Skip to main content

Reference

Glossary

Find the terms Paper Trail uses in lessons, practice, and project rubrics. Definitions link to their supporting references when one is available.

Back to Explore

452 glossary terms

abstention
A deliberate refusal to answer or act when evidence, authority, or safety is insufficient.

Supporting references

acceptance check
An observable test a human uses to decide whether a result serves its intended outcome.

Supporting references

acceptance check
A stated test a receiving owner uses before accepting a handoff output.

Supporting references

acceptance criterion
An observable condition the revised result must meet before it is accepted.

Supporting references

acceptance example
A concrete input, action, and observable outcome used to decide whether a feature meets its brief.

Supporting references

access scope
The specific data, folders, services, or actions a tool is permitted to reach.

Supporting references

  • OpenAI Help Center: Memory FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Memory · Documentation reviewed 2026-07-20 · Memory behavior depends on settings, plan, workspace policy, and product changes. It is not an organizational permission decision.

  • OpenAI Help Center: Projects in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT Projects · Documentation reviewed 2026-07-20 · Project context and memory vary by plan and settings; this source does not approve any data for use.

  • OpenAI Help Center: Apps in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT apps and connected services · Documentation reviewed 2026-07-20 · Availability, actions, data use, scopes, and admin controls vary by app, plan, workspace, and configuration.

  • OpenAI Help Center: Data Controls FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT consumer services · Documentation reviewed 2026-07-20 · Consumer Data Controls behavior only. Business, Enterprise, Edu, API, legal obligations, and an employer's approval policy are outside this source's scope.

account handoff
A traceable record of evidence, communication state, open questions, ownership, and allowed next actions.

Supporting references

account hypothesis
A tentative, evidence-linked implication about a fictional organization that remains unconfirmed.

Supporting references

action ledger
A structured record of actions, owners, dates, status, dependencies, and source evidence.

Supporting references

action state
A precise label such as proposed, approved, completed, failed, or not attempted.

Supporting references

adoption gate
A required evidence and review decision before an organizational change may expand to another stage.

Supporting references

adoption signal
Observed evidence about use, outcomes, burden, confidence, or failure that informs an adoption decision.

Supporting references

adoption stage
A bounded period in which a defined group tests a use under explicit evidence, capacity, and stopping rules.

Supporting references

agent tool
A capability that lets an assistant inspect a resource or perform an action beyond producing chat text.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

agentic workflow
A workflow that can select or sequence actions toward a goal within defined boundaries.

Supporting references

AI operating plan
A proposed system connecting organizational outcomes, use cases, decision rights, evidence, resources, controls, adoption, and exit.

Supporting references

alternative method
A different way to access or complete a process when the default method creates a barrier.

Supporting references

angle card
A compact content idea containing an audience tension, bounded claim, supporting evidence, and intended format.

Supporting references

application evidence
Observable output or interface information showing that required skill rules were used.

Supporting references

  • OpenAI Help Center: Skills in ChatGPT

    Reviewed: 2026-07-20

    Scope: OpenAI Skills · Documentation reviewed 2026-07-20 · Skill availability, installation, discovery, and invocation vary by OpenAI product and workspace configuration.

  • OpenAI Help Center: Plugins in Codex

    Reviewed: 2026-07-20

    Scope: Codex plugins and apps · Documentation reviewed 2026-07-20 · Plugin capability depends on included skills, enabled apps, app permissions, workspace policy, and product surface.

approval gate
A required human decision placed immediately before a consequential action, supported by an exact preview.

Supporting references

approval mode
The configured rule for when an agent must obtain confirmation before an action.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

  • OpenAI Help Center: Data Controls FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT consumer services · Documentation reviewed 2026-07-20 · Consumer Data Controls behavior only. Business, Enterprise, Edu, API, legal obligations, and an employer's approval policy are outside this source's scope.

approved system
A tool or environment authorized by the responsible organization for a defined information classification and use.

Supporting references

approved-use boundary
An organization-authorized statement of which purpose, data, system, actions, owners, and period are permitted.

Supporting references

asset retirement
The controlled end of an asset's current use, including replacement guidance, history, inventory, and verification.

Supporting references

assumption to test
A plausible but unsupported belief kept visibly separate from verified audience evidence.

Supporting references

audience moment
The concrete situation in which a reader needs the content and is ready to act on it.

Supporting references

authoritative guidance
Practice or risk guidance issued by a recognized body within its remit, distinct from original empirical evidence.

Supporting references

authoritative source
A primary or otherwise accountable source suited to establishing the claim being checked.

Supporting references

authoritative source
A source with recognized responsibility or expertise for the claim’s domain and scope.

Supporting references

Authoritative source
A current primary or official source that is responsible for, or directly evidences, the claim being checked.

Supporting references

authorization
Permission from an accountable owner or policy, distinct from technical ability.

Supporting references

authorization mechanism
The specific documented process used to grant an integration access.

Supporting references

  • OpenAI Help Center: Apps in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT apps and connected services · Documentation reviewed 2026-07-20 · Availability, actions, data use, scopes, and admin controls vary by app, plan, workspace, and configuration.

  • OpenAI Help Center: Plugins in Codex

    Reviewed: 2026-07-20

    Scope: Codex plugins and apps · Documentation reviewed 2026-07-20 · Plugin capability depends on included skills, enabled apps, app permissions, workspace policy, and product surface.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

automation readiness
Evidence that a stable process has measurable quality, bounded failures, review ownership, and a rollback path.

Supporting references

base rate
The underlying frequency or starting level used to interpret change or risk.

Supporting references

bearer credential
A secret whose holder may be able to act with its assigned authority.

Supporting references

behavior anchor
An observable example that clarifies what a rubric level means without inferring hidden traits.

Supporting references

behavior trace
A map following an action through interfaces, contracts, state, persistence, and observable output.

Supporting references

blocking branch
A research branch whose evidence threshold must be met before the intended decision can proceed.

Supporting references

blocking criterion
A requirement whose failure prevents acceptance regardless of other scores.

Supporting references

blocking finding
A review issue that must be resolved or explicitly accepted by an authorized owner before progression.

Supporting references

boundary case
An input at an edge such as zero, blank, period end, or category transition that can expose faulty logic.

Supporting references

boundary validation
Checking data as it enters or leaves a step so malformed records do not spread through the workflow.

Supporting references

bounded delegation
Assignment of a narrow objective with explicit inputs, ownership, constraints, evidence, and completion criteria.

Supporting references

bounded task
A delegation with explicit scope, ownership, inputs, deliverable, constraints, and completion evidence.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

break-even point
The point where work saved through delegation becomes greater than the coordination effort it adds.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

buying signal
Evidence that may indicate interest or change, which requires source, scope, and validation before use.

Supporting references

buying trigger
An observable change that may make a relevant problem more timely without proving pain or intent.

Supporting references

calibrated confidence
A person's confidence considered alongside demonstrated accuracy and appropriate uncertainty behavior.

Supporting references

calibration case
A shared example used to align reviewers on rubric interpretation and decision reasoning.

Supporting references

candidate slice
A bounded part of a workflow evaluated for assistance independently from the whole process.

Supporting references

candidate window
A proposed time derived from stated constraints that remains unconfirmed until a human approves it.

Supporting references

capability boundary
The specific limit of what a system can access or do for a defined task and environment.

