STAGE Glossary
Method
| Term | Definition | Avoid |
|---|---|---|
| STAGE | A human-directed method for systems-oriented, tool-mediated, agent-operable, governed, and evidence-backed game engineering. | AI-first architecture, autonomous developer |
| Direction | Accountable human intent governing goals, priorities, taste, and acceptance. | Prompt |
| Delegation | Explicit authority granted to an agent for a bounded class of actions. | Autonomy |
| Operation authority | Permission for one proposed mutation at its actual scope. It may be narrower than ownership or provenance of the containing artifact. | Whole-file permission inferred from one field repair |
| Provenance | Evidence about the origin, derivation, or responsibility of state. Provenance informs an authority decision but does not itself grant permission to mutate. | Authorization |
| Safety Kernel | Eight universal rules protecting human authority, existing work, canonical sources, truthful evidence, and reversibility. It requires no STAGE artifact. | Full STAGE repository, conformance checklist |
| Delivery loop | The recommended Frame -> Inspect -> Plan -> Build -> Rehearse -> Judge -> Checkpoint sequence for one project-native change. Steps may combine or repeat. | Mandatory ceremony |
| Stage Director's Flow | Optional Cue -> Dialogue -> Ready -> Active -> Rehearsal -> Director's Call -> Checkpointed control loop for selecting and judging solo human-agent work around the delivery loop. | Sprint framework, mandatory ticket system |
| Director's Call | Accountable human disposition after a fair rehearsal: accept, revise, reject, or defer. It is separate from mechanical check status. | Automated acceptance, agent approval |
| Rehearsal | Repeatable execution used to gather claim-matched evidence before accepting a change. | Test, when visual, runtime, or interactive evidence is also required |
| Checkpoint | A known source-control state from which work can be accepted or safely reversed. | Backup |
| Project Context Spine | Smallest versioned source chain a fresh agent follows for operating rules, project direction, canonical detail, and active work coordinates. | Transcript archive, duplicated handbook |
| Task Handoff | Disposable revision-bound bridge carrying unfinished execution state across a task boundary. | Project memory, source of truth |
| Orientation Check | Read-only fresh-task reconciliation of repository coordinates, authority, protected state, current outcome, next action, and verification gates before mutation. | Context score, agent self-certification |
| Orientation workflow | Explicit stage-orient route that performs an Orientation Check and stops after reporting the safe next action. | Automatic setup, implicit delivery prerequisite |
| Verification workflow | Explicit stage-verify route that checks an existing delta, revision, artifact, or claim without repairing failures. | Fix loop, retry-until-green |
| Human documentation interface | Derived reading surface built from selected canonical repository sources. | Second handbook, generated source of truth |
| Work-item branch | Short-lived branch and isolated worktree for one Ready item, with declared path ownership and integration order. | Permanent subsystem branch, hidden agent workspace |
| Path lease | Temporary declaration that one active mutator owns a path or asset family, exclusively when its format or tool makes concurrent edits unsafe. | Permanent code ownership, permission to broaden scope |
| Product release version | Human-approved player-facing release identity under the project's declared compatibility rule. | Exact source revision, quality score |
| Build identity | Exact source and environment coordinates for one binary, including revision and dirty state. | Product release version alone |
| Data contract version | Independent compatibility version for a save, content catalog, network protocol, mod API, or artifact schema. | Product release version |
| Gameplay Design Dialogue | Human-led discussion that recovers intent, exposes assumptions and tradeoffs, and produces a human-approved gameplay hypothesis before substantial implementation of a new or materially changed premise. | Agent silently invents the game, design questionnaire dump |
| Human-approved gameplay hypothesis | Bounded statement of target context, fantasy, verbs, intended experience, short loop, first-minute understanding, success, failure, and stop conditions accepted by the accountable human designer. | Generated feature list, permission to keep working |
| Player Experience Gate | Human decision to accept, block, or defer expansion of a new or materially changed player-facing loop after direct play of the real game. | Automated gameplay test, fun score |
| Thin complete loop | Smallest player-facing slice containing production input, presentation, feedback, objective, success or failure, and continuation needed to judge the experience hypothesis. | Architecture skeleton, hidden debug path |
| Player-System Evidence Loop | Diagnostic trace of what the player must perceive, decide, do, read, learn, and adapt to across one repeated systems-heavy activity. It organizes hypotheses and claim-matched evidence; it is not a score or universal model of fun. | Fun formula, theory checklist |
| Smallest complete decision | Narrowest production slice containing the alternatives, information, input, feedback, consequence, explanation, recovery, and continuation needed to judge one material player decision. | Isolated rule, UI mockup, automated policy alone |
| Earned Infrastructure | Persistent tooling, abstractions, operations, or product surfaces retained because an observed recurring job, risk, or named consumer justifies their cost. | Maturity feature, mandatory tooling |
| Maintainer Laboratory | This repository's optional schemas, profiles, migrations, studies, behavior probes, and release machinery. Ordinary game delivery does not adopt it by default. | Safety Kernel |
| Release version | Compatibility and capability version of STAGE's public method and plugin contract. It does not encode proof, adoption, or maturity. | Empirical score |
| Artifact schema version | Version declared by a project map or case-study record, independent from the STAGE method release. | STAGE release version |
Claims And Evidence
| Term | Definition | Avoid |
|---|---|---|
| Production consumer | The actual caller, runtime subsystem, engine loop, input route, middleware, authoring tool, platform, or player path consuming changed behavior. | Nearby preview that omits material behavior |
| Consumer-complete verification | The smallest truthful evidence route containing the production consumer and every integration or environmental boundary material to the claim. It is not automatically a full-game test. | Maximum test scope, nearest passing test |
