Repository terminology guide

Workflow and Harness

Initial proposal: Workflow is a general arrangement of work; deterministic Python code as well as LLM-based work can participate in a workflow.

Initial proposal: a harness is a workflow designed for an LLM or agent. Editorial revision for discussion: distinguish the execution layer from the process it supports. This edition uses supports/orchestrates rather than declaring a Harness is-a Workflow relationship.

Model and LLM

Initial proposal: any input-to-output mapping may be called a model. This remains an intentionally broad, unratified interpretation, not a universal external definition.

Working revision: specify whether model means a representation, an executable representation, or a learned inference component. An LLM is a learned language model; its training, scale and scope should be described rather than reduced to an arbitrary parameter threshold.

Agent

Initial proposal: an agent builds on an LLM and external tools, with its distinguishing attributes still to be decided.

Working revision: reserve the general term for systems with bounded autonomy, goal-directed behavior, environmental feedback and an ability to act. Describe an LLM-based agent as one particular implementation using an LLM. Planning, persistent memory and self-learning are optional capabilities, not automatic universal requirements.

Editorial and adoption policy

Check external evidence in the fixed order: applicable institutional definitions, peer-reviewed papers, industry standards or first-party documents or preprints or academic materials, then internal proposals. Preserve uncertainty and differing scopes.

Use cited statements, typed relations and a quiet reading layout. Strict nesting is reserved for supported is-a relations. All proposals in this guide are unratified; only a documented consensus decision may authorize green adoption tags.

Schema: an unratified editorial extension

Proposed extension: use schema for an explicit, versioned description of data structure, field meanings and stated constraints. Keep the data instance, the description and the validator separate; the same document can occupy different roles under a meta-schema.

Schema extraction is a process that recovers declarations or infers a description from evidence. Record the input population, source-to-field evidence, dialect/version, observed versus declared or inferred constraints, and unresolved coverage. A sample is not proof of every permitted record or intended semantic rule.

Workflow and Harness: flow and governance proposal

Proposed distinction: Workflow describes how work proceeds, including sequence, dependencies, branches and loops. Harness describes how execution is governed through interfaces, policies, validation, state, recovery and observability. These are overlapping responsibilities, not mutually exclusive categories or a universal subtype taxonomy.

Illustrative schema-extraction workflow: candidate discovery, evidence retrieval, attribute extraction, deterministic schema assembly and validation. This is a proposed teaching example, not a report of an implemented or evaluated Repository system.

Proposed experimental harness policies: standardize input snapshots and model adapters, bound attempts and resource consumption, permit an explicit unknown value only where the declared schema allows it, distinguish evidence support from structural validation, checkpoint progress and record inputs, model/configuration versions, prompts, outputs and failure reasons.

For this example, orchestration assigns models, replicates, devices and dependencies. An experimental harness supplies the configured execution and recording environment. These roles may live in the same implementation. Recording experimental conditions helps audit and repeat a setup; it does not establish that every output will be identical or correct.

Revision record

1. 1.0 / 2026-10-10

   Initial internal proposal, summarized for public citation. Editorial revisions are proposals; no consensus decision recorded.

2. 1.1 / 2026-10-10

   Add the Schema editorial proposal while preserving every 1.0 proposal section. No consensus decision recorded.

3. 1.2 / 2026-10-11

   Add the flow/governance comparison and illustrative schema-extraction experiment. Preserve prior proposal sections; no consensus decision recorded.

4. 1.3 / 2026-10-12

   Rename the project Open AI Glossary and scope certification to this repository. All working meanings remain proposed; no new consensus decision recorded.