Supporting references

capability surface
The specific set of operations an integration exposes in its current configuration.

Supporting references

capstone hypothesis
A testable claim that one bounded assistive change will improve a stated task outcome under defined conditions.

Supporting references

capture plan
A pre-recording specification covering purpose, participants, consent, minimization, storage, retention, and review.

Supporting references

case mix
The composition of observations across categories or difficulty levels that can affect aggregate results.

Supporting references

change preview
A before-and-after proposal showing targets, consequences, alternatives, and required approval before execution.

Supporting references

channel-native adaptation
A version redesigned for a format's audience and job without changing the supported meaning.

Supporting references

chart claim
A statement about a visualized pattern that must remain within the underlying data and design.

Supporting references

claim audit
A systematic check that every consequential statement is supported by evidence matching its exact wording and scope.

Supporting references

claim boundary
The limit of what approved evidence can support without inference inflation.

Supporting references

claim headline
A complete sentence stating the evidence-backed takeaway a slide is meant to establish.

Supporting references

claim ledger
A source-of-truth record of claims, evidence, limitations, quotations, and rights used across derivative drafts.

Supporting references

claim matrix
A table connecting each customer-facing claim to approved evidence, scope, freshness, and review status.

Supporting references

Claim scope
The product, version, place, time, audience, and conditions within which the available evidence supports a claim.

Paper Trail working definition. A supporting external reference has not been assigned.

claim status
A label distinguishing observed, calculated, assumed, proposed, approved, and unresolved statements.

Supporting references

claim strength
The degree of certainty and causality expressed by wording, which should not exceed the supporting evidence.

Supporting references

claim-specific freshness
A recency requirement chosen according to how quickly the fact or behavior being claimed can change.

Supporting references

clean re-brief
A self-contained correction that restates the goal, failed criterion, desired result, and useful example.

Supporting references

clean restart
A new attempt with corrected scope, evidence, method, or ownership.

Supporting references

close incident
A failure affecting a period-end financial or operational reporting workflow and its review evidence.

Supporting references

close-the-loop draft
A respectful final message draft that releases pressure and records no assumption of continued interest.

Supporting references

command center
A linked operational view of decisions, commitments, actions, schedule, status, and unresolved issues.

Supporting references

command line
A text interface for issuing commands to tools or the operating system.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

commitment inference
An unsupported conversion of tentative, descriptive, or ambiguous language into an assigned obligation.

Supporting references

comparable unit
A matched topic, rule, field, or claim that can be meaningfully assessed across sources.

Supporting references

compatibility rule
A declared policy for how producers and receivers handle records created under different contract versions.

Supporting references

competing explanation
An alternative account of an observed pattern that could change interpretation or action.

Supporting references

conditional conclusion
A conclusion whose supported population, setting, assumptions, and uncertainty are stated explicitly.

Supporting references

confabulation
Plausible generated content presented without adequate grounding.

Supporting references

Confabulation
Generated content presented with confidence even though it is erroneous, false, inconsistent, or divergent from the input.

Supporting references

confirmation status
A field showing whether a proposed owner, deadline, or decision has been explicitly accepted by the relevant human.

Supporting references

conflict matrix
A structured comparison of differing evidence and its decision consequence.

Supporting references

confounder
A factor related to both exposure and outcome that can offer an alternative explanation.

Supporting references

connector
A configured integration that lets a product access defined data or operations in another system.

Supporting references

consequence of delay
The specific harm or missed opportunity that could occur if a request is not handled by its actual deadline.

Supporting references

consequential action
An operation such as sending, deleting, publishing, paying, or updating a system whose effects may be hard to reverse.

Supporting references

consequential assessment
Assessment used to determine grades, placement, credentials, access, progression, or other meaningful learner outcomes.

Supporting references

construct alignment
The match among intended learning, instruction, practice, evidence, and evaluation criteria.

Supporting references

construct equivalence
Comparable evidence of the same intended learning outcome across different accessible formats.

Supporting references

content pillar
A recurring evidence-backed promise to help a defined audience with one important problem.

Supporting references

content provenance
A trace of the source elements and decisions behind an adapted asset.

Supporting references

context boundary
The set of account, device, workspace, storage, and policy controls that separate one body of information from another.

Supporting references

  • OpenAI Help Center: Memory FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Memory · Documentation reviewed 2026-07-20 · Memory behavior depends on settings, plan, workspace policy, and product changes. It is not an organizational permission decision.

  • OpenAI Help Center: Projects in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT Projects · Documentation reviewed 2026-07-20 · Project context and memory vary by plan and settings; this source does not approve any data for use.

  • OpenAI Help Center: Apps in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT apps and connected services · Documentation reviewed 2026-07-20 · Availability, actions, data use, scopes, and admin controls vary by app, plan, workspace, and configuration.

  • OpenAI Help Center: Data Controls FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT consumer services · Documentation reviewed 2026-07-20 · Consumer Data Controls behavior only. Business, Enterprise, Edu, API, legal obligations, and an employer's approval policy are outside this source's scope.

context drift
A loss of alignment between the active task and the assumptions guiding current work.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

context rot
An informal name for declining clarity when stale or competing material accumulates around a task.

Supporting references

  • OpenAI Help Center: Memory FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Memory · Documentation reviewed 2026-07-20 · Memory behavior depends on settings, plan, workspace policy, and product changes. It is not an organizational permission decision.

  • OpenAI Help Center: Projects in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT Projects · Documentation reviewed 2026-07-20 · Project context and memory vary by plan and settings; this source does not approve any data for use.

context window
A model-dependent limit on the amount of input and conversation material processed together.

Supporting references

  • OpenAI Help Center: Memory FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Memory · Documentation reviewed 2026-07-20 · Memory behavior depends on settings, plan, workspace policy, and product changes. It is not an organizational permission decision.

  • OpenAI Help Center: Projects in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT Projects · Documentation reviewed 2026-07-20 · Project context and memory vary by plan and settings; this source does not approve any data for use.

  • OpenAI Help Center: Temporary Chat FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Temporary Chat for consumers · Documentation reviewed 2026-07-20 · Covers current history, memory, model-improvement, safety-retention, and GPT-action behavior; it is not evidence that a particular organization's data is approved for use.

contrast example
A sample that shows what to avoid so the desired boundary is easier to see.

Supporting references

control evidence
A retained record showing what procedure occurred, on which population, by whom, when, and with what result.

Supporting references

control sheet
A versioned operational specification connecting each workflow transition to its conditions, limits, evidence, and next state.

Supporting references

control total
A known count or aggregate used to detect dropped, duplicated, or changed values after transformation.

Supporting references

control total
An independently supplied aggregate used to check the completeness or accuracy of a transferred dataset.

Supporting references

control total
An independently known count or sum used to test completeness.

Supporting references

control total
An independently obtained total used to detect missing, duplicated, or misclassified values.

Supporting references

controlled replay
Rerunning preserved synthetic fixtures in a local test environment to verify a fix and check regressions.

Supporting references

controlled replay
A rerun using frozen cases while varying one candidate cause at a time.

Supporting references

coordination overhead
The briefing, communication, review, conflict-resolution, and integration work added by delegation.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

correction spiral
A sequence of dependent patches that adds confusion without resolving the underlying failure.

Supporting references

corroboration
Independent evidence that supports or challenges a claim made from a visual source.

Supporting references

coverage gap
A decision-relevant question part lacking sufficient applicable evidence.

