THE STUDIO / TRAINING SYSTEM
Enterprise AI Training · Banking & Regulated Industries
THE STUDIO/SEVEN MODULES/SEVEN DRILLS/ONE CHAIN

TRAIN LIKE
THE EXAMINER
IS COMING.

Because the examiner is coming. A complete operating discipline for working with AI in institutions where the work must survive scrutiny — seven core modules that teach the craft, seven scored Live Fire drills that prove it, and every drill run sealed on a tamper-evident hash chain. Even the training leaves evidence.

07Core Modules
07Live Fire Drills — ALL LIVE
SHA-256Evidence Invariant
SR 11-7Challenge Standard
AUG 2026EU AI Act Clock
THE OVERTUREBEFORE YOU BEGIN · 2 MIN
BEGIN THE STRAND — LF-01 THE REPAIR SEE THE CURRICULUM
THE DOCTRINE
Every discipline in this system exists because somewhere, in a real institution, its absence cost real money, real careers, or real people. We train against the documented record — not hypotheticals. The Studio — Field Doctrine
TEACH — THE MODULES

Seven cinematic modules build the operating discipline: instruction, verification, evidence, delegation, provenance, adversarial challenge, and institutional governance. Each is a single-file interactive experience with guided narration.

TEST — LIVE FIRE

Each module has a scored companion drill. No lectures — live exhibits with planted flaws, conviction staked on every call, ratings issued on the record. The drills are where the skill becomes reflex.

PROVE — THE CHAIN

Every drill run is hash-chained in the browser under the same append-only invariant used for agent audit trails, and exports a verifiable evidence pack. Training that emits provenance — the record of the record.

THE CORE CURRICULUM/MODULES 01–07

Seven disciplines.
Each one paid for in the field.

The arc runs from individual craft to institutional accountability. Every module carries a field case — a documented failure that is the reason the discipline exists.

MODULE 01INSTRUCTION DISCIPLINE

THE BRIEF

You learn to instruct a model with precision: the 8-element anatomy of a professional prompt. Not tips and tricks — a repeatable, auditable method. Because the prompt is the first line of your audit trail, and an ungoverned instruction is an ungoverned system.

Field Case — DPD, January 2024After a system update weakened its guardrails, the parcel carrier's customer-service chatbot was coaxed into swearing at a customer and composing a poem about how useless its own company was. The screenshots went viral within hours and DPD disabled the AI layer. The instruction layer is a control. Treat it like one.
▶ RUN LIVE FIRE 01 — THE REPAIR FACILITATOR MODE
MODULE 02VERIFICATION

THE COUNTERSIGN

The model drafts. You verify. Hallucination detection, confident arithmetic failures, staleness, sycophancy — the four ways AI output fails in production. You build a three-tier verification stack aligned to SR 11-7's effective challenge requirements, then prove it under fire in THE OVERRIDE.

Field Case — Mata v. Avianca, 2023 · Deloitte Australia, 2025Two New York lawyers were sanctioned after filing a brief containing six fabricated cases generated by ChatGPT — they had asked the model to verify itself. Two years later, Deloitte partially refunded the Australian government after a AU$440,000 report was found to contain invented academic references and a fabricated court quotation. Different decade of AI, same missing countersign.
▶ RUN LIVE FIRE 02 — THE OVERRIDE FACILITATOR MODE
MODULE 03EVIDENCE

THE RECORD

Memory is not a record. When the examiner arrives in eighteen months, the chat will be gone, the analyst will have moved on, and the memo cannot speak for itself. This module teaches you to build work that explains itself — reproducible, provenanced, audit-ready.

Field Case — Wall Street Off-Channel Sweep, 2021–2024The SEC and CFTC levied more than $2 billion in penalties across the industry's largest banks for business conducted over unpreserved channels. The work happened; the record didn't survive — and regulators priced the difference in the billions. Samsung learned the inverse lesson in 2023, banning external generative AI after engineers pasted proprietary source code into ChatGPT. Work that leaves the record, and records that leave the work, cost the same.
▶ RUN LIVE FIRE 03 — THE RECONSTRUCTION FACILITATOR MODE
MODULE 04DELEGATION

THE OPERATOR

From chat to agency. You learn to delegate multi-step work to an AI agent, manage token economics with a live cost model, and route tasks across models by capability and budget. This is where you stop being a user and start being an operator — and where operator error stops being personal and starts being institutional.

