Skip to content

Glossary

The words we use,defined once.

44 terms from across LiteSurface: the objects you work with, the rules behind a score, and the controls around them.

A

Artifact
A generated, versioned document such as a validation brief, a PRD, or a build handoff. Artifacts are pinned to the concept version they were built from. Design package
Assumption
A falsifiable statement that must be true for a concept to work. Assumptions are ranked by importance, uncertainty, and cost of being wrong. Validation
Audit log
A record of consequential actions, such as membership changes, project transfers, decisions, budget changes, and exports, with who acted and when. Trust center

B

Brand kit
The versioned set of brand decisions Brand Studio publishes: name, palette, type pairing, logo system, app icon and favicons, imagery, and DESIGN.md guidelines. Published kits never change, so screens and code keep pointing at the version they were designed with. Brand Studio
Brand Studio
The workspace that designs a brand from a concept’s research, slot by slot: a screened name, a palette that passes WCAG 2.2 AA, a type pairing, a logo symbol traced to vector, and imagery. Confirming the required slots publishes a brand kit. Brand Studio
Budget
A workspace spending limit for model usage. When a budget is spent, new runs are refused rather than allowed to overspend. Governance
Build handoff
A 26-section implementation bundle with README, SOURCES, and DECISIONS files, written so a coding agent can build the concept without rediscovering it. Builder apps made from a concept start with it in docs/handoff; it also exports as a ZIP. Design package
Build minutes
Time a Builder sandbox is awake, counted per calendar month (UTC) against the plan: 60 on Solo, 600 on Team, and 3,000 on Enterprise. A sleeping session uses none; when the month’s minutes are spent, sessions sleep until the next month. Pricing
Builder
Where an AI agent builds a real app in a live sandbox. You ask for changes in Build, Plan, or Chat mode, watch the preview update, edit files yourself, and keep a version with a diff for every change. Apps start from a design package, a template, or a prompt. Builder

C

Capability
A technology, platform, or business primitive that can enable a product mechanism, such as on-device speech recognition or a payments API. Each has availability and freshness.
Capability atlas
The searchable, linked collection of capabilities available to a team. Concepts may only depend on capabilities in the atlas. Capability Atlas
Circuit breaker
A control that takes a failing AI provider out of rotation for a cooldown period, so requests fall back to the other provider where policy allows.
Claim
An atomic proposition extracted from a source, with a locator pointing to where it appears, plus credibility and freshness metadata. Evaluations cite claims by id. Evidence
Concept genome
The typed structure of a concept: target user, trigger, mechanism, value loop, business model, moat, feasibility, assumptions, risks, and evidence. Concept genomes
Confidence
How much weight a result can bear. It combines evaluator confidence with disagreement and is lowered when only one evaluator responded or evidence is thin. It never changes the score. Methodology

D

Deploy preview
A hosted copy of one Builder version at a shareable URL, with its own Postgres database. Unclaimed previews are torn down after 72 hours on Team and 168 on Enterprise; claiming one moves its hosting and database into your own accounts. Not yet available for full-stack monorepos. Ship
Design package
The published brand kit and the confirmed screens with their specs, written into a Builder app’s docs/design folder with the favicons and app icons installed, so the agent builds close to 1:1. Re-importing replaces it with the latest confirmed designs. Apps made from a concept also carry the build handoff in docs/handoff. Design package
Disagreement index
The average gap between the two evaluators’ criterion scores, divided by 100. Zero means identical scores; 0.2 means 20 points apart on average. Methodology

E

Evaluation run
One execution of the evaluators against one concept version. Runs are immutable: new evidence or outcomes produce a new run beside the old one.
Evaluator
A model that scores each criterion with a rationale, cited claims, and gate signals. LiteSurface uses one from OpenAI and one from Anthropic, independently. Evaluation
Evidence coverage
The share of an evaluation’s material claims that are supported by claims in the project. Claims resting only on a model’s prior knowledge count as unsupported. Methodology
Evidence-gap loop
A targeted research pass that turns the evaluators’ stated gaps and the riskiest assumptions into queries, gathers new claims, and appends a fresh evaluation. Methodology
Experiment
A planned test designed to reduce uncertainty about one or more assumptions, with success and kill criteria written down before it runs.

F

Fixture mode
A deployment setting that replaces model calls with deterministic outputs, so the full pipeline, exports included, runs without provider keys. Used for demos, tests, and self-hosted trials; the sample evaluation on this site was produced this way.

H

Hard gate
A rule that blocks a recommendation regardless of score, such as a prototype that overruns the build horizon or a capability that is unavailable. Methodology

I

Image credits
What Brand Studio and Mockup Studio image generations spend, with the estimate shown before each run. A draft costs 6, a final 22, and a screen spec 1; names and code-drawn assets cost nothing. Plans include 400 (Solo), 4,000 (Team), or 20,000 (Enterprise) a month, and unused monthly credits expire. Pricing

K

Kill criterion
The result, written down before an experiment runs, that would end or reshape the concept if observed.

L

Lineage
The history connecting a concept to the concepts it was forked or merged from, so revised ideas keep their origin and evidence.

M

Median synthesis
The deterministic step that takes the median of the evaluators’ scores for each criterion and sums median times weight into the concept score. Methodology
Mockup Studio
The workspace that plans an app’s flows from the research and designs each screen per device in the brand kit. Confirmed screens get an editable spec, and each flow can be shared as a prototype link. Mockup Studio

O

Opportunity
A structured pairing of a meaningful user job or problem with a plausible enabling mechanism. Opportunities are the raw material for concepts.
Outcome
The recorded result of an experiment, with its evidence and interpretation. Outcomes feed a new evaluation without editing earlier ones.

P

Portfolio
The ranked set of concepts in a project, each with a human decision: shortlist, validate, build, hold, or reject, recorded with rationale. Portfolio
Prompt version
A versioned prompt with its schema and metadata. Admins can activate a new version or roll back, and every result records which version produced it. Governance
Provenance
The trace connecting any derived result back to its sources, claims, models, prompt versions, and the run that produced it.

R

Relevance triage
A quick judgement of each search result against the research focus before fetching. Results scoring below 0.35 are never fetched.

S

Score band
The range a score falls in: exceptional at 85 and above, strong at 72, hold at 58, and weak below that. Each band maps to a recommendation. Methodology
Scorecard
The weighted criteria and gates used to evaluate concepts. The default, Fast Product Bet, has nine criteria whose weights sum to 1.0. Methodology
Soft gate
A gate that caps rather than blocks. Low evidence coverage is the soft gate: it holds the recommendation below build now and flags the concept as needing evidence. Methodology
Storyboard
Mockup Studio’s plan of an app: its navigation, shared components, and up to five flows of screens, 24 at most, each with a frame per device. It is planned from the concept’s research, and screens that trace to nothing in it are dropped. Mockup Studio
Strategy profile
A project’s constraints: platforms, build horizon, capital and regulatory tolerance, privacy posture, business models, exclusions, and distribution assets.

U

User job
The outcome a target user is trying to achieve in a particular context. Opportunities start from user jobs rather than from features.

V

Validation plan
A ranked list of assumptions and the cheapest experiments that could falsify the riskiest ones, generated when a concept is promoted. Validation

See every termin a working project.

Explore a sample project before you run your own: genomes, gates, and coverage on real screens.

  • Solo and Team are free during early access
  • Explore a sample project first
  • Bring your own OpenAI and Anthropic keys