Supporting references

creative hypothesis
A testable expectation about how one controlled creative choice may affect comprehension or response.

Supporting references

credential
A secret used to authenticate or authorize access, such as a password, private key, token, or recovery code.

Supporting references

critical failure
A failure severe enough to block acceptance regardless of aggregate performance.

Supporting references

critique ledger
A structured record of observations, criteria, impact, priority, disposition, and revision requests.

Supporting references

cross-cutting validation
A check applied to the combined result rather than to one workstream alone.

Supporting references

cross-format check
A test that essential meaning remains available when audio, visuals, or another channel is unavailable.

Supporting references

cue sheet
A timed record of spoken text, music, effects, sources, rights, and review notes for an audio asset.

Supporting references

cut plan
A documented way to remove a blocked or weak scene while preserving the video's essential message.

Supporting references

data contract
An explicit agreement about the names, types, meaning, and validation rules of fields exchanged between steps.

Supporting references

data dictionary
A reference defining each field's meaning, type, valid values, format, and missing-data rule.

Supporting references

data lineage
The trace from source material through extraction, transformation, calculation, visualization, and reported claim.

Supporting references

data lineage
The trace from source data through transformations and calculations to an output claim.

Supporting references

data minimization
Providing only the information necessary for the task instead of sharing everything available.

Supporting references

Data minimization
Limiting an input to the smallest amount and least sensitive form of information needed for the stated task.

Paper Trail working definition. A supporting external reference has not been assigned.

decision boundary
The limit of action justified by available evidence, scope, authority, and uncertainty.

Supporting references

decision brief
A traceable synthesis of evidence, assumptions, options, uncertainties, and ownership prepared for an accountable decision maker.

Supporting references

decision memo
A concise, evidence-linked artifact that makes a specific choice, options, recommendation, and uncertainty reviewable.

Supporting references

decision narrative
A traceable explanation of evidence, options, tradeoffs, authority, and the exact decision requested.

Supporting references

decision question
A bounded evidence-seeking question connected to options an accountable owner can choose.

Supporting references

decision queue
A reviewable list that ranks messages and explains the proposed next action without carrying it out.

Supporting references

decision queue
A prioritized list of questions requiring an accountable owner before work can proceed.

Supporting references

decision right
A defined authority to propose, review, approve, operate, change, or stop a specific use.

Supporting references

decision-deck package
A local artifact joining a decision brief, storyline, slide evidence, accessibility, speaker handoff, and review record.

Supporting references

delimiter
A visible marker separating sections or reference blocks in a prompt.

Supporting references

denominator
The reference total or population used for a rate, fraction, or proportion.

Supporting references

denominator
The population or quantity used as the base of a rate, proportion, or normalized comparison.

Supporting references

dependency disposition
The recorded decision to approve, replace, defer, or reject a dependency with evidence and ownership.

Supporting references

dependency graph
A map showing which task outputs must exist before other tasks can begin.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

detectability
How likely a defined review process is to notice a failure before use.

Supporting references

deterministic automation
A predefined rule sequence that follows fixed conditions from a trigger.

Supporting references

deterministic check
A reproducible test that returns the same result for the same input and rule.

Supporting references

deterministic tool
A method designed to follow fixed rules and produce repeatable results for defined inputs.

Supporting references

diff traceability
The connection from each changed hunk to an approved requirement, risk, or documented defect.

Supporting references

directness
How closely a source's evidence addresses the exact population, setting, measure, and claim under review.

Supporting references

disconfirming check
A test intended to reveal that a claim is false, outdated, or inapplicable.

Supporting references

dry run
A test that produces observable proposed results without performing external side effects.

Supporting references

durable record
A maintained artifact intended to remain useful beyond the current session.

Supporting references

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

earned specificity
Concrete detail supported by approved evidence or firsthand material rather than invented to make prose vivid.

Supporting references

edit contract
A task specification that defines source rights, allowed and prohibited changes, disclosure, accessibility, and acceptance tests.

Supporting references

enablement blocker
An unresolved condition that must prevent a workflow from being activated.

Supporting references

engagement proxy
An activity measure that may correlate with participation but does not by itself demonstrate learning.

Supporting references

entailment
A relation in which evidence within context is sufficient to support a claim.

Supporting references

equivalent evidence
Different accessible artifact forms that demonstrate the same intended capability without changing its standard.

Supporting references

escalation packet
A compact record of context, evidence, uncertainty, consequence, and requested decision.

Supporting references

essential meaning
Information or relationships a learner must access to meet the intended objective.

Supporting references

eval case
A fixed input, expected behavior, and scoring rule used to assess a system or workflow.

Supporting references

evidence applicability
The extent to which evidence covers the exact service, use, period, and requirement under review.

Supporting references

evidence chain
The trace from a requirement through implementation and validation plans to a bounded release decision.

Supporting references

evidence layer
The detailed artifacts that allow a portfolio claim, result, failure, or decision to be inspected and reproduced.

Supporting references

evidence lineage
The underlying origin and dependency chain shared by one or more sources.

Supporting references

evidence synthesis
A traceable organization of source observations by criterion without converting them into an unauthorized decision.

Supporting references

evidence trace
A reviewer test that follows a slide claim through its locator and interpretation back to the source.

Supporting references

evidence trail
A record connecting a claim to its source, review date, and verification result.

Supporting references

evidence unit
A bounded finding from one source connected to one research claim and its limitations.

Supporting references

evidence weighting
Judging sources by relevance, quality, recency, directness, and independence rather than treating each as an equal vote.

Supporting references

exception log
A record of unresolved source, extraction, interpretation, or rendering issues that require review.

Supporting references

exception work
The effort required to recognize and handle cases that do not follow the normal workflow path.

Supporting references

executive decision pack
A concise decision view backed by reproducible evidence, analysis, limitations, and accountable review.

Supporting references

executive signal
A change, decision, risk, or outcome material enough to deserve leadership attention.

Supporting references

fabricated personalization
A tailored-looking statement built from invented, irrelevant, or unsupported details about a person or organization.

Supporting references

factual drift
A claim change unsupported by a corresponding evidence change.

Supporting references

fading scaffold
Support intentionally reduced as a learner demonstrates greater independent control of a task.

Supporting references

failure mode
A specific way a workflow can produce an unacceptable result or state.

Supporting references

failure shape
The way a failure appears, spreads, and can be detected or contained in a workflow.

Supporting references

fallback path
The explicit route for data or runs that do not satisfy the expected condition.

Supporting references

falsification signal
Observed evidence indicating that an assumption no longer supports the scenario in which it is used.

Supporting references

feasibility gate
A required condition a candidate must satisfy before benefit-based ranking is appropriate.

Supporting references

few-shot example
A demonstration showing the intended relationship between input and output.

Supporting references

file allowlist
The explicit set of files or artifact types authorized for a planned change.

Supporting references

finance review package
A versioned collection of source, calculation, reconciliation, control, scenario, and decision-support drafts for qualified review.

Supporting references

first divergence
The earliest workflow step where the actual result differs from the expected result.

Supporting references

first divergence
The earliest point where observed workflow state differs from the expected state.

Supporting references

fit signal
Evidence that a fictional organization has the process, need, and constraints an offer is designed to serve.

Supporting references

fixture pack
A versioned set of inputs and expected outcomes used to test a contract repeatedly.

