audio captured continuously, transcribed at once

作業ログ・画面操作を自動でキャプチャWork logs and on-screen actions captured automatically

「その場」が記録になり、共体験が全員に開くThe room becomes a record; the shared experience opens to everyone

02

Externalization

表出化 Tacit → Explicit · Dialoguing Ba

暗黙知を、対話とメタファーによって言葉・図・概念に変換する段階。SECIの要にして、最も難しい変換。 Converting tacit knowing into words, diagrams and concepts through dialogue and metaphor — the pivot of SECI, and its hardest conversion.

Practice従来Then

対話・メタファー・ホワイトボードでの図解Dialogue, metaphor, whiteboard sketching

熟練者へのインタビューと観察記録Expert interviews and field notes

言語化の質が、聞き手の力量に依存するThe quality of articulation depends on the interviewer

With AIいまNow

LLMとの壁打ちが、問い続ける聞き手になるSparring with an LLM — an interviewer that never stops asking

対話ログから、構造化ドキュメントを自動生成Structured documents generated from the dialogue itself

熟練者の判断基準を、ルールとチェックリストへExpert judgment distilled into rules and checklists

03

Combination

連結化 Explicit → Explicit · Systemizing Ba

形式知同士を編集・分類・接続し、組織の知識体系へと組み上げる段階。 Editing, sorting and connecting explicit knowledge into an organizational body of knowledge.

Practice従来Then

マニュアル・報告書の編纂と改訂Compiling and revising manuals and reports

会議体での集約、フォルダでの文書管理Aggregation through meetings; folder-based document control

探せない・古い・版が割れるUnfindable, outdated, forked into versions

With AIいまNow

ナレッジ基盤が文書同士を自動で連結A knowledge platform links documents to each other automatically

横断検索と要約が、必要な瞬間に届くCross-search and summary arrive the instant they are needed

更新は全員に同期し、版の分裂が消えるUpdates sync to everyone; version drift disappears

04

Internalization

内面化 Explicit → Tacit · Exercising Ba

形式知を実践のなかで使い込み、ふたたび身体の知へと還す段階。次の螺旋が、ここから始まる。 Using explicit knowledge in practice until it becomes embodied skill again. The next turn of the spiral begins here.

Practice従来Then

マニュアルを読む・集合研修Reading manuals; classroom training

反復訓練と、実地の試行錯誤Repetitive drills; trial and error on the job

習得は本人の粘りに委ねられるMastery left to individual persistence

With AIいまNow

AIを相棒に、実務のなかで反復するRepetition inside real work, with an AI partner

その場でフィードバックと矯正が入るFeedback and correction arrive on the spot

習得までの試行回数を、圧縮するFewer trials between knowing and mastery

一周ごとに、知は資産になる。 With each turn, knowing becomes an asset.

BEING

Sworn FDE Fellow

The stance a sworn FELLOW carries onto any floor — bound by the CODE.

The Program · In-house Enablement

“So — how exactly do
our people get trained?”

Three layers of training. Five levels of assessment. And the verdict on the last line of the report card — PROVED IT: your team runs it alone. We train them to that point. Then we leave.

Section 00 — The Whole Picture

The Training Map

Your operator grows through three layers, is measured on five levels, and graduates at PROVED IT.

×9
MINDSET
Stances

The working posture that holds on any floor. Nine stances, drilled to observable behavior.

5×20
SKILLSET
Capabilities

Five domains, twenty units. Every unit carries written passing bars for HAS DONE and PROVED IT.

140
TOOLSET
The Map

140 technical concepts, independent of any product. Products get replaced; the map does not go stale.

LV.1
KNOWS
LV.2
CAN DO
LV.3
HAS DONE
LV.4
CHANGED IT
LV.5
PROVED IT

PROVED IT = self-running confirmed = in-housing complete — this is where we exit.

Section 01 — Mindset

The Nine Stances

We transfer the stance before the tools. Each stance has an observable behavioral indicator — assessed by behavior, never by self-report.

M-01Ambiguity tolerance

Moves before requirements are final. "It isn’t decided yet" is never a reason to stop.

On the floorSplits incomplete instructions into what to confirm and what to start — then starts.
How we train itExercises with deliberately missing specs; we grade how the gaps were filled.
M-02Ownership

Sits in the driver’s seat of their own work. An operator, not a vendor supervisor.

