The words that ran software delivery rationed one scarce thing: engineering hours. When agents build, that scarcity is gone, and most of the words need a successor. These are the terms Rootshift uses, with the word each one replaces where there is one. Each term has its own link.
- Ground Truth
- A versioned, structured model of how a business and its product work, in seven parts, that a person and an agent can both build against and verify against. If it is not structured, owned, versioned, and validated, it is not Ground Truth. REPLACES: BACKLOG
- Artifact
- One file in the Ground Truth. Every artifact carries an id, an owner, a status, a confidence level, dated sources, and the date it was last validated. REPLACES: DOCUMENTATION
- Owner
- The named human accountable for an artifact being true. Roles, not committees.
- Status
- Where an artifact stands: draft, in-review, approved, needs-revalidation, or deprecated. Agents may only build production code from approved artifacts.
- Confidence
- How sure the model is. High requires a primary source and human validation; low means inferred, stale, or conflicting.
- Capability
- A unit of value: a business function, a verb, such as scan a receipt or reconcile a statement. It is made of stories with acceptance criteria and evals, and carries a class, a tier and a size so it is comparable across projects. REPLACES: TICKET
- Story
- A unit of work inside a capability, with Given/When/Then acceptance criteria. A builder agent implements one story at a time.
- Open decision
- What the system writes instead of guessing when sources disagree or a fact is missing: the question, the options, the affected areas, a recommendation, a confidence level, and an owner. Dependent work waits for the answer.
- Decision log
- Part 07 of the Ground Truth. Every material decision as an artifact: who, when, why, the alternatives rejected, what changed. Never rewritten, only superseded.
- Knowledge intake
- Turning existing material (code, documents, interviews, the live product and its data) into a draft Ground Truth, with its inconsistencies and gaps made explicit. REPLACES: WORKSHOPS
- Builder brief
- What a builder agent receives: the story verbatim, its capability, the decisions and rules that touch it, the engineering context, and a pointer to the repository. Assembled deterministically from approved artifacts: same Ground Truth in, byte-identical brief out.
- Fresh-context builder
- A builder agent that starts each story with a clean context, so nothing from a previous task leaks into the next one. It opens a pull request with a trace back to the specification.
- Independent review
- Reviewer agents that did not build the change inspect it for correctness, security, architecture, reliability, and test adequacy. The reviewer is never the fixer, and the fixer is never the merger. No agent judges its own work.
- Evals
- One named check per acceptance criterion, plus the checks that catch regressions. Part 06 of the Ground Truth. REPLACES: QA
- Gate
- A check that refuses work. The readiness gate refuses to build from unapproved truth; the merge gate refuses a change without verification evidence. Gates are calibrated, not maximal.
- Delivery loop
- Knowledge intake, Ground Truth, agent fleet execution, closed loops, quality gates, telemetry, and back to intake. A change closes when its evidence and outcome are recorded. REPLACES: SPRINT
- The record
- The telemetry of every run: outcomes, findings found and held, decision latency, and bounces. Forecasts come from it. REPLACES: VELOCITY
- Delivered history
- What comparable capabilities actually took, from the record. The basis for data-backed simulations instead of estimates. REPLACES: STORY POINTS
- Evidence class
- How a claim is backed: verified live, verified against the installed tool, or still a claim. Every number on the site carries one.
- Operator
- The person who sets the objectives, resolves the decisions the system queues, and approves what is true. The system works within those boundaries.