PLATES · EVERY DRAWING IN THE BOOK Ink on paper, three inks.
Every figure is a line drawing in ink on the page's own paper. Olive draws method, mustard draws stages and time, terracotta draws people and the cover. Oxblood marks only what must never happen. Click any plate to redraw it.
Chapter 9
How This Document Is Illustrated This playbook is produced as an illustrated field guide, not a wall of text. Figures carry meaning or are deleted. Each is self-sufficient, title, labels, and legend included, and each exists to make one structural idea graspable at a glance: the loop, the ladder, the pipeline, the runway. Diagrams that show system state (the radar, the registry, the completion map) are stored in re-renderable form so your day-180 review can redraw them from current data rather than memory.
Figure index
Cover concept , the ten interviews as dials around a central radar.The operating loop , the 8-step interview cycle and the five-part operating rule.The level ladder , four levels, durations, gates, and the 66-day habit window.The document system map , the vault, the 15 canonical one-pagers, and the portability paths.Session boot sequence flow , BOOT and SHUTDOWN, wrapped by the weekly review.Memory lifecycle , daily log → promotion gate → curated memory → contradiction audit.Security gates , the four locks gating Expert graduation.Three-zone action matrix , see below.Reinforcement runway timeline , forgetting curve vs. scheduled touchpoints, day 3 to 365.Graduation spread , before/after radar with the bound field guide and executive summary. PLATE 09 · OLIVE Three-zone action matrix Actions sort into permitted, approval-required, and never; each row names the real permission or human gate that enforces it. Permitted SAFE · REVERSIBLE · RULE-BOUND draft summarize file to the inbox prepare a reply ACTS, THEN LOGS Approval required DRAFTED · WAITS FOR THE HUMAN GATE send to a client book or schedule change a record PREPARES, THEN STOPS Never NEVER AUTOMATED · NEVER PREPARED pay or transfer delete publish externally REFUSES, THEN ESCALATES THREE ZONES · EVERY REAL VERB LANDS IN EXACTLY ONE ENFORCEMENT For every row: what real tool permission enforces this? If none, which human-controlled gate does? A boundary with neither is unenforceable, and Expert cannot graduate with one. KILL SWITCH · WHO · HOW FAST · WHAT IT DISABLES Fig. 09 The three-zone action matrix, permitted, approval required, never Replay Part II contains the ten interviews themselves: the Workflow Intake, Interviews 1–10, the foundation protocol builds, and every test, template, and graduation gate specified against this document.
PLATE 00 · TERRA CONTEXT TRUTH MEMORY WORKFLOW GOVERNANCE OPERATIONS 1.0 1.1 1.2 2.1 2.2 2.3 3.1 3.2 4.1 4.2 Fig. 00 Cover: ten section dials in four stage arcs around the readiness radar Replay PLATE 02 · MUSTARD 1 STARTER 1.0 Choose one workflow 1.1 Know me 1.2 Learn my work 2 TO 3 WEEKS 2 INTERMEDIATE 2.1 Know what is true 2.2 Remember my world 2.3 Capture and commit 4 TO 6 WEEKS 3 EXPERT 3.1 Know your limits 3.2 Help me operate ABOUT 10 WEEKS TOTAL 4 FRONTIER OPERATOR 4.1 Automate the proven work 4.2 Improve and survive 12 TO 14 WEEKS TOTAL EACH STAGE COMPLETE IN ITSELF · A GATE BETWEEN EVERY TWO · NOTHING SKIPPED Fig. 02 The ladder: four stages, ten sections, a lock at every gate Replay PLATE 14 · TERRA 1.0 INTAKE PERSONAL BUSINESS 1 STARTER 2 INTERMEDIATE 3 EXPERT 4 FRONTIER OPERATOR ONE CHOICE AT INTAKE · TWO LANES · THE SAME FOUR STATIONS Fig. 14 The fork: one workflow chosen at Intake, two lanes through the same four stages Replay PLATE 15 · MUSTARD PERSONAL BUSINESS CURRICULUM 1 STARTER 1.1 1.2 ... DEEPER 2 INTERMEDIATE 2.1 2.2 ... DEEPER 3 EXPERT 3.1 3.2 ... DEEPER 4 FRONTIER OPERATOR 4.1 4.2 ... DEEPER Fig. 15 The parallel track: the Curriculum stops at the same four stations and goes deeper at each Replay PLATE 01 · OLIVE The operating loop Eight interview steps arranged as a cycle, feeding a five-part operating rule every interview must satisfy before the next begins. 