On the floorSays "I’ll decide the spec," not "I’ll ask the vendor."
How we train itOne full assignment owned end-to-end under their own name. We never take over midway.
M-03So what?

Attaches "so here is what we should do" to every report and analysis. No fact-dumping.

On the floorThe first line of every report is a conclusion. Answers "so what?" on the spot.
How we train itWeekly reports in conclusion-first format; reports without a conclusion are returned.
M-04Hypothesis first

Places a tentative answer before researching, then breaks it with tests.

On the floorDeclares at the start: "The bottleneck is probably X — I’ll verify that first."
How we train itResearch tasks require a written hypothesis memo before work begins.
M-05Small and fast

A working prototype beats a perfect plan. One process first, one week first.

On the floorShows something working in week one. Prototypes arrive before slide decks.
How we train itFixed weekly prototype deadlines; showing unfinished work is mandatory.
M-06Facts, not feelings

Judges by numbers, logs, and measurements — never by "it feels better now."

On the floorEvery claim of impact comes with measured values: time, volume, error rate.
How we train itPre-implementation measurement is a required step; unmeasured claims are not accepted.
M-07Boundary crossing

Crosses between tech, operations, and management on their own initiative.

On the floorWalks to other departments to hear their problems firsthand.
How we train itInterview drills run in other departments; must bring back one problem from across the border.
M-08Always learning

The technology map is redrawn every year. We install how to learn, not just what to know.

On the floorHas an established routine for handling a concept they’ve never seen.
How we train itRepeated drills: given an unknown concept, reach KNOWS within 30 minutes.
M-09Letting go

Builds systems others can run. Never counts key-person dependence as an achievement.

On the floorTheir manuals work in their absence. "Only I can touch it" does not exist.
How we train itHandover assignments, graded by the successor’s operational success.
Section 02 — Skillset

Capabilities — 5 Domains, 20 Units

Every unit carries a double bar. HAS DONE = applied in real work with us alongside. PROVED IT = reproduced without us. Only the latter counts as in-housing.

S1Technical Execution
TECHNICAL EXECUTION

Decompose the work, prototype it, grow it into production. The hands-on core of in-housing.

S1-1Process decomposition
  • Splitting any process into seven elements: input, judgment, transfer, check, notification, approval, exception
  • Writing decomposition tables; telling automatable elements from those humans must keep
  • The order of operations: automation starts from the table, not from picking tools
HAS DONE ── LV.3Decomposed one process from their own department and identified the automation targets.
PROVED IT ── LV.5Given a new process, runs decomposition through target identification alone, unassisted.
S1-2Prototyping
  • The shortest path from idea to a working prototype with AI; validation speed over polish
  • Running the short loop: build → show the floor → fix
  • What a prototype must verify vs. what can wait for production
HAS DONE ── LV.3Turned one decomposed process into a working prototype within one to two weeks.
PROVED IT ── LV.5Carries a prototype from proposal to floor deployment alone.
S1-3Data flow design
  • Diagramming where data is born, where it pools, and who uses it
  • Structured vs. unstructured data (documents, images, audio) and how they differ
  • Drawing the line on personal and confidential data; what may be handed to AI
HAS DONE ── LV.3Drew the data-flow diagram for the target process and flagged every sensitive point.
PROVED IT ── LV.5Designs new systems with no blind spots in the data dimension, alone.
S1-4Productionizing
  • What separates a prototype from production: permissions, exceptions, records, backups
  • Where to place human approval — the checkpoints that must never be unmanned
  • Reversible design: decide how to stop before you start
HAS DONE ── LV.3Took one prototype into production with approval checkpoints and logging.
PROVED IT ── LV.5Detects, handles, and prevents recurrence of failures and exceptions without support.
S2Business Acumen
BUSINESS ACUMEN

Choosing what should be done, not what could be done. Thinking that survives an investment decision.