Supporting references

formative assessment
Low-stakes evidence used to guide instruction, practice, feedback, and learner reflection.

Supporting references

formula lineage
The recorded source fields and calculation steps that produced a derived value used in a visual or claim.

Supporting references

fresh brief
A self-contained statement of the current goal, constraints, sources, and next action.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

freshness trigger
An event or date that prompts a claim to be reviewed again.

Supporting references

full cost
The combined setup, operating, review, failure, maintenance, adoption, and exit cost of a use case.

Supporting references

functional alt text
Alternative text written for the image's purpose in context rather than as an exhaustive inventory of pixels.

Supporting references

governance packet
A versioned collection of use-case evidence and control proposals prepared for authorized expert and organizational review.

Supporting references

governing question
The central decision-focused question that determines what belongs in a presentation.

Supporting references

handoff check
A test that ensures meaning and evidence survive transfer between stages.

Supporting references

handoff state
The concise current status and next-step information needed to resume work.

Supporting references

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

human authorship decision
A meaning-bearing creative choice intentionally selected and owned by a person rather than delegated without review.

Supporting references

hunk annotation
A note explaining the purpose, evidence, risk, and test relationship of one contiguous patch section.

Supporting references

idempotency
A design property that prevents the same event or retry from producing duplicate effects.

Supporting references

idempotency
A design property that prevents a repeated trigger from duplicating an intended effect.

Supporting references

idempotency key
A stable identifier used to prevent the same logical request from producing duplicate actions.

Supporting references

idempotent action
An action designed so repeating it with the same key does not create an additional effect.

Supporting references

independent recalculation
A check from source inputs using a method separate from the original.

Supporting references

independent verification
A check using evidence or method separate from the process that produced a claim.

Supporting references

indirect identifier
A detail that can reveal a person or case when combined with other available information.

Supporting references

inference inflation
A claim broader, more certain, or more causal than evidence permits.

Supporting references

input method
The mechanism used to provide material, such as pasted text, an uploaded file, or a scoped connection.

Supporting references

  • OpenAI Help Center: Apps in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT apps and connected services · Documentation reviewed 2026-07-20 · Availability, actions, data use, scopes, and admin controls vary by app, plan, workspace, and configuration.

  • OpenAI Help Center: File storage and Library in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT file storage and Library · Documentation reviewed 2026-07-20 · File retention and use depend on product, plan, workspace retention, and data-control settings.

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI Help Center: Data Controls FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT consumer services · Documentation reviewed 2026-07-20 · Consumer Data Controls behavior only. Business, Enterprise, Edu, API, legal obligations, and an employer's approval policy are outside this source's scope.

instruction precedence
The rule for deciding which applicable instruction governs when directions conflict.

Supporting references

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

instruction scope
The files, directories, or tasks to which an instruction applies.

Supporting references

instruction scope
The files, directory, workflow, people, or decisions to which a standing rule applies.

Supporting references

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

instruction-content boundary
The distinction between governing directions and material analyzed only as data.

Supporting references

insufficient evidence
A review state indicating that available information cannot support a rubric judgment.

Supporting references

integration gate
A check that all required validated inputs are present and compatible before results are combined.

Supporting references

integration ledger
A review record linking each delegated result to evidence, conflicts, decisions, and final validation.

Supporting references

investment learning case
A staged evidence proposal for testing value, cost, and risk before an authorized funding decision.

Supporting references

invocation
The tool-specific act or matching process that selects a skill for a task.

Supporting references

  • OpenAI Help Center: Skills in ChatGPT

    Reviewed: 2026-07-20

    Scope: OpenAI Skills · Documentation reviewed 2026-07-20 · Skill availability, installation, discovery, and invocation vary by OpenAI product and workspace configuration.

  • OpenAI Help Center: Plugins in Codex

    Reviewed: 2026-07-20

    Scope: Codex plugins and apps · Documentation reviewed 2026-07-20 · Plugin capability depends on included skills, enabled apps, app permissions, workspace policy, and product surface.

item validity
The extent to which an assessment task supports the intended interpretation for its stated purpose.

Supporting references

job relevance
A supported connection between a criterion or assessment and observable requirements of the work.

Supporting references

kill switch
A tested control that stops new runs or prevents further effects when a workflow behaves unexpectedly.

Supporting references

knowledge page
A maintained document that answers one durable project question.

Supporting references

knowledge promotion
The controlled process of turning reviewed source material into reusable, maintained knowledge.

Supporting references

known state
A documented workflow condition whose completed checks, owner, and permitted next actions are understood.

Supporting references

layout verification
Comparing extracted content with the rendered page to confirm reading order, headers, signs, units, and footnotes.

Supporting references

learner agency
A learner's meaningful ability to understand, question, choose, and act within a learning process.

Supporting references

learner context
Relevant task, prior-opportunity, access, support, and environment information used cautiously in learning design.

Supporting references

learning budget
A proposed cap on time, money, scope, or cases for a reversible evidence-gathering stage.

Supporting references

learning evidence
Observable task work used to guide instruction, distinct from inferred traits or an employment decision.

Supporting references

learning review package
A versioned set of aligned instructional, access, assessment, feedback, and evaluation drafts for authorized expert review.

Supporting references

learning threshold
A predeclared measured result required to justify another bounded stage of testing.

Supporting references

least privilege
Giving a tool only the access needed for the current task and no more.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

least privilege
Granting only the access needed for a defined task and no more.

Supporting references

listening review
A human check of intelligibility, timing, pronunciation, artifacts, meaning, and source fidelity in rendered audio.

Supporting references

managed device
A device whose software, access, and data-handling rules are controlled by an organization.

Supporting references

  • OpenAI Help Center: Memory FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Memory · Documentation reviewed 2026-07-20 · Memory behavior depends on settings, plan, workspace policy, and product changes. It is not an organizational permission decision.

  • OpenAI Help Center: Projects in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT Projects · Documentation reviewed 2026-07-20 · Project context and memory vary by plan and settings; this source does not approve any data for use.

  • OpenAI Help Center: Apps in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT apps and connected services · Documentation reviewed 2026-07-20 · Availability, actions, data use, scopes, and admin controls vary by app, plan, workspace, and configuration.

  • OpenAI Help Center: Data Controls FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT consumer services · Documentation reviewed 2026-07-20 · Consumer Data Controls behavior only. Business, Enterprise, Edu, API, legal obligations, and an employer's approval policy are outside this source's scope.

management assumption
A planning input supplied by an authorized human owner rather than directly observed or independently verified.

Supporting references

manual path
A maintained non-AI way to complete or obtain the service when the assisted path is unsuitable or unavailable.

Supporting references

markdown
A plain-text convention that uses lightweight markers to express document structure.

Supporting references

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

material change
A change to purpose, data, capability, system, population, consequence, or evidence that requires renewed review.

Supporting references

material claim
A claim that could change a decision, obligation, or action if wrong.

Supporting references

meaningful dissent
A substantiated disagreement that could affect the decision and must not be erased by synthesis.

Supporting references

media handoff package
A reviewable set of source, consent, transcript, edit, accessibility, disclosure, and approval records accompanying media.

Supporting references

media provenance
The record of a recording's origin, context, custody, edits, and intended use.

Supporting references

meeting decision
The specific choice, approval, or alignment the meeting is intended to produce.

Supporting references

message angle
A chosen framing that connects an audience need to a supported value proposition.

