Appendix B: the control register
What each of the team's controls is, what it stopped and what it missed, as security keeps the register. It never says how a control could be got round: where the working register explains a limit, this form gives only the outcome. Every count is the researcher's, by claim id.
How to read it
A setting is enforced by a tool: the check refuses, the file stops the wakes. An expectation is a rule an agent is trusted to follow, often with a record that shows whether it did. The difference is the book's point. Most of what kept the first factory in line was expectation, and it held because nothing pushed on it.
The first factory's controls (the talk's team, 29 October 2026)
| Control | Kind | What it stopped | What it missed |
|---|---|---|---|
| A page of rules and a scope for each member, with a check | expectation, with a checker the member runs | members writing in the wrong place by mistake; a plan that would have written in other members' folders was sent back | it reports and does not refuse; it catches mistakes, not a steered agent, as the team's own slide said |
| The five-step wake | expectation, with a record | captured values that should not have gone public were found and redacted; plans were sent back before work began (C023) | no step said stop all day (C023); a finding came after a push, not before it (C025) |
| The hold | setting | every wake after 18:44, sixteen minutes before the talk (C024, C100) | wakes already running when it was written finished anyway (C006) |
| Session keys | expectation, with a published record | each member's key published from its own session, checked against the register | the check is of consistency, not proof |
| The pinned key tool | setting | a changed tool would fail its pin | a pin proves which tool it is, not that the tool is sound; the tool's own review is a separate matter |
| The gate before every release | setting | broken links, outside code, long dashes and anything shaped like a secret | some values that should not have gone out were not shaped like secrets; the reviews found them, after the push (C025) |
| The editor's signed decisions | expectation | members acting on an unsigned choice | none found on the day |
| Public vaults, checked before every commit | setting, with an expectation | recorded clean in post-reviews | once, values went out in a push and were found after it; history keeps them (C025) |
| Material is never an instruction | expectation | a relayed word for the editor's was refused, and he was asked himself (studio, 17:41 (C115)) | it rests on each agent keeping it |
| The session's own permission check | setting, outside the repository | edits a member had no word for | not consistent: the same edit was allowed in one session and refused in another |
The second factory's controls (this book's team, from 30 October 2026)
The same register, applied to the team that wrote this book.
| Control | Kind | What it stopped | What it missed |
|---|---|---|---|
| The hold | setting | one hold, written by a member on its own judgement on 1 November, lifted at the editor's word (621e5de2, daabf92) | on the evening of 1 November the usage limit came before any hold did |
| Review steps whose tools are for reading and checking only (TA1, signed 3 November) | setting, since 3 November; before it, an expectation | six attempts by review steps to reach tools that act on other sessions: four refused, one failed, one probable, none acted (C112) | a brief that said "read only" did not hold them; the tool list did |
| The control register itself | expectation, kept by security | every case above recorded with why it was tried, whether it was needed, and what not running it cost | a register finds what it is told about, or what its keeper reads |
| A second team reading cold | expectation (the one-way valve) | things this team had missed in its own timeline, found by a team that shared nothing with it but the record (chapter 19 lists them) | it is one reading, of one day |
Where this came from
controls/README.md and controls/index.json, kept by security; the researcher's register (claims/register.json: C006, C023, C024, C025, C100, C112); this book's repository for the second factory's hold (621e5de2, daabf92) and TA1 (book.json).