S2-1Impact estimation
  • Converting impact to money: hours saved × labor cost + quality gains
  • Comparing against build-and-run costs; calculating payback periods
  • The discipline of saying no: shelving work whose impact never clears its cost
HAS DONE ── LV.3Estimated the automation impact of a target process in money and recommended go / no-go.
PROVED IT ── LV.5Produces estimates precise enough for management investment decisions, case by case, alone.
S2-2Hypothesis thinking
  • Placing a tentative answer first, then closing it out with the smallest possible test
  • The fast loop: hypothesize → break it → rebuild it
  • Cutting off the time thief called "exhaustive research first"
HAS DONE ── LV.3Led with a hypothesis and reached a verdict through minimal verification.
PROVED IT ── LV.5Reaches answers alone on unknown problems, starting from hypotheses.
S2-3Issue decomposition
  • Splitting big questions into answerable sizes
  • Checking for gaps and overlaps; knowing when to stop splitting
  • Prioritizing the pieces: impact × feasibility
HAS DONE ── LV.3Turned a vague directive like "do something with AI" into an actionable issue list.
PROVED IT ── LV.5Structures ambiguous management problems all the way into execution plans, alone.
S2-4Business context
  • Setting priorities from how the company earns and what the department exists for
  • Translating between the floor’s reality and management’s reality
  • Never confusing "technically interesting" with "good for the business"
HAS DONE ── LV.3Re-ordered initiatives by business impact and explained why.
PROVED IT ── LV.5Re-selects initiatives on their own judgment when management direction shifts.
S3Communication
COMMUNICATION & INFLUENCE

A system only becomes a system when people use it. The language skills that move an organization.

S3-1Translation
  • Writing the same content three ways: one minute for executives, a manual for the floor, a spec for engineers
  • Explaining technology without jargon
  • Re-ordering content by what each audience needs to hear first
HAS DONE ── LV.3Explained one initiative to three different audiences, each in their own language.
PROVED IT ── LV.5Communicates with every layer of the company without an interpreter.
S3-2Conclusion-first reporting
  • The format: conclusion → grounds → request
  • The rule: the worse the news, the faster and shorter the report
  • Replacing "still under review" with "I will decide X by date Y"
HAS DONE ── LV.3Built an executive report on one page, conclusion first.
PROVED IT ── LV.5Carries a proposal through an executive meeting — Q&A included — alone.
S3-3Winning people over
  • Drawing out the real reason behind a colleague’s resistance
  • Experience beats persuasion: staging moments where skeptics touch the system
  • Building small wins first, and sequencing how they are seen
HAS DONE ── LV.3Turned one skeptical floor member into a user.
PROVED IT ── LV.5Initiates expansion into new departments, flag in hand.
S3-4Storytelling
  • One page of before / after in numbers and scenes
  • Reusing it for internal comms, executive reports, floor briefings
  • Telling results as the team’s story, not personal credit
HAS DONE ── LV.3Packaged the results into a one-page story and shared it internally.
PROVED IT ── LV.5Has become the narrator who can tell the company’s story externally as well.
S4Delivery & Impact
DELIVERY & IMPACT

Carry every initiative to the point where the numbers move. No projects that quietly die halfway.

S4-1Scope management
  • Declaring "what we will not do this time" before starting
  • Receiving — and declining — the requests that swell mid-project
  • Finishing small before moving on
HAS DONE ── LV.3Declared an out-of-scope list at kickoff and declined at least one mid-project expansion.
PROVED IT ── LV.5Runs multiple projects in parallel without the scope ever unfolding.
S4-2The weekly loop
  • Operating the weekly cycle: build → show → fix
  • Disarming perfectionism: show at 80%, fix the last 20% with the floor
  • Setting one’s own weekly review — and protecting it
HAS DONE ── LV.3Held the weekly show-your-work rhythm for four straight weeks.
PROVED IT ── LV.5The rhythm survives as a team habit after our support ends.
S4-3Risk & checkpoints
  • Knowing what must never be unmanned: payments, contracts, HR, external communications
  • Where to put human approval; how exceptions get escalated
  • Reversible design: writing the accident scenario before it happens
HAS DONE ── LV.3Mapped the danger points of their own project and designed the approval checkpoints.
PROVED IT ── LV.5Smells danger points on new projects and places checkpoints preemptively.
S4-4Impact measurement
  • The iron rule: define the metrics before implementation (no retrofitting)
  • Collecting hours saved, error rates, volumes, lead times
  • Connecting measurement → reporting → the next improvement, year-round
HAS DONE ── LV.3Reported impact using measured before / after data.
PROVED IT ── LV.5Runs the annual cycle of measure → report → propose improvements, alone.
S5Knowledge Transfer
KNOWLEDGE TRANSFER

The moment knowledge becomes one person’s property, in-housing dies. Making systems the organization’s property.