Supporting references

message architecture
The evidence-backed relationship among audience need, value, proof, action, and channel role.

Supporting references

message map
A hierarchy connecting a core conclusion to supporting claims, evidence, and uncertainty.

Supporting references

minimum exposure
Supplying the least amount of accessible data that still supports the task.

Supporting references

  • OpenAI Help Center: Apps in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT apps and connected services · Documentation reviewed 2026-07-20 · Availability, actions, data use, scopes, and admin controls vary by app, plan, workspace, and configuration.

  • OpenAI Help Center: File storage and Library in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT file storage and Library · Documentation reviewed 2026-07-20 · File retention and use depend on product, plan, workspace retention, and data-control settings.

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI Help Center: Data Controls FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT consumer services · Documentation reviewed 2026-07-20 · Consumer Data Controls behavior only. Business, Enterprise, Edu, API, legal obligations, and an employer's approval policy are outside this source's scope.

monitoring signal
An observed pattern that warrants qualified investigation but is not itself a legal or causal conclusion.

Supporting references

multimodal campaign system
A coordinated set of image, audio, video, variant, evidence, access, and review plans governed by one brief.

Supporting references

must-pass check
A criterion whose failure blocks acceptance until correction or escalation.

Supporting references

named human SME review
Review assigned to an identified person with relevant subject expertise and recognized organizational authority.

Supporting references

near miss
An output that appears strong but fails one important criterion.

Supporting references

need hypothesis
A tentative account need framed for validation rather than asserted as buyer truth.

Supporting references

no-action default
A rule that timeout, silence, ambiguity, or failure cannot trigger the consequential action.

Supporting references

no-authorization statement
An explicit record of actions and decisions the draft does not permit or claim to complete.

Supporting references

non-color cue
A label, pattern, shape, text, or position that communicates meaning without depending on color perception alone.

Supporting references

non-goal
An explicitly excluded outcome or capability that prevents a project from silently expanding its scope.

Supporting references

nondeterminism
The possibility that similar generations produce different outputs.

Supporting references

normalization rule
A documented deterministic transformation from the original source representation to a standard stored value.

Supporting references

objection diagnosis
Clarifying what a concern actually means before choosing a response or next step.

Supporting references

observable outcome
A result that a named owner can inspect against stated success criteria.

Supporting references

occlusion
A condition in which part of a visual subject or label is hidden by another element or by the crop.

Supporting references

OCR uncertainty
Uncertainty introduced when an image of text is converted into machine-readable characters.

Supporting references

OCR uncertainty
The risk that optical character recognition has misread, omitted, or reordered visible content.

Supporting references

operating effectiveness
A qualified conclusion about whether a control operated as designed over an appropriate period and population.

Supporting references

operating ownership
Accountability for maintaining a use, its evidence, controls, decisions, incidents, and retirement over time.

Supporting references

operating specification
The complete, versioned definition of how a workflow should run, be tested, be observed, and be governed.

Supporting references

operational constraint
A condition that affects what work can occur without exposing the underlying sensitive detail.

Supporting references

operational handoff
A transfer record covering ownership, prerequisites, reproduction, limits, maintenance, recovery, next decisions, and retirement.

Supporting references

operations evidence pack
A versioned collection of records supporting review of a workflow's measured behavior and controls.

Supporting references

opportunity to learn
The instruction, practice, resources, and access a learner has received before performance is interpreted.

Supporting references

orchestrator
The accountable agent or person who decomposes, assigns, reviews, and integrates delegated work.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

organizational commitment
An authorized decision that allocates policy, funding, people, systems, or accountable action.

Supporting references

organizational deliberation
A governed human process for reviewing evidence and making decisions through assigned authority.

Supporting references

output contract
An explicit description of the structure and quality a result must satisfy.

Supporting references

  • OpenAI Help Center: Skills in ChatGPT

    Reviewed: 2026-07-20

    Scope: OpenAI Skills · Documentation reviewed 2026-07-20 · Skill availability, installation, discovery, and invocation vary by OpenAI product and workspace configuration.

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

output format
The requested structure, length, and organization of an answer.

Supporting references

output shape
The organization, fields, sequence, and length required to support inspection and use.

Supporting references

outward action
An action that sends, publishes, uploads, shares, or otherwise moves information beyond the current local task boundary.

Supporting references

ownership boundary
The files, decisions, or artifacts one task may change and those it must leave alone.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

paired locator
A page, section, table, or paragraph reference recorded for each side of a comparison.

Supporting references

parking lot
A topic intentionally retained for later discussion without treating it as a current decision or action.

Supporting references

partial write
A state in which only some intended changes occurred before interruption or failure.

Supporting references

percentage point
The arithmetic difference between percentages, distinct from relative percent change.

Supporting references

pillar boundary
An explicit rule describing which ideas belong in a pillar and which should be excluded or redirected.

Supporting references

pipeline staleness
A lack of recent buyer evidence or a dated next action that signals a record needs review, recycling, or closure.

Supporting references

plausible continuation
Text that fits language and context patterns without necessarily being evidence-supported.

Supporting references

plug
An unsupported adjustment inserted only to force totals to agree.

Supporting references

portfolio gate
A required condition a use case must satisfy before benefit-based prioritization is appropriate.

Supporting references

Preflight
A short check completed before an AI task to record the data boundary, approved environment, evidence requirements, and safe next action.

Paper Trail working definition. A supporting external reference has not been assigned.

preserved behavior
An existing contract the change must leave intact even though nearby behavior is modified.

Supporting references

primary source
Original evidence or a direct official record that supports a claim at its source.

Supporting references

primary source
Original evidence or first-hand record whose relevance and quality still need evaluation.

Supporting references

problem frame
A bounded description of an outcome, current evidence, stakeholders, constraints, consequences, and decisions needed.

Supporting references

process disclosure
A bounded explanation of how information or automation is used in a workflow, based on approved facts.

Supporting references

process owner
The human authorized to confirm the workflow rules, exceptions, and acceptable outcomes.

Supporting references

product surface
The interface and surrounding features through which a person supplies context, invokes tools, and receives output.

Supporting references

prompt injection
An attempt inside processed content to redirect an AI system away from the authorized user instruction.

Supporting references

proof boundary
The sample, setting, period, and limitation that define how far a result can safely be generalized.

Supporting references

protected element
A logo, message, likeness, disclosure, or other feature that must not be altered outside defined limits.

Supporting references

protected revision element
A verified claim, right, disclosure, brand feature, or successful design choice that a revision must retain.

Supporting references

provenance
The recorded origin and path that let a reviewer trace a claim back to its supporting source.

Supporting references

provenance
Evidence about where a package came from and how its current contents were produced.

Supporting references

  • OpenAI Help Center: Skills in ChatGPT

    Reviewed: 2026-07-20

    Scope: OpenAI Skills · Documentation reviewed 2026-07-20 · Skill availability, installation, discovery, and invocation vary by OpenAI product and workspace configuration.

  • OpenAI Help Center: Plugins in Codex

    Reviewed: 2026-07-20

    Scope: Codex plugins and apps · Documentation reviewed 2026-07-20 · Plugin capability depends on included skills, enabled apps, app permissions, workspace policy, and product surface.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

provenance chain
The documented sequence from source material through edits, synthetic elements, variants, and selected output.

