Reference
Glossary
The terms Horizon Audit uses in its results, its dashboard and these documents, each with one meaning.
Each term on this page has one meaning. The dashboard, the results and these documents all use a term in the same way.
Tasks and submissions
| Term | Meaning |
|---|---|
| Task | One Harbor task: an instruction, a task.toml, an environment, a reference solution and a verifier. |
| Submission | The tasks you send in at one time, as one archive under one collection name. |
| Digest | A hash of the task's files as they were read from your archive. It identifies those exact files. |
| Intake standard | The published list of what a task must have before it can be audited. See the intake standard. |
| Band | Basic, Standard, Complex or Specialist. A band says how much work an audit of the task is. |
The parts of an audit
| Term | Meaning |
|---|---|
| Grader (verifier) | The code of the task that scores an answer and writes the reward. "Grader" and "verifier" mean the same thing. |
| Reference solution | The author's own solution to the task, at solution/solve.sh. |
| Sandbox | The isolated remote environment where the code of a task runs. |
Results
| Term | Meaning |
|---|---|
| Verdict | Clean or flawed. It holds for the exact files that were audited. "Needs from you" is not a verdict. |
| Flag | Something worth a look. A flag never decides a verdict. |
| Finding | The write-up of a flaw: where it is, how it was established, and what it does not claim. |
| Material | A flaw is material when it can change a realistic agent's score. Only a material flaw makes a task flawed. |
Words that are not verdicts
Three words describe a task that has no verdict.
- Needs from you. The task cannot be audited until you supply something. It is never counted as clean.
- In review. An engineer is looking at the task before its result is published. You never see a draft.
- Waiting for review. The audit of the task does not start until an engineer acts.
See task states for every label, and read a verdict for what a verdict proves.