| Evidence coordinates | Evidence class, provenance, scope, outcome, revision, effective toolchain, and material limitations. | Result |
| Evidence receipt | Optional durable record connecting a claim to evidence and uncertainty for a named consumer. | Required proof package |
| Quality gate | One selected check or human decision boundary protecting a material claim. STAGE groups gates as Source, Model, Engine, Consumer, Distribution, and Director without treating them as a maturity ladder. | Universal quality score, green game |
| Behavior probe | Fresh consumer-level agent invocation used to observe packaged routing, stopping, authority, or response behavior. It is model- and toolchain-bound and semantically human-judged. | Unit test, release proof |
| Mechanical input rehearsal | Scripted or virtual-device execution showing that a production input path reaches an outcome without claiming human usability or feel. | Playtest |
| Synthetic player | Automated policy exploring the observations, actions, and utility function it encodes. It supports mechanical and strategy-space claims, not human comprehension, feeling, or fun. | Representative player |
| Decision-bearing artifact consumer | A concrete person, tool, or operation needing durable output to compare, audit, release, retain, or make a decision the live project cannot serve. | Reviewers, human and AI review, standing capability |
| Player-build evidence | Revision-bound evidence that an engine assembled a standalone artifact, without implying launch, performance, signing, distribution, or release acceptance. | Release build |
Optional Contracts
| Term | Definition | Avoid |
|---|---|---|
| Optional contract | A normative interoperability interface applying only after deliberate adoption. More adopted contracts do not imply a better project. | Maturity level |
| Project map | Optional human-readable overview of direction, native architecture, operations, ownership, evidence, risks, and decision rights. | Required architecture document |
| Project manifest | Machine-readable companion to an adopted project map. | Agent config required for delivery |
| Map profile | Proportional map depth: compact, standard, or extended. | Maturity level |
| Named operation | Declared project action with purpose, capabilities, preconditions, effects, safety, completion signal, and claim boundary. | Command name alone |
| Operation state | documented, available, passed, failed, inconclusive, unavailable, or not attempted. | Test status without scope |
| Operation capability | Execution-environment affordance such as graphics, editor interactivity, network, credentials, or middleware connectivity. | Project precondition |
| Formal source | Repository state treated as immutable evidence during an external study. | Working copy |
| Rehearsal copy | Disposable exact-revision copy used when tools may import, compile, cache, or generate state. | Source-control branch |
| Target-maintainer review | Response from authority for the project being studied, separate from the STAGE researcher's conclusion. | Researcher acceptance |
| Researcher disposition | STAGE researcher's decision to retain, reject, or synthesize a study as evidence. It cannot grant project adoption. | Target-maintainer review |
Human-facing STAGE terms use spaces or hyphens for readability. Project-map
0.3 manifests use the exact snake_case enum spellings declared by their
schema, such as not_attempted, project_generated, external_generated, and
external_vendored. The two forms name the same concepts.
Visual Work
| Term | Definition | Avoid |
|---|---|---|
| Visual work | Ordinary implementation, diagnosis, inspection, or polish whose truth depends on production rendering or presentation over time. | Separate STAGE workflow |
| Native inspection surface | Existing engine or game route containing the production consumer and material context needed by one visual question. | Engine-native by location alone |
| Production-integrated gallery | Optional game or authoring feature such as an almanac, codex, bestiary, model viewer, or world previewer that reuses canonical production content for an independent player, author, or recurring developer job. | STAGE evidence product |
| Project-native capture/export | Screenshot, motion clip, comparison, build, or decision record produced through project or engine tooling for a named consumer. | Default contact sheet |
| Fixed seed | Permanent procedural coordinate retained for replay or regression. | Favorite seed |
| Exploratory seed | Bounded random sample used to discover variation and outliers; it becomes permanent only through explicit promotion. | Random run without bounds |
A still does not establish continuous motion, a clip does not establish input feel, and a gallery supports only the production consumers and material context it actually contains. Human product authority accepts visuals, audio, feel, balance, pacing, and fun.
Project Surfaces
| Term | Definition |
|---|---|
| Intent surface | Data, documents, assets, and tools expressing desired content or behavior. |
| Runtime surface | Mutable state and lifecycle through which the project realizes intent. |
| Operation surface | Stable command or tool through which a human or agent inspects or changes project state. |
| Evidence surface | Repeatable way to observe whether a bounded claim is true. |
| Ownership boundary | Declared distinction among human, project-generated, external, runtime, legacy, and unresolved artifacts. |
| Effective toolchain | Actual executable, editor, middleware, and version that produced evidence. |
| Unresolved authority | State permitting safe inspection but blocking the proposed mutation whose authority boundary is unresolved. It does not erase separate narrow authority that preserves the unresolved state. |
Roles
| Term | Definition |
|---|---|
| Director | Human or accountable group holding final product authority. |
| Agent | Delegated computational contributor able to inspect, reason, use tools, and modify project artifacts. |
| Domain owner | Code or module with authoritative responsibility for a game concept or lifecycle. |
| Engine shell | Engine-bound rendering, physics, input, audio, persistence, and integration around game decisions. |
Common Ambiguities
- Test may mean automated assertion, runtime smoke, visual inspection, or playtest. Name the evidence class.
- Config may mean authored intent, runtime debug state, or fallback default. Name the source and lifecycle.
- Generated may mean source generation, middleware output, runtime procedural content, or editor-created assets. Name the owner and regeneration policy.
- Available is not synonymous with executed or passed. Use the operation state.
- Done may mean implemented, mechanically checked, human accepted, or checkpointed. State which boundary was reached.
- Baseline may mean an authored reference, deterministic starting state, or accepted visual evidence. Name the kind and owner.