The terms the Atlas and the Workflows use, explained in plain language. Nobody should need the White Paper to read this site.
A function and area of responsibility within a specific piece of work — for example Developer, Architect, Reviewer, Tester. A managed type of the knowledge graph: a role carries versions, relations, provenance and evidence.
labels a column in the evidence tablesA reusable method of performing a specific professional operation. A managed type of the knowledge graph.
An area of professional ability — for example devops-operations. It appears on this site in the Atlas accounting and filters.
A formalized way of performing a class of work: the sequence of actions, tools, checkpoints and quality criteria. A managed type of the knowledge graph.
An ability the system applies in real tasks; its knowledge lives in the managed artifacts — roles, skills, blueprints. Professional Capability is the higher-order product term: a delegable ability to perform a class of professional work to a defined standard; it is not a synonym of capability, competency, skill or role.
A bounded set of knowledge, rules and obligations for a specific task: what the executor must know, what it may do, what it must not do, and how success is defined.
Verifiable information about execution: sources, tests, logs, reasoning. On this site evidence accompanies every record — from an Atlas capability to a cost row.
A published record of real work being executed: the steps, roles, artifacts, time, cost and evidence. Real tasks, real agents, a verifiable process.
The accounted unit of execution: one step of a task with its own Workflow Ledger record carrying time, cost and validation status. A row of the time-and-cost table is a subtask, not a whole task.
labels a column in the evidence tablesThe time from the start of a unit of work to its finish, including waiting, pauses and everything in between. It is the elapsed duration of the execution window, not the amount of work done; across parallel lanes the sum of windows is legitimately larger than the calendar time.
labels a column in the evidence tablesThe time during which work was actually executing, with the waiting and the idle stretches inside the window removed. It is always less than or equal to wall time; the difference between them is waiting, not lost work.
labels a column in the evidence tablesThe cost of the language-model calls for this unit of work: input and output tokens at the provider's price. Where no metered provider invoice exists for the session, the figure is marked as an estimate and its cost basis is published beside the row.
labels a column in the evidence tablesThe cost of machine time outside the model: servers, builds, test runs and other infrastructure spent on this unit of work. A zero here means measured absence of such spend, not absence of measurement.
labels a column in the evidence tablesThe human time on this unit of work, converted to money at the rate the operator declared. Both the rate and the time itself are reconstructed after the fact from evidence rather than measured while the work ran, so every figure in this column is marked as reconstructed — and a row with no human participation has no human cost at all.
labels a column in the evidence tablesThe sum of model, compute and human cost for this unit of work. Only what was measured is summed: if nothing in the column was measured the total is not zero — it is unknown, and says so.
labels a column in the evidence tablesThe managed knowledge-graph type an artifact belongs to: role, skill, blueprint plus the governance types constraint, success criterion, policy and capability description. The type decides which contract the artifact lives under and how it may be reused.
labels a column in the evidence tablesThe version of the knowledge artifact as it took part in this task. Artifacts are versioned in the private knowledge repository and a task is bound to a specific version rather than to «current» — otherwise the evidence would stop reproducing at the next edit.
labels a column in the evidence tablesThe artifact's identifier and the link to its Capability Atlas entry where one is published. The source is the address at which the claim can be checked; where no public entry exists the identifier is shown alone rather than an invented link.
labels a column in the evidence tablesThe stage of a knowledge artifact: draft, reviewed, validated, shared-internal, public-sanitized, deprecated. Only public-sanitized artifacts reach a public surface; the rest are accounted for as withheld rather than quietly omitted.
labels a column in the evidence tablesHow a knowledge artifact took part in the task: created — authored for the first time, reused — used unchanged, modified — revised. The reuse metric stands on this mark, so it is recorded in the ledger entry rather than inferred afterwards.
labels a column in the evidence tablesThe identifier of the task in which this artifact was touched. A task is the unit of project work; it contains the subtasks that time and cost are accounted against, and this identifier is what ties a row of evidence to the ledger records.
labels a column in the evidence tablesThe control verdict for a unit of work: passed, failed, pending, skipped or not applicable. Failed records are kept alongside the rest: work done around the contract is not accepted as sound, and is not erased from the accounting either.
labels a column in the evidence tables