Supporting references

provisional feedback
Feedback explicitly limited by current evidence and open to correction through human review.

Supporting references

purpose change
Use of data or a system for an objective materially different from the one originally reviewed.

Supporting references

quarantine
A separate holding area for records that cannot be safely corrected without review.

Supporting references

query branching
Decomposition of a research question into distinct evidence-seeking paths.

Supporting references

raw material
An original or minimally changed input preserved for review, subject to storage and permission rules.

Supporting references

re-performance
Independent repetition of a procedure or check using preserved evidence to verify the result after change.

Supporting references

reach
The people, records, systems, or later decisions an output can affect.

Supporting references

readiness decision
An evidence-backed recommendation to continue authorized review, revise, or reject without implying deployment approval.

Supporting references

readiness disposition
An owned decision to continue bounded testing, revise and rerun, or stop based on stated evidence.

Supporting references

reading order
The sequence in which text and visual elements should be interpreted, which can differ from extraction order.

Supporting references

recommendation status
A label showing whether a recommendation is supported, conditional, deferred, or abstained.

Supporting references

reconciliation
Comparison of records and totals across stages with explanations for differences.

Supporting references

reconciliation failure
A mismatch between transformed output and a trusted count or control total beyond the predefined tolerance.

Supporting references

reconciling item
A documented difference explaining part of the relationship between two independently defined balances.

Supporting references

recovery path
A tested way to restore the prior state if a change is wrong.

Supporting references

reference example
A safe sample that demonstrates what a successful result should resemble.

Supporting references

refusal condition
A stated situation in which a skill must stop or return the decision to a person.

Supporting references

  • OpenAI Help Center: Skills in ChatGPT

    Reviewed: 2026-07-20

    Scope: OpenAI Skills · Documentation reviewed 2026-07-20 · Skill availability, installation, discovery, and invocation vary by OpenAI product and workspace configuration.

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

regression
A new defect introduced while changing previously acceptable output.

Supporting references

release evidence state
The exact status of a readiness item, such as proposed, executed, passing, reviewed, or approved.

Supporting references

release handoff
A package stating artifact status, completed checks, unresolved stops, and the authorized next owner.

Supporting references

relevance chain
A sequence linking verified context to a tentative implication, relevant outcome, credible proof, and small next step.

Supporting references

replay
A controlled reconstruction of a run from retained inputs, versions, decisions, and step evidence.

Supporting references

reporting checkpoint
The dated baseline against which a project update describes material change.

Supporting references

representative example
A sample that clearly demonstrates the particular quality the learner wants to transfer.

Supporting references

reproducibility
The ability of another reviewer to apply documented inputs and steps and obtain the same result.

Supporting references

requirement traceability
The explicit connection between a requirement or risk and the test evidence intended to cover it.

Supporting references

research closeout
A record explaining why research stopped and what action evidence supports.

Supporting references

research log
A trace of questions, queries, sources, findings, conflicts, and decisions.

Supporting references

reset brief
A compact statement of the current goal, decisions, constraints, and next step.

Supporting references

  • OpenAI Help Center: Memory FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Memory · Documentation reviewed 2026-07-20 · Memory behavior depends on settings, plan, workspace policy, and product changes. It is not an organizational permission decision.

  • OpenAI Help Center: Projects in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT Projects · Documentation reviewed 2026-07-20 · Project context and memory vary by plan and settings; this source does not approve any data for use.

residual variance
The portion of a variance not yet explained by supported reconciling drivers.

Supporting references

retirement signal
A predefined condition indicating that an asset should be reviewed, replaced, or removed from use.

Supporting references

retirement trigger
A condition indicating that a rule is obsolete, duplicated, or no longer belongs in the active instruction file.

Supporting references

retrieval
Selection of material from a source or collection for use in a response.

Supporting references

retry budget
A predeclared limit on repair attempts, time, or cost before a workflow pauses.

Supporting references

reversibility
How fully and quickly an action’s original state can be restored at acceptable cost.

Supporting references

review burden
The time and expertise required to detect and correct errors before use.

Supporting references

review capacity
The available qualified time and attention to evaluate outputs and exceptions against an agreed standard.

Supporting references

review packet
The proposal, evidence, target, uncertainty, and controls shown before an approval decision.

Supporting references

review register
A record of questions, named reviewers, supplied evidence, decisions, conditions, expiry, and follow-up ownership.

Supporting references

review-only patch
A proposed code change prepared for inspection but not applied, executed, or represented as validated.

Supporting references

reviewable action
An action whose target, proposed change, and consequence can be inspected before or during execution.

Supporting references

reviewer disagreement
A documented difference in human interpretation that remains visible until resolved through an approved process.

Supporting references

revision brief
A bounded instruction naming changes, preserved elements, evidence, and checks.

Supporting references

revision brief
A versioned plan linking observed evidence to one bounded change, expected effect, tests, owner, and rollback.

Supporting references

revision instruction
A focused request that identifies a defect and describes the exact change needed.

Supporting references

revision ledger
A traceable record connecting proposed instructional changes to evidence, ownership, tests, and decisions.

Supporting references

rollback path
The steps for disabling a connection and withdrawing its access.

Supporting references

  • OpenAI Help Center: Apps in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT apps and connected services · Documentation reviewed 2026-07-20 · Availability, actions, data use, scopes, and admin controls vary by app, plan, workspace, and configuration.

  • OpenAI Help Center: Plugins in Codex

    Reviewed: 2026-07-20

    Scope: Codex plugins and apps · Documentation reviewed 2026-07-20 · Plugin capability depends on included skills, enabled apps, app permissions, workspace policy, and product surface.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

rollback unit
The smallest coherent code, configuration, or data change that can be reversed to restore prior behavior.

Supporting references

rotation
Replacing a credential and retiring the previous value.

Supporting references

row grain
The real-world entity or event represented by one row of a table.

Supporting references

rubric anchor
A concrete description of observable performance at one scoring level.

Supporting references

run ledger
An ordered record of workflow inputs, actions, versions, outputs, checks, decisions, attempts, and states.

Supporting references

run receipt
A concise evidence record of one workflow run's inputs, decisions, action states, approvals, results, and errors.

Supporting references

run stamp
A compact record of report period, data-as-of time, run ID, versions, validation result, exceptions, and reviewer status.

Supporting references

safe crop
A framing boundary within which an image can be resized without losing essential meaning or protected elements.

Supporting references

safe degradation
A fallback that preserves controls and reduces capability when the preferred path is unavailable.

Supporting references

scenario assumption
A stated, versioned input or relationship used to explore a possible future outcome.

Supporting references

scenario suite
A set of representative tasks used to test whether standing instructions lead to consistent, safe decisions.

Supporting references

scene provenance
The source, rights, evidence, and transformation record attached to one storyboard scene.

Supporting references

schema
A definition of expected fields, types, relationships, and constraints.

Supporting references

schema invariant
A rule that every accepted record must satisfy regardless of how it was produced.

Supporting references

scoped disposition
A decision whose claim is limited to the exact task, cases, evidence, and conditions evaluated.

Supporting references

search result
A candidate returned from a collection, not yet established as applicable evidence.

Supporting references

selection proxy
An indirect signal used in place of direct evidence about job-related capability.

Supporting references

selection rate
The proportion of a clearly defined population receiving a specified selection-stage outcome.

Supporting references

selection review package
A versioned set of job, evidence, access, workflow, and monitoring drafts prepared for authorized expert review.