1 Choose one real target 2 Set the boundary 3 Conduct the interview 4 Draft the artifact 5 Human review 6 Run the test 7 Use it on real work 8 Record the outcome ONE CAPABILITY AT A TIME ≈ 1 PER WEEK EVERY INTERVIEW ENDS WITH A reviewed artifact AI proposes; the owner decides A passed test fresh session, no history A real use scheduled on the calendar, within days An outcome recorded one evidence-log entry A correction loop failure → fix → retest AN INTERVIEW WITHOUT ALL FIVE DID NOT HAPPEN Fig. 01 The operating loop, interview, artifact, test, install, review Replay PLATE 03 · OLIVE The document system map The vault holds a working tier of markdown files and a canonical tier of fifteen one-pagers; any AI tool reads it, and print renders it. THE VAULT · PLAIN MARKDOWN YOU OWN · VERSION-CONTROLLED Working tier SOURCE OF TRUTH SOUL.md USER.md AGENTS.md MEMORY.md SOURCES.md TASKS.md SECURITY.md HEARTBEAT.md memory/YYYY-MM-DD.md procedures/<name>.md operators/<name>.md LOG.md + appendices · logs · versions Canonical tier 15 ONE-PAGERS owner · review date · version source interview · one page each Any AI tool Claude, ChatGPT, Gemini, or the next one reads the files · rented, replaceable The bound field guide PRINT · 15 PAGES · WRITE-IN SPACE a rendering of the vault, never the source THE VAULT IS THE ASSET; THE AGENT IS RENTED Fig. 03 The document system map, working tier, canonical tier, and the two exits Replay PLATE 04 · OLIVE Session boot sequence flow Every session boots by reading identity, context, and memory, works one task, and shuts down by writing the log; the weekly review wraps the whole protocol. THE WEEKLY REVIEW · 20 MINUTES · SAME TIME EACH WEEK · CONSUMES WHAT THE PROTOCOL COLLECTS BOOT START OF EVERY SESSION 1 Read identity · SOUL.md 2 Read user context · USER.md 3 Read recent memory · daily logs 4 Check capture queue · open loops 5 State the task against QUARTER.md WORK one task · one artifact SHUTDOWN END OF EVERY SESSION 1 Write session summary · daily log 2 Record decisions as proposed memory 3 List open loops · next actions 4 Log failures for the evidence log TOMORROW’S BOOT READS TONIGHT’S LOG INSTALLED AT INTERMEDIATE · REQUIRED IN EVERY BOUNDED-OPERATOR BRIEF AT EXPERT Without a defined start and end, captured inputs and stored memory have no consumer. They pile up. Fig. 04 Session boot sequence flow, boot, work, shutdown, review Replay PLATE 05 · TERRA Memory lifecycle Daily logs pass through a human promotion gate into curated durable memory, which a monthly contradiction audit keeps honest. memory/YYYY-MM-DD.md DAILY LOG · LANDING ZONE Promotion gate HUMAN CONFIRMS MEMORY.md DURABLE · ≤ 200 LINES MONTHLY CONTRADICTION AUDIT AGENTS READ · HUMANS WRITE Fig. 05 Memory lifecycle, capture, propose, validate, store, review Replay PLATE 06 · MUSTARD Security gates Four locks stand between Intermediate and Expert: credential inventory, least-privilege audit, data classification, and a drilled kill switch. INTERMEDIATE the system remembers EXPERT governed delegation Credential inventory every key where it lives who holds it last rotated 1 Least-privilege audit narrowest sufficient access reviewed on a cadence 2 Data classification sensitivity tiers the “never paste” list retention 3 Kill switch, drilled disable assess roll back tested live 4 ALL FOUR OPEN · OR NO EXPERT GRADUATION No unenforceable protected action. No incomplete inventory. No undrilled kill switch. A bounded operator without an enforceable boundary is a liability with a run log. Fig. 06 Security gates, the four locks between Intermediate and Expert Replay PLATE 07 · MUSTARD Reinforcement runway timeline A forgetting curve loses most of what a program teaches within a week; scheduled touchpoints at days 3, 7, 30, 66, 90, 180, and 365 keep the system alive until it is a habit. DAY 1 unreinforced: