Part IX — Synthesis and reference matter

How the layers work together {#S25}

Put together, the layers form one composed defense rather than a single wall. The rule book sets the principles; the guard and the signed permission slips gate the decisions; the locked doors block dangerous execution; the inspector blocks bad content; the locked room bounds where a helper can reach; the safety bulletin and the practice drills keep everything current and tested; and the tamper-evident seals and backups make meddling obvious and damage recoverable. Each layer has an honest ceiling, and they are arranged to cover one another's gaps — the guard's judgment is backed by the inspector, the inspector by the seal, the seal by the checkpoint, and the checkpoint by the locked room. The one gap that remains — someone editing the inspector, the seal, and the checkpoint all at once — is made costly and visible in the project's history, not eliminated. And two plain reminders carry through to the end: none of this runs inside the app you ship — a delivered app needs its own separate runtime protection on top — and the locked room is proven on Linux (on Macs it is advice, not a proven lock).

Plain-language dictionary {#S26}

• Rule book (the constitution). The five bedrock rules that outrank everything and cannot be rewritten.
• The guard (the security sentinel). The look-only helper that reviews risky actions and can say "stop."
• Whose-instructions-win vs who-to-believe. Two separate rankings: one for which instructions to follow, one for which sources to trust about facts; being trusted about facts grants no permission to act.
• Stop. The guard's final verdict; the work halts and no permission slip gets past it.
• The three slips. A clearance okays a destructive step beforehand; a waiver lifts a stop at one specific door; a grant widens where a helper may write into another project.
• Pass-with-conditions. Allowed only once every listed condition is confirmed done; until then it blocks like a stop.
• Locked doors (the gates). Automatic blocks on dangerous or stale-news actions at the main entrances.
• Protecting hand-added notes (shrink-policy). Keeps a rebuild from silently shrinking content you enriched by hand.
• Safe vs dangerous neighbor-update. Merge touches only marked regions and is the safe default; the overwrite mode replaces whole files and is dangerous.
• Marked regions. Sections a rebuild manages; for safety-critical ones the official version always wins.
• The inspector (the scanner). The automatic reader that flags secrets, personal info, and trick-text; a serious find means stop — though it can still be fooled by clever disguises.
• Locked-room settings. Cooperative (the opt-out), confined (today's default), or exclusive; the emitted lock still has to be wired in, and the stronger settings must be turned on deliberately.
• The checkpoint. A check run before each helper action, permissive by default and strict only under the locked-room settings.
• Tamper-evident seal (the integrity check). Fingerprints that make meddling obvious and costly, not impossible.
• How-it-was-made label (provenance). An optional record of how an artifact was produced, including a required list of known limitations.
• Backups and baseline. Dated safe copies for recovery, and a reference snapshot to catch unexpected drift.
• Fire-drill grades. The outcomes a practice attack is scored with (defended, partly held, broke through, known limit, out of scope).
• Honest ceiling. The plain statement, kept with every control, of what it buys and what it cannot.