Supporting references

semantic validation
A check that values and relationships make sense and are supported beyond structure.

Supporting references

semantic validation
A check that a structurally valid value is plausible and meaningful for its intended use.

Supporting references

Sensitive information
Material whose unauthorized disclosure or use could harm a person, organization, or system, including personal, confidential, or access-control details.

Paper Trail working definition. A supporting external reference has not been assigned.

sensitivity case
A recalculation showing how a conclusion changes when an uncertain input or assumption changes.

Supporting references

sensitivity check
A test of how a decision changes when uncertain assumptions take different plausible values.

Supporting references

sensory-channel independence
The ability to access essential meaning without relying exclusively on sight, sound, color, or motion.

Supporting references

session handoff
A concise record that preserves the current state needed to resume focused work.

Supporting references

  • OpenAI Help Center: Memory FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Memory · Documentation reviewed 2026-07-20 · Memory behavior depends on settings, plan, workspace policy, and product changes. It is not an organizational permission decision.

  • OpenAI Help Center: Projects in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT Projects · Documentation reviewed 2026-07-20 · Project context and memory vary by plan and settings; this source does not approve any data for use.

  • OpenAI Help Center: Temporary Chat FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Temporary Chat for consumers · Documentation reviewed 2026-07-20 · Covers current history, memory, model-improvement, safety-retention, and GPT-action behavior; it is not evidence that a particular organization's data is approved for use.

shared asset
A maintained reusable package that supports one bounded task with instructions, examples, checks, and ownership.

Supporting references

sign-off boundary
The point at which only an authorized named human may record review, approval, or a qualified conclusion.

Supporting references

skill
A tool-specific reusable instruction package for a bounded task.

Supporting references

  • OpenAI Help Center: Skills in ChatGPT

    Reviewed: 2026-07-20

    Scope: OpenAI Skills · Documentation reviewed 2026-07-20 · Skill availability, installation, discovery, and invocation vary by OpenAI product and workspace configuration.

  • OpenAI Help Center: Plugins in Codex

    Reviewed: 2026-07-20

    Scope: Codex plugins and apps · Documentation reviewed 2026-07-20 · Plugin capability depends on included skills, enabled apps, app permissions, workspace policy, and product surface.

slide evidence ledger
A record linking each planned slide claim to its source, scope, visual, uncertainty, and implication.

Supporting references

software change dossier
A reviewable local package connecting product need, planned files, proposed patch, tests, risks, decisions, and recovery.

Supporting references

source ledger
A record of each creative input, its origin, rights status, allowed use, attribution, and unresolved questions.

Supporting references

source ledger
A record connecting sources and locations to findings, scope, decisions, and freshness.

Supporting references

source lineage
The chain showing whether multiple publications derive from the same original evidence or from independent work.

Supporting references

source location
A reproducible pointer such as document, page, section, table, row, column, or footnote.

Supporting references

source of truth
The designated record that controls when multiple versions or recollections conflict.

Supporting references

  • OpenAI Help Center: Memory FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Memory · Documentation reviewed 2026-07-20 · Memory behavior depends on settings, plan, workspace policy, and product changes. It is not an organizational permission decision.

  • OpenAI Help Center: Projects in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT Projects · Documentation reviewed 2026-07-20 · Project context and memory vary by plan and settings; this source does not approve any data for use.

  • OpenAI Help Center: Temporary Chat FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Temporary Chat for consumers · Documentation reviewed 2026-07-20 · Covers current history, memory, model-improvement, safety-retention, and GPT-action behavior; it is not evidence that a particular organization's data is approved for use.

source of truth
The designated version of a record that should be consulted when copies differ.

Supporting references

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

source provenance
Information about where a claim or piece of context came from.

Supporting references

  • OpenAI Help Center: Memory FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Memory · Documentation reviewed 2026-07-20 · Memory behavior depends on settings, plan, workspace policy, and product changes. It is not an organizational permission decision.

  • OpenAI Help Center: Projects in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT Projects · Documentation reviewed 2026-07-20 · Project context and memory vary by plan and settings; this source does not approve any data for use.

  • OpenAI Help Center: Apps in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT apps and connected services · Documentation reviewed 2026-07-20 · Availability, actions, data use, scopes, and admin controls vary by app, plan, workspace, and configuration.

  • OpenAI Help Center: Data Controls FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT consumer services · Documentation reviewed 2026-07-20 · Consumer Data Controls behavior only. Business, Enterprise, Edu, API, legal obligations, and an employer's approval policy are outside this source's scope.

source role
The purpose a source serves, such as governing evidence, data, background, or lead.

Supporting references

source-value ledger
A traceable table preserving reported values and metadata separately from transformations and interpretation.

Supporting references

speaker attribution
The assignment of a spoken segment to a participant, with uncertainty preserved when identity is not established.

Supporting references

speaker handoff
The notes and evidence context that let another presenter deliver the deck accurately without inventing explanations.

Supporting references

stable slice
A bounded workflow segment with repeatable inputs, explicit rules, and containable failure modes.

Supporting references

stage exit evidence
The buyer-verified facts required before an opportunity can legitimately move to the next stage.

Supporting references

staged learning
A sequence of bounded tests in which evidence and review are required before scope can expand.

Supporting references

staged test
A narrow first run designed to reveal problems while limiting consequences.

Supporting references

  • OpenAI Help Center: Skills in ChatGPT

    Reviewed: 2026-07-20

    Scope: OpenAI Skills · Documentation reviewed 2026-07-20 · Skill availability, installation, discovery, and invocation vary by OpenAI product and workspace configuration.

  • OpenAI Help Center: Plugins in Codex

    Reviewed: 2026-07-20

    Scope: Codex plugins and apps · Documentation reviewed 2026-07-20 · Plugin capability depends on included skills, enabled apps, app permissions, workspace policy, and product surface.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

standing instruction
A durable rule that should guide repeated work rather than one isolated task.

Supporting references

status freshness
The date and evidence showing when an operational status was last confirmed.

Supporting references

status rule
A predefined evidence standard that determines labels such as on track, at risk, or blocked.

Supporting references

stop authority
Explicit organizational authority to pause or end a use when a defined condition occurs.

Supporting references

stop rule
A predefined observable condition that pauses a workflow before further action.

Supporting references

storyboard provenance
Source and transformation notes attached to each planned scene, sound, quotation, or claim.

Supporting references

structured interview
An interview using comparable job-related questions, probes, evidence records, and scoring anchors.

Supporting references

structured selection
A process that asks comparable job-related questions and evaluates evidence against predeclared anchors.

Supporting references

style transfer
Applying observable writing characteristics from a reference to new content without carrying over its facts.

Supporting references

  • OpenAI Help Center: Memory FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Memory · Documentation reviewed 2026-07-20 · Memory behavior depends on settings, plan, workspace policy, and product changes. It is not an organizational permission decision.

  • OpenAI Help Center: Projects in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT Projects · Documentation reviewed 2026-07-20 · Project context and memory vary by plan and settings; this source does not approve any data for use.

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

  • OpenAI Help Center: Data Controls FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT consumer services · Documentation reviewed 2026-07-20 · Consumer Data Controls behavior only. Business, Enterprise, Edu, API, legal obligations, and an employer's approval policy are outside this source's scope.

success criterion
An observable condition used to decide whether output meets a defined need.

