12. Glossary

Short definitions of product terms. The link after each entry goes to the page where it is used.

12.1 Specification

Interview

The feature that finds and asks about decisions your idea has not made yet. Ch. 5

Ambiguity Score

A 0-to-1 value for the ambiguity left in the requirements. At 0.2 or below a Seed can be generated. Ch. 5

Seed

The work-spec YAML file holding goal, constraints, completion criteria, and data structure. Both the runtime and evaluation work against it. Ch. 6

Goal

The Seed field stating what this run should produce. Ch. 6

Constraints

The Seed field holding the limits a run must respect. Ch. 6

Non-goals

What you agreed not to build this time. Recorded in the Ledger, not the Seed body. Ch. 6

Acceptance Criteria

The list of criteria for judging completion, written so each can be verified on screen or by a test. Ch. 6

Ontology

The structure naming and typing the data the work handles. Ch. 6

Grade A

The Seed state with no blockers and quality scores past their thresholds, allowing execution. ooo auto starts execution only at grade A. Ch. 6

Ledger

The record of assumptions, exclusions, and decision grounds confirmed while questioning. Ch. 6

12.2 Execution

Core

The Ouroboros engine handling the Interview, Seeds, execution management, and evaluation. Ch. 3

runtime

The AI coding tool that actually touches files and code, such as Claude Code or Codex CLI. Ch. 3

MCP

The protocol that lets the runtime call Ouroboros Core's features. Ch. 4

Double Diamond

The execution phases run in the order Discover, Define, Design, Deliver. Ch. 7

12.3 Evaluation

Mechanical

Evaluation stage 1: runs commands such as lint, build, and test and checks the results. Ch. 8

Semantic

Evaluation stage 2: an LLM judges whether the result meets the Acceptance Criteria. Ch. 8

Consensus

Evaluation stage 3: multiple models vote independently on approval. Runs only under set conditions. Ch. 8

Drift

A 0-to-1 value for how far a result moved from the original Seed's intent, weighted Goal 50%, Constraints 30%, Ontology 20%. Ch. 8

12.4 Iteration

Evolve

The feature that applies evaluation findings to the Seed and runs again. Ch. 9

Wonder · Reflect

The first two steps of one Evolve generation. Wonder finds what is still unknown; Reflect turns that into a Seed change proposal. Ch. 9

lineage

The history grouping the chain of Seed revisions from the same piece of work. Ch. 9

Convergence

The state where repeated runs barely change the Seed's data structure. The criterion is an Ontology similarity of 0.95 or higher between consecutive generations. Ch. 9

Stagnation

The state where iteration no longer improves the result. Used as the signal to stop and change approach. Ch. 9

Ralph

The feature that repeats revision runs in the background until a stop condition is reached. Ch. 9

12.5 Records and IDs

EventStore

The local SQLite database storing every execution event. Used for session recovery. Ch. 3

session_id · job_id · execution_id · lineage_id · auto_session_id

IDs used for session queries and evaluation, background-job queries, cancelling runs, revision history, and auto resume, respectively. Ch. 5 · Ch. 7