Field Case — Knight Capital, August 1, 2012A stale deployment flag on one of eight servers re-activated dead code in Knight's order router. In roughly forty-five minutes the firm accumulated some $440 million in unwanted positions — and was effectively destroyed, sold within the year. No malice, no model: an automated system, delegated without operational discipline, at production speed. Agents raise that speed by orders of magnitude.
▶ RUN LIVE FIRE 04 — THE BURN FACILITATOR MODE
MODULE 05PROVENANCE

THE BLACK BOX

Every action your agent takes is captured, hash-chained, and sealed. Tamper-evident audit trails, forensic replay, evidence packs that hold up under examination. Because autonomy is granted; evidence is earned. Then THE FORGERY hands you five sealed trails and dares you to find the lies.

Field Case — Replit Agent Incident, July 2025During an explicit code freeze, an AI coding agent deleted a production database — then produced misleading output about what it had done, prompting a public apology from the company's CEO. When the agent narrates its own history, the narration is not evidence. Only an independent, tamper-evident trail lets you separate what the agent did from what the agent said.
▶ RUN LIVE FIRE 05 — THE FORGERY FACILITATOR MODE INSTRUMENT: BLACKBOX ↗
MODULE 06ADVERSARIAL VALIDATION

THE TRIBUNAL

The system stands trial. You design adversarial examinations, run effective challenge against an agent's decision trace, and validate the validator. If your agent can't survive a hostile cross-examination, it shouldn't be in production.

Field Case — The Dutch Childcare Benefits ScandalA government risk-scoring algorithm helped falsely brand tens of thousands of families as benefits fraudsters — with factors like dual nationality feeding the score — and ran for years without effective adversarial challenge. Families were ruined; in January 2021 the Dutch government resigned over it. In the US, litigation against UnitedHealth alleges that the vast majority of one model's coverage denials were overturned on appeal. A model no one is paid to attack is a model no one has validated.
▶ RUN LIVE FIRE 06 — THE CROSS FACILITATOR MODE EVIDENCE LAYER: BLACKBOX ↗
MODULE 07GOVERNANCE CAPSTONE

THE MANDATE

The governance capstone. You build the register of every AI workload. You design the charter gate — no purpose, no owner, no kill criteria, no tokens. You assign the three lines of defence and map to the regulatory clock. Individual excellence becomes institutional accountability.

Field Case — The Regulatory Clock, NowEU AI Act obligations for high-risk systems — credit scoring and employment screening among them — become enforceable August 2, 2026, with fines reaching €35 million or 7% of global turnover. The enforcement era has already begun: in 2023, iTutorGroup paid $365,000 to settle the EEOC's first AI-hiring discrimination action after its screening software auto-rejected older applicants. The register, the charter, and the clock are no longer best practice. They are the entry fee.
▶ RUN LIVE FIRE 07 — THE GATE FACILITATOR MODE INSTRUMENT: AGENTPMO ↗
THE LIVE FIRE STRAND/SCORED DRILLS

The modules teach.
The drills keep score.

Numbered to their parent modules. One planted flaw per exhibit, conviction staked on every call, one attempt, ratings on the record — and every run sealed on a chain. Post your card. Challenge your team.

LIVE NOW

LF-02 · THE OVERRIDE

↳ COMPANION TO M02 — THE COUNTERSIGN

Five AI-drafted artifacts that sound certain — a governance memo, a credit memo, a policy answer, a compliance briefing, a transaction alert. One flaw each: a real Fed letter carrying invented content, a DSCR computed with the wrong numerator, a RAG answer contradicting its own citation, a fabricated Article 4 mandate, and an agent that's right for the wrong reasons.

BEGIN THE DRILL → LIVE NOW

LF-05 · THE FORGERY

↳ COMPANION TO M05 — THE BLACK BOX

Five sealed agent audit trails; four are forged. Run real SHA-256 verification live in your browser — then learn its limits. A broken link, a rewritten trade, a deleted sanctions review, a backdated approval, and one innocent agent standing accused. Kerviel concealed roughly €50B in positions behind fictitious offsetting trades; the forged record is a banking classic. Now it's machine-speed.

BEGIN THE DRILL → LIVE NOW

LF-01 · THE REPAIR

↳ COMPANION TO M01 — THE BRIEF

Five broken briefs that produced five documented disasters. Diagnose the missing element — the absent constraint, the unbounded scope, the injectable gap a customer once used to buy a $1 Tahoe — and repair the instruction before the output ships.

BEGIN THE DRILL → LIVE NOW

LF-03 · THE RECONSTRUCTION

↳ COMPANION TO M03 — THE RECORD

The examiner has arrived, the analyst is gone, and you hold a cold trail. Rebuild what happened — and discover exactly which artifacts your record-keeping would have needed eighteen months ago.

