§ 01 — Start here
/start
Pick the row that sounds like you. It tells you which chapter to read first, and where to go after.
01 · You have
02 · Start with
03 · Leads to
Open
Twenty minutes and want the heart of the argument
Chapter 3The right facts, not every fact20 minutesCh. 3
Then Chapter 5 and the take-home kit (Ch. 11)
An agent about to get production credentials
Chapter 10Safety, cost and the human in the loopread firstcredentials
Harness and review habits before you widen access
Agents burning tokens on context I already have in source
Chapter 3Finding 1 — optimised repo cost+35–48%tokens
Lean AGENTS.md; add layers only when they pay
Changes that depend on a repository the agent cannot read
Chapter 3Findings 2–3 — facts across a boundary3 → 18–19/20
Contracts and freshness gates in Chapter 5
No clear “done” before the agent starts writing
Chapter 2Briefs, plans and specificationsbriefacceptance
One verifiable change at a time (Ch. 4)
Reports that say passed while the real check is unverified
Chapter 5Proof, not promises12 runsfreshness gate
Independent checks the agent cannot rewrite
The same environment failures every agent session
Chapter 6The harness and the friction loopfrictionscript
Fixes in the harness, not only in prompts
Need to match review depth to blast radius
Chapter 8Review, merging and risktiersblast radius
Tiered review from reversibility and verification strength
Wondering whether a set-up change actually helped
Chapter 9Tools and honest measurementmeasurelimits
Repeated runs with stated limits, not Tuesday anecdotes
Ready to adopt — what do we put in place first
Chapter 11The take-home kitStage 0justify
Default / Conditional / Experimental labels for each layer
Want the findings and thesis before any chapter
ForewordWhy I ran the experiments4 findingsthesis
Eleven chapters with checklists and the worked example
Numbers in chips (for example +35–48%,14 / 16, 3 → 18–19/20,12 runs) come from the Foreword findings. Each has limits stated in the Guide.