Skip to content

Glossary

Career and résumé terms · Software and AI agents · Mechanical and manufacturing · Terms about this site

Career and résumé terms

Contract manufacturer
A company that builds a product in its own factories for the brand that designed it.
FATP
Final assembly, test and pack. The last stage of building a product, where the finished parts are put together, tested and packed to ship.
OEM
Original equipment manufacturer. A company that makes a product or machine and sells it under its own name, often with parts or systems bought from others.
Yield
The share of the units a line builds that come out good.

Software and AI agents

ADR (decision record)
A short written decision with its context, the options I rejected and the trade-offs. The public site summarizes a few as Selected design decisions.
Autonomation
The name I use for the whole system: the agent fleet, its guards, the control plane and the review gates. It is the English term for Toyota's jidoka: automation that stops itself when something is abnormal (Autonomation).
Backstop
A git-level check that still runs if a hook was bypassed. One example is the refusal to commit in a shared primary checkout (Guard rails).
Brief
The bounded instructions given to a dispatched worker before it edits or runs commands (How I run AI coding agents).
Builder
A worker agent that owns a defined area for a defined task, in its own worktree (How I run AI coding agents). The Autonomation page calls the same thing a worker agent.
Checker
The role that reads or runs a change and records a decision on the exact version; it never writes the change it checks. It is the independent reviewer (How agent output is trusted).
Closeout gate
A hold on a run that claims it pushed work or produced a result: the run is not counted as done until the claim is checked against the repository or report it points at (Measuring agent reliability).
Control plane
The ledger, the role-based messaging and the plans with owned paths and gates that route and track agent work (Lanes and the run ledger).
Coordinator
The agent session that triages work, picks a lane, dispatches workers, and verifies results (Lanes and the run ledger).
Courier
A full-write worker in its own git worktree, tracked in the run ledger, reporting completion through an inbox (Lanes and the run ledger).
Deadman timer
A self-canceling rollback timer armed before a risky change so an operator must disarm it after verifying the outcome (Guard rails).
Design lock
A recorded limit on what an agent must be unable to do, written before the dispatch and deploy paths were built (Selected design decisions).
Dispatched run
A row in the run ledger for a tracked worker run: who launched it, which worktree, state, and last progress (Lanes and the run ledger).
Doom-loop guard
A guard that stops a worker after the same failing action repeats, allows one corrected re-dispatch, then escalates to a person instead of retrying again (Measuring agent reliability).
Drift
Live configuration or documentation that no longer matches the authoritative template or source; renders and freshness checks treat drift as a failure class (Incidents and the fixes they produced).
Fail closed
When a guard cannot evaluate a call, it denies the call instead of allowing it (Guard rails for AI coding agents).
False-done
A worker reports success but nothing useful landed; the ledger makes this class visible (Lanes and the run ledger).
Fleet coordinator
The role that triages work, decides the lane, dispatches, verifies results and names who merges. The Autonomation page calls it the coordinator agent (How I run AI coding agents).
Guard hook
A mechanical intercept on tool calls or git operations, configured in Claude Code and Cursor, plus git backstops (Guard rails).
Handoff
The note an agent writes at the end of a session so the next one can pick up: what was done, what is still open, and what to do next. Test output and the pass or fail summary go into it (Verification standard).
Hollow-done
A worker exits cleanly with an empty result; treated as a failure class in the ledger (Lanes and the run ledger).
Independent reviewer
In the main repositories, a separate agent whose recorded decision is a required pull-request check; only its identity can post it (How agent output is trusted).
Lane
A dispatch pattern with a fixed contract: courier, phone line, scout, remote argv, or orchestrate (Lanes and the run ledger).
Ledger
The run ledger: relational records of dispatched runs, with reapers and stall timers (Lanes and the run ledger).
Liveness probe
A check that every registered guard script exists, parses and imports, so a guard that has broken shows up (Guard rails for AI coding agents).
Non-vacuity proof
A test that plants a violation and expects refusal, so a guard cannot pass by inspecting nothing (Incidents and the fixes they produced).
Operator gate
A pause in a plan where a human must answer approved, rejected or modified before the next phase can run; gates are never answered automatically (Orchestrating multi-step work).
Owned paths
The files a lane or phase is allowed to change. The runner compares each phase's diff against them afterwards and fails the phase on a stray file (Orchestrating multi-step work).
Phone line
Resuming a finished courier on the same ledger row for follow-up work (Lanes and the run ledger).
Poka-yoke
Error-proofing: a design that does not depend on the operator remembering (Guard rails for AI coding agents).
Probe
A synthetic check, such as a blackbox HTTP, TCP, DNS or ICMP request, whose result feeds into metrics (Observability and alerting).
Ratchet
Changes to the guard set are reviewed: additions are registered, and removals or weakened protections require an explicit reason (Guard rails for AI coding agents).
Relay
Passing a decision from where it was made to where it's needed, without changing it. For example, the reviewer's decision is copied onto the pull request by a step that can only repeat what the reviewer recorded (How agent output is trusted).
Render
Applying a template to produce live configuration. If a container mounts a single rendered file, an atomic replace can leave it pointing at the old copy (Incidents and the fixes they produced).
Runbook
An alert-linked page describing what a rule means and first steps; generated stubs fill gaps without overwriting hand-written pages (Observability and alerting).
Scout
An untracked, read-only investigation on the cheapest capable model tier (Lanes and the run ledger).
Silence
Muting an alert route for a set time, with a written reason. It can be undone without deleting the route (Observability and alerting).
Wedged
A dispatch row whose process is alive but not making progress (Lanes and the run ledger).
Worktree
An isolated git checkout for agent edits so the shared main checkout stays clean (Lanes and the run ledger).

Mechanical and manufacturing

8D
Eight disciplines. A structured method for solving a quality problem step by step: contain it, find the root cause, fix it and stop it from coming back.
CFD
Computational fluid dynamics. A simulation of how air or liquid flows through or around a design, used to check cooling and airflow before anything is built.
DFM/DFA
Design for manufacturing and design for assembly. Shaping a part, and the way parts go together, so it is easy to make and to assemble the same way every time.
FEA
Finite element analysis. A simulation that splits a part into many small pieces to see how it bends, heats up or vibrates under load.
Gauge R&R
Gauge repeatability and reproducibility. A test of how much of the variation in a set of measurements comes from the measuring system itself, the gauge and the people using it, rather than from the parts.
GD&T
Geometric dimensioning and tolerancing. The drawing language that says how far each feature of a part may vary and still fit and work.
Golden line
A production line that has been proven to meet spec, kept as the reference that a new line is matched against before it is allowed to run production.
NPI
New Product Introduction. The stage where a new product's design is built for the first time and brought up to full production.
PFMEA
Process failure mode and effects analysis. A step-by-step review of a manufacturing process that asks how each step could fail, what that would cause and how it would be caught.
PLC
Programmable logic controller. A rugged industrial computer that runs a machine's sequence of moves and its safety checks.
SPC
Statistical process control. Tracking measurements from a running process on control charts so drift shows up before parts go out of spec.

Terms about this site

Allowlist
An explicit list of what is permitted. Items outside the list are refused or excluded; the permitted items may be commands, resources, connections or pages.
Leak test
A scan of the built public site, search index included, that fails the build on any finding (About this site).
Status header
The banner at the top of each page, generated from its front matter, showing the state and the review date (or a short "last reviewed" line on selected pages) (Documentation that stays true).

Last reviewed Oct 27, 2026