Supporting references

superseding record
A newer authoritative record that replaces an earlier decision or instruction.

Supporting references

supersession
An explicit relationship in which a newer source replaces an earlier one for defined scope.

Supporting references

supported performance
Task practice completed with available guidance, feedback, and access supports before independent expectations apply.

Supporting references

surface imitation
Copying visible wording or format without demonstrating the underlying reasoning or capability.

Supporting references

synthesis
A newly organized explanation built from inputs, which still requires evidence for factual claims.

Supporting references

task context
The facts and constraints a collaborator needs to complete one specific job.

Supporting references

task decomposition
Division of an outcome into stages with distinct methods, evidence, and review.

Supporting references

task fit
The match between a task and a method’s capability, risk, evidence, and review needs.

Supporting references

task portfolio
A structured inventory of bounded work candidates with owners, baselines, evidence needs, consequences, and dispositions.

Supporting references

test oracle
An independent rule or source used to decide whether a test result is correct.

Supporting references

text-only pass
A review of slide text and order without visual styling to test whether the logic and reading sequence stand alone.

Supporting references

time to containment
The elapsed time from detecting a failure until its effects are prevented from spreading.

Supporting references

timestamp locator
A time range that lets a reviewer return to the source recording for verification.

Supporting references

tolerance
A predefined acceptable difference arising from stated precision or rounding.

Supporting references

topic-only citation
A link related to the general subject that does not contain evidence supporting the specific claim.

Supporting references

transferred burden
Work, cost, risk, or delay moved from one group or process to another rather than eliminated.

Supporting references

transformation brief
A specification of audience, purpose, allowed changes, protected elements, and acceptance criteria for a format change.

Supporting references

transformation contract
A channel-specific rule set describing what may change and what meaning must remain stable.

Supporting references

transformation disclosure
A clear notice that media contains synthetic, edited, reenacted, or otherwise transformed elements.

Supporting references

transformation log
A sequenced record of changes from raw evidence to analysis-ready data.

Supporting references

trigger condition
The observable situation in which a reusable instruction should apply.

Supporting references

  • OpenAI Help Center: Skills in ChatGPT

    Reviewed: 2026-07-20

    Scope: OpenAI Skills · Documentation reviewed 2026-07-20 · Skill availability, installation, discovery, and invocation vary by OpenAI product and workspace configuration.

  • OpenAI Help Center: Plugins in Codex

    Reviewed: 2026-07-20

    Scope: Codex plugins and apps · Documentation reviewed 2026-07-20 · Plugin capability depends on included skills, enabled apps, app permissions, workspace policy, and product surface.

trust boundary
A point where data or actions cross between actors, systems, permissions, or confidence levels.

Supporting references

truth pass
The first editing pass that checks factual claims, citations, quotations, dates, and permissions before style.

Supporting references

two-strike rule
A recovery trigger used after two targeted attempts fail the same material check.

Supporting references

typed failure
A failure assigned a stable category with defined evidence and permitted recovery states.

Supporting references

uncertainty register
A record of unresolved claims, consequences, allowed actions, and owners.

Supporting references

unknown handling
A rule for representing missing information without inventing a value.

Supporting references

use boundary
A proposed limit on purpose, data, system, actions, ownership, and time prepared for organizational approval.

Supporting references

validation matrix
A table connecting each acceptance rule to its method, evidence, owner, and failure response.

Supporting references

value pre-mortem
A structured forecast of how apparent benefit could fail, be overstated, or shift costs to others.

Supporting references

variance driver
An evidenced component proposed to explain part of the difference between two measured values.

Supporting references

variant constant
A claim, right, disclosure, brand element, or access requirement that must remain unchanged across options.

Supporting references

vendor evidence gap
A material review question for which supplied information is missing, stale, unclear, or out of scope.

Supporting references

verification tier
A defined evidence and review level matched to consequence and uncertainty.

Supporting references

version drift
A meaningful change in workflow behavior associated with changes to models, prompts, tools, data, contracts, or review rules.

Supporting references

visual evidence layer
A structured separation of source context, literal observation, extraction, interpretation, and uncertainty.

Supporting references

visual invariant
A compositional, typographic, color, or subject rule that remains stable across system variants.

Supporting references

visual provenance
The documented origin, custody, context, and known transformations of an image or screenshot.

Supporting references

visual review dossier
A local packet joining a visual artifact with its provenance, rights, claim checks, edit history, accessibility, and review evidence.

Supporting references

voice guide
A maintained set of approved samples and concrete rules for recurring writing choices.

Supporting references

  • OpenAI Help Center: Memory FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Memory · Documentation reviewed 2026-07-20 · Memory behavior depends on settings, plan, workspace policy, and product changes. It is not an organizational permission decision.

  • OpenAI Help Center: Projects in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT Projects · Documentation reviewed 2026-07-20 · Project context and memory vary by plan and settings; this source does not approve any data for use.

  • OpenAI Codex repository: AGENTS.md documentation pointer

    Reviewed: 2026-07-20

    Scope: Codex AGENTS.md guidance · Repository documentation reviewed 2026-07-20 · Repository implementation documentation may change with Codex versions and feature flags.

  • OpenAI Help Center: Data Controls FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT consumer services · Documentation reviewed 2026-07-20 · Consumer Data Controls behavior only. Business, Enterprise, Edu, API, legal obligations, and an employer's approval policy are outside this source's scope.

workflow contract
An explicit definition of inputs, transformations, decisions, outputs, actions, and failure behavior.

Supporting references

workflow trigger
The specific event and required state that make a workflow eligible to begin.

Supporting references

working context
The information available and relevant to the model for the current interaction.

Supporting references

  • OpenAI Help Center: Memory FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Memory · Documentation reviewed 2026-07-20 · Memory behavior depends on settings, plan, workspace policy, and product changes. It is not an organizational permission decision.

  • OpenAI Help Center: Projects in ChatGPT

    Reviewed: 2026-07-20

    Scope: ChatGPT Projects · Documentation reviewed 2026-07-20 · Project context and memory vary by plan and settings; this source does not approve any data for use.

  • OpenAI Help Center: Temporary Chat FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT Temporary Chat for consumers · Documentation reviewed 2026-07-20 · Covers current history, memory, model-improvement, safety-retention, and GPT-action behavior; it is not evidence that a particular organization's data is approved for use.

working context
The material available for the current generation, not a guarantee of permanent knowledge or permission.

Supporting references

working directory
The folder a command treats as its current location and default scope.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

workspace boundary
The files, resources, and tools that a particular task or environment can reach.

Supporting references

  • OpenAI Developers: Codex use cases

    Reviewed: 2026-07-20

    Scope: Codex · Use-case catalog reviewed 2026-07-20 · Illustrative workflows, not a guarantee of availability, permissions, or behavior in every Codex surface.

  • OpenAI: Running Codex safely at OpenAI

    Reviewed: 2026-07-20

    Scope: Codex safety practices · Article reviewed 2026-07-20 · Describes OpenAI practices and product safety concepts; users must still follow their own policy and environment controls.

  • OpenAI Help Center: Data Controls FAQ

    Reviewed: 2026-07-20

    Scope: ChatGPT consumer services · Documentation reviewed 2026-07-20 · Consumer Data Controls behavior only. Business, Enterprise, Edu, API, legal obligations, and an employer's approval policy are outside this source's scope.