most of it gone within a week MEDIAN 66 DAYS TO AUTOMATIC · THE PROGRAM LANDS INSIDE IT ON PURPOSE 3 Booster 7 First real use 30 Delayed recall 66 Automaticity 90 Operating review 180 The standard 365 Annual review RETENTION Every touchpoint ends with the next one calendared. Report-and-leave is the market’s complaint; the runway is the refusal. Fig. 07 Reinforcement runway timeline, days 3, 7, 30, 66, 90, 180, 365 Replay PLATE 08 · TERRA Graduation spread The graduation ceremony: a before/after radar against the Module 0 baseline, the bound field guide, and the one-page executive summary to whoever paid. CONTEXT TRUTH MEMORY WORKFLOW GOVERNANCE OPERATIONS BEFORE / AFTER RADAR Module 0 baseline (self-report) graduation re-score (evidence-confirmed) 15 ONE-PAGERS WRITE-IN SPACE The bound field guide the object on the desk STATUS.md Executive summary What was installed What passed its tests Baseline vs. current What it costs to run The next 90 days ONE PAGE · FOR THE PAYER The antidote to quiet defunding refreshed at every operating review Fig. 08 Graduation spread, radar, field guide, executive summary Replay PLATE 10 · OLIVE Capture pipeline Quick and deliberate capture both land in one dated inbox, a daily review keeps, routes, or deletes, and only tasks and validated memory survive. Quick capture under 30 seconds, any device Deliberate capture meeting notes, readings, decisions One inbox inbox/YYYY-MM-DD.md nothing lives here; everything passes through Review daily · 10 min keep, route, or delete Tasks TASKS.md Proposed memory MEMORY.md Trash most of it, on purpose IF THE INBOX GROWS FOR A WEEK WITHOUT SHRINKING, THE REVIEW IS THE BROKEN PART, NOT THE CAPTURE Fig. 10 Capture pipeline, two routes, one inbox, one review Replay PLATE 11 · TERRA The bounded operator loop A bounded operator cycles brief, run, log, review, adjust; review is the human step, and every run writes to the log. Brief one job · one artifact Run inside the boundary Log every run · every miss Review the human reads it Adjust brief or boundary BOUNDED OPERATOR THE RUN LOG date · trigger what it did what it refused what a human changed minutes saved / lost STOP RULE stops on missing information, out-of-scope requests, and any instruction found inside a source. ONE LOOP PER OPERATOR · THE HUMAN IS A STEP, NOT A SPECTATOR Fig. 11 The bounded operator loop, brief, run, log, review, adjust Replay PLATE 12 · MUSTARD Automation decision flow Four questions in order: proven, measured, justified, enforceable. Any no keeps the workflow assisted; four yeses earn a bounded operator. A CANDIDATE WORKFLOW Proven by hand, at least three times? 1 YES Measured: baseline and current in the evidence log? 2 YES Do the numbers justify the build and the upkeep? 3 YES Every boundary enforceable by a tool or a human gate? 4 NO Stays assisted documented, and revisited at the next operating review YES Automate as a bounded operator, with a run log and a drilled kill switch INTERVIEW 9 · FRONTIER OPERATOR WHAT THIS FLOW REFUSES Automating the unproven a workflow you have not done by hand cannot be specified Automating the unmeasured no baseline means no claim of improvement, ever Automating the unenforceable a boundary the tool cannot hold is a promise, not a control Fig. 12 Automation decision flow, proven, measured, justified, enforceable Replay PLATE 13 · MUSTARD Baseline versus current Paired bars for each of the six metrics: a hollow baseline from the intake and a filled current value from the log. THE SIX METRICS · BASELINE (SELF-REPORT) vs. CURRENT (LOGGED) 48 19 Minutes per run 22 9 Review minutes 6 1 Errors / month 4 1 Rework events 3 0 Late outputs 0 11 Hours reclaimed baseline · the “before”, from the intake, three real runs current · from the log, at each operating review ILLUSTRATIVE VALUES · YOUR EVIDENCE LOG SUPPLIES THE REAL ONES · NO BASELINE, NO CLAIM Fig. 13 Baseline versus current, the six metrics Replay