Glossary
Every term the docs lean on, in plain words. Hover a term anywhere on the site
to see its short definition.
Testing and the gate
| Term |
What it is |
More |
| Mutation testing |
Seed small deliberate bugs into code and rerun the tests to see if they notice |
Wikipedia, Techniques |
| Mutant |
One such deliberate bug: >= turned into >, 18 into 19 |
Techniques |
| Killed |
A mutant some test failed on. Good: the tests noticed |
Gate |
| Survivor |
A mutant every test still passes on. The tests would miss that bug too |
Gate |
| Kill rate |
Share of mutants killed. The one number that tells a green suite from a blind one |
Rules → Testing |
| Waiver |
An entry in .mutation-gate-waivers.toml that accepts one survivor, with a reason why no test should kill it |
Gate |
| Equivalent mutant |
A mutant that changes the code but not its behaviour, so no test can kill it |
Wikipedia |
| Diff-scoped |
Only the lines your change touched are mutated, not the whole repo |
Techniques |
| Baseline |
The unmutated test run the gate does first; if it fails, nothing after it means anything |
Config |
| Covering test |
A test that executes a given line; each mutant runs only against its covering tests |
Techniques |
| Code coverage |
Which lines a test run executed. The gate records it per test to build the covering-test map |
Wikipedia |
closure_depth |
How many import hops from a test to a file still count as covering it |
Config |
| Adversary |
An isolated review after a pass that sees the intent and the tests, never the code |
Techniques |
| Blind pass |
The reverse: a review of model code that sees the code, never the spec |
Techniques |
| Slice test |
One test per ticket that enters where a real user enters and checks the outcome the ticket asks for |
Techniques |
| Vertical slice |
A thin piece of a feature that works end to end, through every layer |
Wikipedia |
| Red first |
Write the test, watch it fail for the right reason, only then write the code |
Rules → Testing |
| Red loop |
One fast command, already run, that shows the bug. /diagnose starts here |
Debugging |
| Metamorphic test |
A test that checks a relation between runs, such as "rotating the input rotates the output", when the exact answer is unknown |
Wikipedia |
Model V&V
| Term |
What it is |
More |
| Model V&V |
Verification and validation for estimation and physics code: filters, frames, units, noise |
Techniques |
model_paths |
The files a repo declares as model code. Declared, never guessed |
Config |
| Model spec |
The pinned issue holding what the model covers, what it ignores, and how accurate it must be |
Rules → Model V&V |
MS-n |
One numbered line of the model spec. Model tests cite it in their docstring |
Gate |
| Envelope |
The range of states, time steps and manoeuvres the model is claimed to work in |
Rules → Model V&V |
| Golden |
A generated reference, such as the F and Q matrices, carrying the hash of the SymPy source it came from |
Rules → Model V&V |
| SymPy |
A Python library for symbolic maths. The model is written in it and the matrices are generated |
sympy.org |
| Kalman filter |
An estimator that blends a motion model with noisy measurements |
Wikipedia |
| PSD |
Positive semi-definite. A covariance matrix must be, or the filter's maths breaks |
Wikipedia |
| Jacobian |
The matrix of partial derivatives a nonlinear filter linearises with |
Wikipedia |
| NEES / NIS |
Normalised estimation error / innovation squared: statistics that say whether a filter's claimed uncertainty matches its real error |
Rules → Model V&V |
| Innovation |
The gap between a measurement and what the filter predicted it would be |
Wikipedia |
| Chi-square band |
The range NEES/NIS should fall in if the filter is consistent |
Wikipedia |
| Monte Carlo |
Running many randomised trials and reading the statistics off them |
Wikipedia |
| Units at every boundary |
A distinct type per physical quantity, so mixing metres and seconds fails to compile |
Wikipedia |
Naming
| Term |
What it is |
More |
| Vocabulary |
The dictionary of approved words for names, core plus the repo's own |
Techniques |
| Canonical word |
The one word the dictionary picks for a concept |
Rules → Naming |
| Rejected synonym |