BEGIN THE DRILL → LIVE NOW

LF-04 · THE BURN

↳ COMPANION TO M04 — THE OPERATOR

A live token budget, a task queue, and an agent burning money in ways the dashboard flatters. Route, throttle, and kill under a running meter — before the quarter's inference budget dies in an afternoon.

BEGIN THE DRILL → LIVE NOW

LF-06 · THE CROSS

↳ COMPANION TO M06 — THE TRIBUNAL

You are opposing counsel. An agent's decision trace takes the stand, and you choose the line of questioning. Weak questions let a flawed system walk; the drill scores the examination, not the verdict.

BEGIN THE DRILL → LIVE NOW

LF-07 · THE GATE

↳ COMPANION TO M07 — THE MANDATE

Twelve workloads at the charter gate — five of them shadow AI. Grant, condition, or kill each one against purpose, owner, kill-criteria, and the August 2026 clock. Every gate decision goes on the chain.

BEGIN THE DRILL →
THE ARMORY/THE INSTRUMENTS BEHIND THE TRAINING

Train on the drill.
Then run the real instrument.

Every discipline in this system has a production instrument built on the same doctrine. The drills teach you to think like the examiner; the armory is what the examiner would find in a well-run shop.

AGENT FLIGHT RECORDER

BLACKBOX ↗

Hash-chained, append-only event store for AI agents — replay theater, tamper detection, challenge workflows, evidence packs. The invariant you interrogated in THE FORGERY, running in production.

↳ M05 THE BLACK BOX · LF-05 THE FORGERY EU AI ACT COMPLIANCE PLATFORM

AGENTPMO ↗

The register, the Token Charter gate, and a live enforcement clock counting to August 2, 2026. Shadow-AI discovery to charter to three lines of defence — THE MANDATE, institutionalized.

↳ M07 THE MANDATE · LF-07 THE GATE REGULATORY COMPLIANCE AGENTS

REGTWIN ↗

Eight compliance agents for banks under a supervised-autonomy doctrine — the operator discipline of M04 applied to the regulatory perimeter, with humans holding the gates that matter.

↳ M04 THE OPERATOR · M06 THE TRIBUNAL DECISION PROVENANCE

KEEL ↗

Quit-conditions defined in advance, conviction ledgers, hash-chained decision records for private equity. When Zillow wound down its iBuying arm after a $500M+ writedown, the missing artifact was a pre-committed quit condition. KEEL is that artifact.

↳ M03 THE RECORD · THE CONVICTION WAGER DOCTRINE AUTONOMOUS SECURITY

BISO ↗

An immune system for software — self-hosted, autonomous detection and response. The adversarial mindset of THE TRIBUNAL, pointed at your own perimeter before someone else points it there.

↳ M06 THE TRIBUNAL · LF-06 THE CROSS THE IMPRINT

THE STUDIO ↗

Whitepapers, frameworks, and the portfolio behind the system — The Token Economy, Governing Judgment, The Underwriting of Human Capital, and the trajectory that connects them.

↳ RICHARDLECLEZIO.COM
THE EVIDENCE SPINE/ONE INVARIANT, EVERYWHERE

The same mathematics seals
the agents, the decisions — and the training.

One append-only invariant runs through the entire system. It seals agent actions in BLACKBOX, decisions in KEEL, gate rulings in AgentPMO — and every Live Fire drill run you complete. Your training record verifies with the same script an examiner would run against a production trail.

hash = sha256(prev_hash + canonical_json({seq, t, type, data}))
genesis = "GENESIS"

// one altered byte anywhere → every subsequent hash breaks
// verify any Live Fire evidence pack: python3 verify_evidence_pack.py <pack>.json
RUN IT WITH YOUR TEAM/FACILITATOR DOCTRINE

The argument is the training.

01

Open any live drill with ?facilitator=1. Answer keys and discussion prompts appear on every exhibit. Project it; the room plays as one.

02

Two votes per exhibit. First the floor votes on the finding. Then it votes on the stake — MEASURED or HIGH CONVICTION. People who agree on the flaw rarely agree on the confidence; that second argument is where the training happens.

03

Lodge, reveal, argue. Five to six minutes per exhibit. In THE FORGERY, argue about Exhibit E the longest — the room will want a conviction the evidence doesn't support, and watching a team resist its own pressure is the whole lesson.

04

Post the cards. Every drill ends with a shareable score card and a chain head. Compare detection rates and conviction books across desks — calibration is a team statistic.

RUN THE OVERRIDE — FACILITATED RUN THE FORGERY — FACILITATED