S5-1Writing manuals
  • Writing what you did at a granularity someone else can reproduce
  • The readable manual: purpose, prerequisites, steps, pitfalls, contact
  • Keeping manuals alive: update dates, owners, review cycles
HAS DONE ── LV.3Wrote the manual for their own system; someone else operated it from the manual alone.
PROVED IT ── LV.5The team’s manuals are maintained with zero key-person dependence.
S5-2Teaching
  • The three steps: show → do together → watch over
  • Auditing "I taught them" — measured by whether they did it alone
  • Building a relationship where questions come easily
HAS DONE ── LV.3Handed over to one colleague and stayed until that person could run it alone.
PROVED IT ── LV.5Now the teacher — raises the next operator to self-running.
S5-3The improvement circuit
  • Collecting complaints, exceptions, and near-misses — through channels people actually use
  • Keeping an improvement backlog; sorting do from don’t
  • The judgment to retire systems no longer used
HAS DONE ── LV.3Implemented one improvement from floor feedback and verified the effect.
PROVED IT ── LV.5The improvement cycle keeps turning without anyone asking.
S5-4The system ledger
  • Keeping the ledger: what runs where, under whose responsibility
  • Maintaining change logs, naming conventions, owner tables
  • Structurally preventing "only the builder can fix it"
HAS DONE ── LV.3Built the ledger (what, where, who) for their own project.
PROVED IT ── LV.5Audits, handovers, and incidents can be traced from the ledger alone.
Section 03 — Toolset

The Map — 140 Concepts

A map of products goes stale in a year; a map of concepts does not. We hand over 140 technical concepts as the map — the map is shared, the vehicle is yours.

LLM
SLM
MoE
Multimodal
Reasoning models
Distillation
Quantization
Speculative decoding
Batching
Streaming
KV cache
Model routing
Cascade
Fallback
RAG
GraphRAG
Agentic RAG
Hybrid search
Vector search (ANN/HNSW)
BM25
Embeddings
Chunking
Reranking
Query rewriting
Semantic search
Long/short-term memory
Semantic cache
Context compression
Agent loop
ReAct
Plan-and-Execute
Multi-agent
Handoff
Swarm
Blackboard
State machine
Scratchpad
Reflexion
Tree-of-Thought (ToT)
HITL
Tool-use
Function calling
MCP
A2A
Computer-use
Browser agent
Sandboxed execution
API/Webhook integration
Workflow/DAG
Chain-of-Thought
Few-shot/Zero-shot
System prompt design
Context engineering
Token budget
Structured output (JSON/schema)
Prompt cache
Prompt compression
Fine-tuning
LoRA/QLoRA
Continued pretraining
Instruction tuning
RLHF
RLAIF
DPO
Preference optimization
Synthetic data generation
Data curation
Eval
Golden set
LLM-as-judge
Rubric scoring
Regression eval
Offline/online eval
A/B testing
Hallucination detection
Grounding
Citations
Faithfulness
Coverage
Semantic guardrails
I/O filtering
PII redaction
Prompt-injection defense
Jailbreak resistance
Policy enforcement
Approval flow
Red-teaming
Tracing
Observability
Logs/metrics
Cost/token metering
Latency monitoring
Drift detection
Canary/shadow deploy
Versioning
Replay
ETL/ELT
CDC/sync
Data contracts
Lineage
Feature store
Document parsing (OCR/layout)
Structured extraction
Entity resolution
Knowledge graph
Ontology
Semantic layer
Taxonomy
Metadata
Triple store
Image generation
Speech (ASR/TTS)
Video
Diffusion models
Realtime voice agents
Next Best Action
Recommendation
Scoring
Anomaly detection
Early-warning detection
Simulation
Optimization
F
D
E
F
E
L
L
O
W
S
A
R
T
&
S
C
I
E
N
C
E
!
RowDomainThe question this row answers
01ModelsWhat does the thinking — model types, compression, serving
02Retrieval & knowledgeWhat grounds the answers — connecting and searching internal knowledge
03Memory & orchestrationHow it remembers, how agents are organized
04ExecutionHow it moves its hands — tool calls, safe runtimes
05Instruction & tuningHow to instruct it, how to condition it
06Training & evaluationHow to teach it, how to grade it
07Quality & safetyHow errors are caught, data protected, attacks resisted
08Governance & operationsHow to watch it, meter it, roll it back
09Data foundationWhat it is fed — pipelines, document parsing, meaning
10Generation & applicationsHow far it extends — image, voice, video, prediction, detection

Not a memorize-all-140 course. Concepts that matter to your target go deep (to CAN DO); the rest are held as a map (to KNOWS). The engraving on the bottom rows is our crest.

