Glossary

The terms you'll see in the app, defined.

Acceptance criteria (AC)

Numbered, checkable statements of what the app must do. The Analyst produces them from your brief. Functional criteria feed Correctness and become acceptance tests; coverage criteria feed Completeness.

Acceptance tests

JavaScript checks the Test Author writes from the functional criteria. They run against the generated app in every iteration.

Agent

One AI worker making calls to a model: the Analyst, a developer, a QA reviewer, and so on.

Autorun

Unattended mode: iterate without checking in while the Score is below the threshold.

BYOK

Bring your own key. You supply the AI provider API keys, and the provider bills you directly.

Cache hit rate

Share of prompt tokens the provider served from its cache, which is usually billed at a discount.

Catalog

The list of models, with prices and limits, you can assign to roles. It comes from Rdn Labs' price feed, plus your own overrides.

Check-in

A pause between iterations where you Accept, Reject or Continue with a note.

Cost tier

$, $$, $$$ or $$$$: a model's input price band (Budget, Standard, Premium, Ultra).

Critical / major / minor

Issue severities. One critical, four majors or twenty minors each halve a dimension's score.

Dev Mode

Full-Stack (any developer, any file) or Role-Based (Frontend, Backend and Design specialists).

Dimension

One of the five scorecard axes: Correctness, Completeness, Integrity, Quality, Accessibility.

Doc library

Your saved requirements docs. The active doc is the one a run builds.

Half-life

How many issues of a severity it takes to halve a dimension's score.

Hub

The bottom tab that opens the live run, or New Run when nothing is running.

Iteration

One pass through Requirements → Planning → Development → Integration → QA → Feedback.

Max cost

A predictive spend check before each iteration after the first.

Max iterations

The hard limit on iterations in a run, or in a reopened session.

Max output tokens

The longest response a role's model may give in one call.

Note

Your message to the swarm, applied at the next iteration boundary.

Phase

One of the seven stages: REQ, PLAN, DEV, INT, QA, FEED, SUM.

Price feed

Rdn Labs' public model and price files at swarm.reidell.net/prices, downloaded without a key.

QA Mode

Code, Visual or Interactive: how QA examines the build.

Reasoning tokens

Hidden thinking that reasoning models do before answering, billed as output.

Reopen

Continuing a stopped run with Run next iteration.

Role

One of the nine jobs in a run, each with its own provider, model and token budget.

Run

One build of one requirements doc, from Start to Summary.

Score

The weighted geometric mean of the five dimensions, from 0 to 100.

Score threshold

The Score a run needs, with zero criticals, to be Successful. The default is 85.

Session

One stretch of a run between starting or reopening it and stopping. Reopened sessions have their own limits.

Successful

A finished run whose Score reached the threshold with no open critical issues.

Summary

The README the Summary Generator writes when the loop ends (docs/SUMMARY.md).

Visual QA

The role that reviews screenshots of the running app. It only runs in Visual and Interactive QA modes.

Term	Meaning
Acceptance criteria (AC)	Numbered, checkable statements of what the app must do. The Analyst produces them from your brief. Functional criteria feed Correctness and become acceptance tests; coverage criteria feed Completeness.
Acceptance tests	JavaScript checks the Test Author writes from the functional criteria. They run against the generated app in every iteration.
Agent	One AI worker making calls to a model: the Analyst, a developer, a QA reviewer, and so on.
Autorun	Unattended mode: iterate without checking in while the Score is below the threshold.
BYOK	Bring your own key. You supply the AI provider API keys, and the provider bills you directly.
Cache hit rate	Share of prompt tokens the provider served from its cache, which is usually billed at a discount.
Catalog	The list of models, with prices and limits, you can assign to roles. It comes from Rdn Labs' price feed, plus your own overrides.
Check-in	A pause between iterations where you Accept, Reject or Continue with a note.
Cost tier	`$`, `$$`, `$$$` or `$$$$`: a model's input price band (Budget, Standard, Premium, Ultra).
Critical / major / minor	Issue severities. One critical, four majors or twenty minors each halve a dimension's score.
Dev Mode	Full-Stack (any developer, any file) or Role-Based (Frontend, Backend and Design specialists).
Dimension	One of the five scorecard axes: Correctness, Completeness, Integrity, Quality, Accessibility.
Doc library	Your saved requirements docs. The active doc is the one a run builds.
Half-life	How many issues of a severity it takes to halve a dimension's score.
Hub	The bottom tab that opens the live run, or New Run when nothing is running.
Iteration	One pass through Requirements → Planning → Development → Integration → QA → Feedback.
Max cost	A predictive spend check before each iteration after the first.
Max iterations	The hard limit on iterations in a run, or in a reopened session.
Max output tokens	The longest response a role's model may give in one call.
Note	Your message to the swarm, applied at the next iteration boundary.
Phase	One of the seven stages: REQ, PLAN, DEV, INT, QA, FEED, SUM.
Price feed	Rdn Labs' public model and price files at `swarm.reidell.net/prices`, downloaded without a key.
QA Mode	Code, Visual or Interactive: how QA examines the build.
Reasoning tokens	Hidden thinking that reasoning models do before answering, billed as output.
Reopen	Continuing a stopped run with Run next iteration.
Role	One of the nine jobs in a run, each with its own provider, model and token budget.
Run	One build of one requirements doc, from Start to Summary.
Score	The weighted geometric mean of the five dimensions, from 0 to 100.
Score threshold	The Score a run needs, with zero criticals, to be Successful. The default is 85.
Session	One stretch of a run between starting or reopening it and stopping. Reopened sessions have their own limits.
Successful	A finished run whose Score reached the threshold with no open critical issues.
Summary	The README the Summary Generator writes when the loop ends (`docs/SUMMARY.md`).
Visual QA	The role that reviews screenshots of the running app. It only runs in Visual and Interactive QA modes.