A word the dictionary maps to its canonical word instead |
Rules → Naming |
| Mold |
The word-order pattern a kind of name must fit; compute_total fits the function mold |
CLI |
| WordNet |
A lexical database of English senses, used to flag new words that mean the same as old ones |
Wikipedia |
Workflow
| Term |
What it is |
More |
| Ticket |
A GitHub issue that holds the intent. The approved ticket is the plan |
SDLC |
| Triage |
Deciding a ticket is ready and adding its model:* label |
Reporting |
model:<name> label |
haiku, sonnet, opus or fable. It approves the ticket and picks the model that runs it |
Reporting |
| Follow-up |
A ticket filed for work deferred or found mid-task, with the follow-up label and one category |
Rules → Tickets |
size:tiny |
A follow-up that names one location and leaves no design choice. Tiny ones can share one PR |
Rules → Tickets |
| Kata |
The ticket-to-merge loop /kata runs. Named after the practice drill |
Commands, Wikipedia |
| Worker |
The fresh agent kata starts for one ticket, in its own worktree |
Commands |
| Worktree |
A second checkout of the same repo on its own branch, so parallel work can't collide |
git docs |
| LGTM |
"Looks good to me". Typed at the prompt, it lets kata merge |
Usage |
| Squash merge |
All of a PR's commits land on main as one |
GitHub docs |
Closes #N |
The PR-body line that closes ticket N when the PR merges |
SDLC |
| Handoff |
The note /done leaves so a fresh session can pick up |
Commands |
| Diff discipline |
The smallest change that satisfies the ask; no drive-by edits |
Rules → Diff discipline |
| Pre-registration |
Writing down what result would confirm a hypothesis before running the test |
Wikipedia |
| Blind test |
A test where the process can't see the answer it might be biased towards |
Wikipedia |
| Term |
What it is |
More |
| Rule |
A file in rules/ that says how to work. Synced into a repo, the agent reads it every session |
Rules |
| Skill |
A SKILL.md playbook the agent loads when a task matches, or when you call it |
Skills |
| Command |
A slash command in commands/ that runs a workflow: /kata, /done |
Commands |
<skill-dir> |
The directory a SKILL.md was read from; its scripts are called through it |
Skills |
AGENTS.md |
The instruction file Codex and other agents read; rules sync writes a block into it |
agents.md |
| Hook |
A command Claude Code runs on an event, without being asked |
Claude Code docs, Hooks and tools |
| PreToolUse |
The hook event before each tool call. The git guardrail runs there |
Hooks and tools |
| Stop hook |
The hook event when the agent finishes a turn. The gate runs there |
Hooks and tools |
| Git guardrail |
The PreToolUse hook that denies reset --hard, add -A, force-push and unmerged branch -D |
Hooks and tools |
--force-with-lease |
A force-push that refuses if the remote moved since you last fetched |
git docs |
| pre-commit |
The framework that runs checks at git commit; this repo ships hooks for it |
pre-commit.com, Hooks and tools |
| commit-msg stage |
The pre-commit stage that runs on the commit message, after you write it |
Hooks and tools |
| Plugin |
A packaged agent add-on. This repo ships one for Claude Code and one for Codex |
Hooks and tools |
| ast-grep |
A search-and-rewrite tool that matches code by syntax tree, not text |
ast-grep.github.io |
| tree-sitter |
The parser library ast-grep builds on; GDScript needs its grammar built separately |
Wikipedia |
| RFC1918 |
The private address ranges 10.x, 172.16–31.x, 192.168.x. no-leaks blocks them |
Wikipedia |
| Kokoro |
The text-to-speech model /say speaks through |
GitHub |
| Term |
What it is |
More |
| SOL |
Speed of light: the fastest a stage could run on this hardware, measured, not guessed |
Techniques |
| Ratio |
SOL ÷ measured. Near 1 means the code is already at the hardware's limit |
Techniques |
MODEL_DEFECT |
A stage measured faster than its own floor, so the machine model is wrong |
Techniques |
| Budget |
The share of the SOL each stage may spend, set in perf/budget.md |
Skills |
| Roofline |
A chart of whether a kernel is limited by compute or by memory bandwidth |
Wikipedia |
| llvm-mca |
An LLVM tool that estimates how a CPU would schedule a block of machine code |
llvm.org |