Experiments
ActiveAn experiment record is the unit of evidence in this archive. Every number that reaches a chapter or a paper comes from one, and every one of them is complete enough that the run can be repeated by someone else.
| ID | Tests | Project | Status | Result | Reported in |
|---|---|---|---|---|---|
| EXP-001 | HYP-001 | PRJ-001 | planned | None | CH05 |
| EXP-002 | HYP-002 | PRJ-002 | planned | None | CH06 |
Lifecycle
failed means the run did not execute correctly. invalidated means it
executed but the result cannot be trusted, usually because a defect was found
in the setup afterwards. Both stay on the site: the record of what was tried is
part of the result.
What every record must contain
A record missing any of these is not yet an experiment record.
| Field | Why it is mandatory |
|---|---|
| Hypothesis | Ties the run to a prediction made before it ran |
| Configuration | The exact parameters, in a file, not in prose |
| Environment | Hardware, versions, commit pins |
| Inputs | Dataset identifiers and versions |
| Results | Raw observations, with uncertainty |
| Analysis | How raw observations became reported numbers |
| Interpretation | What it means, kept separate from what was observed |
| Limitations | What this run cannot support |
| Reproduction | The commands, in order, that repeat it |
Adding an experiment
Copy templates/experiment-template.md, take the next number from
registry/experiments.yaml, register it in the same commit, and write the
hypothesis link before the run starts. The ordering rule in
research methodology depends on it.