Section 04 — Assessment

Assessment — Five Levels

“Attended the training” is not an assessment. We measure one thing: whether they became able. Each level has a defined method and passing bar.

LevelWhat it meansHow it is measuredPassing bar
LV.1KNOWSCan explain the concept in their own words.Comprehension tests after each unit, graded instantly.Meets the passing score on core-concept questions.
LV.2CAN DOHands actually move in a practice environment.Submitted practicum, graded against a criteria sheet.The deliverable meets every item on the sheet.
LV.3HAS DONEUsed it in real work — with us running alongside.Field assessment — one real case entering operation.One real process actually entered operation.
LV.4CHANGED ITThe numbers moved. Impact was measured.Post-implementation measurement of pre-defined metrics.Measured improvement in hours, error rate, or volume.
LV.5PROVED ITIt kept running — without us.The absence test — solo operation with no support, plus improvement.A set period of solo operation (exceptions included) + one self-initiated improvement.

In driving terms — HAS DONE means they drove on real roads with an instructor in the passenger seat. If the instructor gets out now, there will be an accident.

PROVED IT means they come home safely, every time, with no instructor. We only leave in the second case.

Section 05 — The Report Card

The Report Card

Results are not delivered verbally. Every operator receives this card. When the final verdict reads PROVED IT, in-housing is complete. Below is a specimen.

Field Report Card ── In-House Operator
In-House Operator Report Card
SPECIMEN
I. MINDSET — the nine stances (behavioral)
M-01
Ambiguity tolerance
●
M-02
Ownership
●
M-03
So what?
●
M-04
Hypothesis first
●
M-05
Small and fast
●
M-06
Facts, not feelings
●
M-07
Boundary crossing
●
M-08
Always learning
●
M-09
Letting go
●

● = confirmed by recorded behavioral evidence (never self-reported)

II. SKILLSET — 5 domains × 5 levels
DomainLV.1
KNOWS		LV.2
CAN DO		LV.3
HAS DONE		LV.4
CHANGED IT		LV.5
PROVED IT
S1 Technical Execution●●●●●
S2 Business Acumen●●●●●
S3 Communication & Influence●●●●●
S4 Delivery & Impact●●●●●
S5 Knowledge Transfer●●●●●
V. Final Verdict

Cleared the __-week absence test. Handled __ exceptions solo. Implemented __ self-initiated improvements. We judge this operator able to run and improve the target process alone.

PROVED
SELF-RUNNING CONFIRMED

※ Specimen. Verdicts rest solely on measurement and field assessment, by two assessors. Study volume has no influence.

Section 06 — How It Unfolds

How We Train — Three Acts

Run together, put into words, let go — people grow in that order only. Standard duration: three to six months.

ACT 1 · M1 – M2
Run together

We enter your floor and run the target process together. Your operator watches from the next seat, takes parts, and gradually takes the wheel. What never transfers through slides — the breathing of judgment, the nose for danger — transfers only here.

KNOWS → CAN DO
ACT 2 · M3 – M4
Into words, into systems

What we did together becomes decomposition tables, manuals, and training material written for your company. Here the roles reverse — your operator leads, we assist. Only what they can write in their own words remains as an organizational asset.

HAS DONE → CHANGED IT
ACT 3 · M5 – M6
Let go and prove

We move to the back seat, then get out of the car. The absence test — a period with no support, your operator alone running operations, exceptions, improvements. We confirm with numbers that it kept running. And we leave.

PROVED IT
RhythmWhat happens
Daily, 30 minSelf-study with video units and quizzes. Concept input is asynchronous — it never steals floor time.
Weekly, 90 minSide-by-side session. Your real work is the teaching material; we clear that week’s wall together.
MonthlyLevel verdict. Progress against the criteria sheets is checked and recorded on the report card.

Our job is not to remain needed.
It is to become unneeded.

Train. Transfer. Vanish.
Train. Transfer. Vanish.
Back to FDE